GNU poke Manual

Table of Contents

GNU poke Manual

This manual describes GNU poke (version 1.0, 25 February 2021).

Copyright © 2019-2021 The poke authors.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included in the section entitled “GNU Free Documentation License”.

Using poke

Pickles

Reference Material

Internals

Appendices

Indexes

Here are some other nodes which are really subnodes of the ones
already listed, mentioned here so you can get to them in one step:

Introduction

Basic Editing

Structuring Data

Maps and Map-files

Writing Pickles

Configuration

Time

Colors

Audio

Dot-Commands

Commands

The Poke Language

The Standard Library

The Machine-Interface

The Poke Virtual Machine

1 Introduction

1.1 Motivation

The main purpose of GNU poke is to manipulate structured binary data in terms of abstractions provided by the user. The Poke type definitions can be seen as a sort of declarative specifications for decoding and encoding procedures. The user specifies the structure of the data to be manipulated, and poke uses that information to automagically decode and encode the data. Under this perspective, struct types correspond to sequences of instructions, array types to repetitions or loops, union types to alternatives or conditionals, and so on.

1.1.1 Decode-Compute-Encode

Computing with data whose form is not the most convenient way to be manipulated, like is often the case in unstructured binary data, requires performing a preliminary step that transforms the data into a more convenient representation, usually featuring a higher level of abstraction. This step is known in computer jargon as unmarshalling, when the data is fetch from some storage or transmission media or, more generally, decoding.

Once the computation has been performed, the result should be transformed back to the low-level representation to be stored or transmitted. This is performed in a closing step known as marshalling or, more generally, encoding.

Consider the following C program whose purpose is to read a 32-bit signed integer from a byte-oriented storage media at a given offset, multiply it by two, and store the result at the same offset.

void double_number (int fd, off_t offset, int endian)
{
   int number, i;
   int b[4];

   /* Decode.  */
   lseek (fd, offset, SEEK_SET);
   for (i = 0; i < 4; ++i)
      read (fd, &b[i], 1);

   if (endian == BIG)
     number = b[0] << 24 | b[1] << 16 | b[2] << 8 | b[3];
   else
     number = b[3] << 24 | b[2] << 16 | b[1] << 8 | b[0];

   /* Compute.  */
   number = number * 2;

   /* Encode.  */
   if (endian == BIG)
   {
     b[0] = (number >> 24) & 0xff;
     b[1] = (number >> 16) & 0xff;
     b[2] = (number >> 8) & 0xff;
     b[3] = number & 0xff;
   }
   else
   {
     b[3] = (number >> 24) & 0xff;
     b[2] = (number >> 16) & 0xff;
     b[1] = (number >> 8) & 0xff;
     b[0] = number & 0xff;
   }

   lseek (fd, offset, SEEK_SET);
   for (i = 0; i < 4; ++i)
      write (fd, &b[i], 1);
}

As we can see, decoding takes care of fetching the data from the storage in simple units, bytes. Then it mounts the more abstract entity on which the computation will be performed, in this case a signed 32-bit integer. Considerations like endianness, negative encoding (which is assumed to be two’s complement in this example and handled automatically by C) and error conditions (omitted in this example for clarity) should be handled properly.

Conversely, encoding turns the signed 32-bit integer into a sequence of bytes and then writes them out to the storage at the desired offset. Again, this requires taking endianness into account and handling error conditions.

This example may look simplistic and artificial, and it is, but too often the computation proper (like multiplying the integer by two) is way more straightforward than the decoding and encoding of the data used for the computation.

Generally speaking, decoding and encoding binary data is laborious and error prone. Think about sequences of elements, variable-length and clever compact encodings, elements not aligned to byte boundaries, the always bug-prone endianness, and a long etc. Dirty business, sometimes risky, and always boring.

1.1.2 Describe-Compute

This is where poke comes into play. Basically, it allows you to describe the characteristics of the data you want to compute on, and then decodes and encodes it for you, taking care of the gory details. That way you can concentrate your energy on the fun part: computing on the data at your pleasure.

Of course, you are still required to provide a description of the data. In the Poke language, these descriptions take the form of type definitions, which are declarative: you specify what you want, and poke extracts the how from that.

For example, consider the following Poke type definition:

type Packet =
  struct
  {
    uint<16> magic = 0xef;
    uint<32> size;
    byte[size] data @ 64#B;
  };

This tells poke that, in order to decode a Packet, it should perform the following procedure (a similar procedure is implied for encoding):

• - Read four bytes from the IO space, mount them into an unsigned 16-bit integer using whatever current endianness, and put it in magic. If this unsigned 16-bit integer doesn’t equal to 0xef, then stop and emit a “data integrity” error.
• - Read eight bytes, mount them into an unsigned 32-bit integer using the same endianness, and put it in size.
• - Seek the IO space to advance 16 bits.
• - Do size times:
  • - Read one byte and mount it into an unsigned 8-bit integer.
  • - Put the integer in the proper place in the data array.

If during this procedure an end of file is encountered, or some other erroneous condition happens, an appropriate error is raised.

In the procedure sketched above we find a sequence of operations, implied by the struct type, and a loop, implied by the array type. As we shall see later in this book, it is also possible to decode conditionally. Union types are used for that purpose.

1.2 Nomenclature

GNU poke is a new program and it introduces many a new concept. It is a good idea to clarify how we call things in the poke community. Unless everyone uses the same nomenclature to refer to pokish thingies, it is gonna get very confusing very soon!

First of all we have poke, the program. Since “poke” is a very common English word, when the context is not clear we either use the full denomination GNU poke, or quote the word using some other notation.

Then we have Poke, with upper case P, which is the name of the domain-specific programming language implemented by poke, the program.

This distinction is important. For example, when people talk about “poke programmers” they refer to the group of people hacking GNU poke. When they talk about “Poke programmers” they refer to the people who write programs using the Poke programming language.

Finally, a pickle is a Poke source file containing definitions of types, variables, functions, etc, that conceptually apply to some definite domain. For example, elf.pk is a pickle that provides facilities to poke ELF object files. Pickles are not necessarily related to file formats: a set of functions to work with bit patterns, for example, could be implemented in a pickle named bitpatterns.pk.

We hope this helps to clarify things.

1.3 Invoking poke

Synopsis:

poke [option…] [file]

The following options are available.

‘-l’

‘--load=file’

Load the given file as a Poke program. Any number of ‘-l’ options can be specified, and they are loaded in the given order.

‘-L file’

Load the given file as a Poke program and exit. The rest of the command-line is not processed by poke, and is available to the Poke script in the argv variable. This is commonly used along with a shebang (see Scripts) to implement Poke scripts.

Commanding poke from the command line:

‘-c’

‘--command=cmd’

Execute the given command. Any number of ‘-c’ options can be specified, and they are executed in the given order.

‘-s’

‘--source=file’

Load file as a command file. Any number of ‘-s’ options may be specified, and they are loaded in the given order. See Command Files.

Styling text output:

‘--color=how’

Whether to use styled output, and how. Valid options for how are ‘yes’, ‘no’, ‘auto’, ‘html’ and ‘test’.

‘--style=file’

Use file as the CSS to use for styling poke, instead of the default style.

Machine interface:

--mi

Run poke in MI mode. In this mode, poke communicates with a client using JSON messages in the standard input and output.

Other options:

-q

--no-init-file

Do not load the ~/.pokerc init file.

--no-auto-map

Do not load map-files automatically when poke opens IO spaces.

--no-hserver

Do not run the terminal hyperlinks server.

--quiet

Be as terse as possible.

--help

Print a help message and exit.

--version

Show version and exit.

1.4 Commanding poke

GNU poke is primarily an interactive editor that works in the command line. However, it is also possible to use it in a non-interactive way. This chapter documents both possibilities.

1.4.1 The REPL

If poke is invoked with an interactive TTY connected to the standard input, it greets you with a welcome message, licensing information and such, and finally a prompt that looks like:

(poke)

At this point, the program is ready to be commanded. You are expected to introduce a line and press enter. At that point poke will examine the command, notify you if there is some error condition, process the line and maybe displaying something in the terminal.

Repeatedly typing complex commands can be tiresome. To help you, poke uses the readline library See GNU Readline Library. This provides shortcuts and simple keystrokes to repeat previous commands with or without modification, fast selection of file names and entries from other multiple choice contexts, and navigation within a command and among previous commands. When the REPL starts, the history of your previous sessions are loaded from the file .poke_history located in your home directory (if it exists).

There are several kinds of lines that can be provided in the REPL:

• A dot-command invocation, that starts with a dot character (.).
• A command invocation.
• A Poke statement.
• A Poke expression.

These are explained in the following sections.

1.4.2 Evaluation

You can evaluate a Poke statement by typing it at the REPL’s prompt. Only a single statement, including expression statements (see Expression Statements) and compound statements (see Compound Statements), can be evaluted this way. It needs not be terminated by a semicolon.

When an expression is evaluated, the result of the evaluation is printed back to you. For example:

(poke) 23
23
(poke) [1,2,3]
[1,2,3]
(poke) Packet @ 0#B
Packet {i=1179403647,j=65794L}

When a statement other than an expression statement is executed in the REPL no result is printed, but of course the statement can print on its own:

(poke) fun do_foo = void: {}
(poke) do_foo
(poke) for (i in [1,2,3]) printf "elem %i32d\n", i;
elem 1
elem 2
elem 3

If there is an error compiling the line, you are notified with a nice error message, showing the location of the error. For example:

(poke) [1,2,3 + "foo"]
<stdin>:1:6: error: invalid operands in expression
[1,2,3 + "foo"];
     ^~~~~~~~~

1.4.3 Commands and Dot-Commands

There are two kinds of commands in poke: the dot-commands, which are written in C and have their own conventions for handling sub-commands and passing arguments and flags, and normal commands, which are written in Poke.

1.4.3.1 Dot-Commands

Dot-commands are so called because their names start with the dot character (.). They can feature subcommands. Example:

(poke) .vm disassemble mapper int[] @ 0#B
(poke) .vm disassemble writer int[] @ 0#B

When there is no ambiguity, the command name and the subcommands can be shortened to prefixes. The commands above can also be written as:

(poke) .vm dis m int[] @ 0#B
(poke) .vm dis w int[] @ 0#B

Some commands also get flags, which are one-letter indicators that can be appended to the command name (including subcommands) after a slash character (/). For example, the .vm disassembler commands accept a n flag to indicate we want a native disassemble. We can pass it as follows:

(poke) .vm disassemble mapper/n int[] @ 0#B
(poke) .vm disassemble writer/n int[] @ 0#B

If a dot-command accepts more than one argument, they are separated using comma characters (,). Spaces are generally ignored.

1.4.3.2 Commands

Regular poke commands are written in Poke and use different conventions. The name of commands follow the same rules as normal Poke identifiers, and do not start with a dot character.

An example is the dump command:

(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 f700 0100 0000 0000 0000 0000 0000  ................
00000020: 0000 0000 0000 0000 8001 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0800 0700  ....@.....@.....
00000040: 1800 0000 0000 0000 0000 0000 0000 0000  ................
00000050: 7900 0000 0000 0000 b701 0000 9a02 0000  y...............
00000060: 7b10 0000 0000 0000 1800 0000 0000 0000  {...............
00000070: 0000 0000 0000 0000 7900 0000 0000 0000  ........y.......

After the name of the command, arguments can be specified by name, like this:

(poke) dump :from 0#B :size 8#B
(poke) dump :from 0#B :size 8#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100                      .ELF....

The dump command is discussed in greater detail below (see dump). The order of arguments is irrelevant in principle:

(poke) dump :from 0#B :size 8#B :ascii 0 :ruler 0
00000000: 7f45 4c46 0201 0100
(poke) dump :ruler 0 :from 0#B :size 8#B :ascii 0
00000000: 7f45 4c46 0201 0100

However, beware side effects while computing the values you pass as the arguments! The expressions themselves are evaluated from left to right.

Which arguments are accepted, and their kind, depend on the specific command.

Note that the idea is to restrict the number of dot-commands to the absolutely minimum. Most of the command-like functionality provided in poke shall be implemented as regular commands.

1.4.4 Command Files

Command files contain poke commands. A poke command may be a dot command, a Poke statement or a Poke expression. Lines starting with # are comments will be ignored. However a comment must start at the beginning of a line. Here is an example of a script:

# The following two lines are dot commands
.load my-pickle.pk
.set obase 16

# The following line is a Poke statement
dump :size 0x100#B :from 0x10#B

# The following line is a Poke expression statement without any side effect.
# Consequently it is valid, but rather useless.
4 == 4

A command file contains commands, not Poke code. This means it gets read line by line and commands cannot occupy more than one line. Hence the following is a valid command file:

type foo = struct {int this; int that;}

but this is not valid as a command file (although it is a valid Poke statement) and will provoke an error:

type foo = struct
{
 int this;
 int that;
}

Command files can be loaded at startup using the -s command line option (see Invoking poke). The ~/.pokerc startup file is also an example of a poke command file (see pokerc).

1.4.5 Scripts

Following the example of Guile Scheme, the Poke syntax includes support for multi-line comments using the #! and !# delimiters. This, along with the -L command line option, allows to write Poke scripts and execute them in the command line like if they were normal programs. Example of a script:

#!/usr/bin/poke -L
!#

print "Hello world!\n";

The resulting script can process command-line options by accessing the argv array. The following Poke script prints its arguments:

#!/usr/bin/poke -L
!#

for (arg in argv)
  printf ("Argument: %s\n", arg);

If you want to pass additional flags to the poke command, you need to use a slightly different kind of shebang:

#!/usr/bin/env sh
exec poke -L "$0" "$@"
!#

load elf;
printf ("%v\n", Elf64_Ehdr @ 0#B);

2 Basic Editing

In this chapter you will learn how to shuffle binary data around with poke, in terms of fundamental predefined entities: bits, bytes, integers, and the like.

2.1 Binary Files

GNU poke is an editor for binary files. Right, so what is a binary file? Strictly speaking, every file in a computer’s file system is binary. This is because, in a very fundamental level, files are just sequences of bytes.

Colloquially, however, it is very common to talk about “binary files” as opposed to “text files”. In this informal meaning, a text file is basically a file composed, mostly, of bytes (and byte sequences) that can be translated into printable characters in some character set, such as ASCII, EBCDIC or Unicode. It follows that binary files would then be files composed, mostly, of bytes not intended to be interpreted as encoded characters.

Some text files contain non-printable characters, such as form feed characters, and many binary files contain printable strings, such as a string table in an ELF object file. That is why we used the word “mostly” in the definitions above. In practice, however, the distinction is almost always clear and there is common consensus on whether a given file format can be considered as a binary format, or not.

GNU poke can edit any file, and as we shall see, it provides some nice features to manipulate sequences of bytes interpreted as character strings. However, it is called a “binary editor” because it is especially designed to be particularly useful editing binary files, in the sense of the term defined above.

In this chapter, we will be using ELF object files as the experiment subject in most of the examples. ELF files are good for this purpose, because they are eminently binary, highly structured, and still strings play a role in them, encoding names of entities like sections and symbols. You don’t need to have a perfect knowledge of the ELF format in order to follow the examples, but being familiarized with the concept of object file formats should surely help.

Obtaining a simple ELF object file is easy, if you have a C compiler installed:

$ echo 'int foo () { return 0; }' | gcc -c -xc -o foo.o -

The command above compiles a very simple ELF object file that contains the compiled form of a little dummy function. This object file will be our companion for a while, and will be the subject of much analysis and abuse, as we poke it.

2.2 Files as IO Spaces

Now that we have a binary file (foo.o) it is time to open it with poke. There are two ways to do that.

One way is to pass the name of the file in the poke invocation. The program will start, open the file, and present you with the REPL, like in:

$ poke foo.o
[…]
(poke)

The other way is to fire up poke without arguments, and then use the .file dot-command to open the file:

$ poke
[…]
(poke) .file foo.o
The current IOS is now `./foo.o'.
(poke)

Note how poke replies to the dot-command, stating that the current IOS is now the file we opened.

You may be wondering, what is this IOS thing? It is an acronym for Input/Output Space, often written IO Space. This is the denomination used to refer to the entities being edited with poke. In this case the IO space being edited is a file, but we will see that is not always the case: poke can also edit other entities such as memory buffers and remote block-oriented devices over the network. For now, let’s keep in mind that IOS, or IO space, refers to the file being edited.

And why “current”? GNU poke is capable of editing several files (more generally, several IO spaces) simultaneously. At any time, one of these files is the “current one”. In our case, the current IO space is the file foo.o, since it is the only file poke knows about:

(poke) .info ios
  Id	Type	Mode	Size		Name
* #0	FILE	rw	0x00000398#B	./foo.o

The command .info ios gives us information about all the IO spaces that are currently open. The first column tells us a tag that identifies the IOS. In this example, the tag corresponding to foo.o is #0. The second column tells us the type of IO space. The third column tells us that foo.o allows both reading and writing. The fourth column tells us the size of the file, in hexadecimal.

You may wonder what is that weird suffix #B. It is a unit, and tells us that the size 0x398 is measured in bytes, i.e. the size of foo.o is 0x398 bytes (or, in decimal, 920 bytes.)

Finally, the asterisk character at the left of the entry for foo.o identifies it as the current IO space. To see this more clearly, let’s open another file:

(poke) .file bar.o
The current IOS is now `./bar.o'.
(poke) .info ios
  Id	Type	Mode	Size		Name
* #1	FILE	rw	0x00000398#B	./bar.o
  #0	FILE	rw	0x00000398#B	./foo.o

Ah, there we have both foo.o and bar.o. Now the current IO space (the entry featuring the asterisk at the left) is the file that we just opened, bar.o. This is because poke always sets the most recently open file as the current one. We can switch back to foo.o using yet another dot-command, .ios, which gets an IO space tag as an argument:

(poke) .ios #0
The current IOS is now `./foo.o'.
(poke) .info ios
  Id	Type	Mode	Size		Name
  #1	FILE	rw	0x00000398#B	./bar.o
* #0	FILE	rw	0x00000398#B	./foo.o

We are back to foo.o. Since we are not really interested in bar.o, let’s close it:

(poke) .close #1
(poke) .info ios
  Id	Type	Mode	Size		Name
* #0	FILE	rw	0x00000398#B	./foo.o

Awesome. Now we can focus on foo.o’s contents…

2.3 Dumping File Contents

Data stored in modern computers, in both volatile memory and persistent files, is fundamentally a sequence of entities called bytes. The bytes can be addressed by its position in the sequence, starting with zero:

+--------+--------+--------+ ... +--------+
| byte 0 | byte 1 | byte 2 |     | byte N |
+--------+--------+--------+ ... +--------+

Each byte has capacity to store a little unsigned integer in the range 0..255. Therefore, the IO spaces that we edit with poke (like the file foo.o) can be seen as a sequence of little numbers, like depicted in the figure above.

GNU poke provides a command whose purpose is to display the values of these bytes: dump¹ . It is called like that because it dumps ranges of bytes to the terminal, allowing the user to inspect them.

So let’s use our first poke command! Fire up poke, open the file foo.o as explained above, and execute the dump command:

(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 f700 0102 0000 0000 0000 0000 0000  ................
00000020: 0102 0000 0000 0000 9801 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0800 0700  ..............
00000040: 2564 0a00 0000 0000 0000 0000 0000 0000  %d..............
00000050: b702 0000 0100 0000 1801 0000 0000 0000  ................
00000060: 0000 0000 0000 0000 8510 0000 ffff ffff  ................
00000070: b700 0000 0000 0000 9500 0000 0000 0000  ................
(poke)

What are we looking at?

The first line of the output, starting with 76543210, is a ruler. It is there to help us to visually determine the location (or offset) of the data.

The rest of the lines show the values of the bytes that are stored in the file, 16 bytes per line. The first column in these data lines shows the offset, in hexadecimal and measured in number of bytes, from which the row of data starts. For example, the offset of the first byte shown in the third data line has offset 0x20 in the file, the second byte has offset 0x21, and so on. Note how the data rows show the values of the individual bytes, in hexadecimal. Generally speaking, when dealing with bytes (and binary data in general) it is useful to manipulate magnitudes in hexadecimal, or octal. This is because it is easy to group digits in these bases to little groups of bits (four and three respectively) in the equivalent binary representation. In this case, each couple of hexadecimal digits denote the value of a single byte². For example, the value of the first byte in the third data row is 0x01, the value of the second byte 0x02, and so on.

Using the ruler and the column of offsets, locating bytes in the data is very easy. Let’s say for example we are interested in the byte at offset 0x68: we use the first column to quickly find the row starting at 0x60, and the ruler to find the column marked with 88. Cross column and row and… voila! The byte in question has the value 0x85. The reverse process is just as easy. What is the offset of the first 0x40 in the file? Try it!

The section at the right of the output is the ASCII output. It shows the row of bytes at the left interpreted as ASCII characters. Non-printable characters are shown as . to avoid scrambling the terminal, and yes, there is actually way to customize what character to use, so they are not confused from real ASCII dot characters (0x2e) :P In this particular dump we can see that near the beginning of the file there are three bytes whose value, if interpreted as ASCII characters, conform the string “ELF”. As we shall see, this is part of the ELF magic number. Again, the ruler is very useful to locate the byte corresponding to some character in the ASCII section, or the other way around. What is the value of the byte corresponding to the F in ELF? Try it!

Something to notice in the dump output above is that these are not, by any mean, the complete contents of the file foo.o. The .info ios dot-command informed us in the last section that foo.o contains 920 bytes, of which the dump command only showed us… 0x80 bytes, or 128 bytes in decimal.

dump is certainly capable of showing more (and less) than 128 bytes. We can ask dump to display some given amount of data by specifying its size using a command argument. For example:

(poke) dump :size 64#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 f700 0102 0000 0000 0000 0000 0000  ................
00000020: 0102 0000 0000 0000 9801 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0800 0700  ..............

The command above asks poke to “dump 64 bytes”. In this example :size is the name of the argument, and 64#B is the argument’s value. Again, the suffix #B tells poke we want to dump 64 bytes, not 64 kilobits nor 64 potatoes.

Another interesting aspect of our first dump (ahem) is that the dumped bytes start from the beginning of the file, i.e. the offset of the first byte is 0x0. Certainly there should be other areas of the file with interesting contents for us to inspect. To that purpose, we can use yet another option, :from:

(poke) dump :size 64#B :from 128#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000080: 1400 0000 0000 0000 0000 0000 0000 0000  ................
00000090: 0000 0000 0000 0000 0000 0000 0000 0000  ................
000000a0: 0000 0000 0300 0100 0000 0000 0000 0000  ................
000000b0: 0000 0000 0000 0000 0000 0000 0300 0300  ................

The command above asks poke to “dump 64 bytes starting at 128 bytes from the beginning of the file”. Note how the first row of bytes start at offset 0x80, i.e. 128 in decimal.

Passing options to commands is easy and natural, but we may find ourselves passing the same values again and again to certain command options. For example, if the default size of dump of 128 bytes is not what you prefer, because you have a particularly tall monitor, or you are one of these people using sub-atomic sized fonts, it can be tiresome and error-prone to pass :size to dump every time you use it. Fortunately, the default size can be customized by setting a global variable:

(poke) pk_dump_size = 160#B

This tells poke to set 160 bytes as the new value for the pk_dump_size variable. This is a global variable that the dump command uses to determine how much data to show if the user doesn’t specify an explicit value with the :size option. Many other commands use the same strategy in order to alter their default behavior, not just dump.

And now that we are talking about that, it is also cumbersome to have to set the default size used by dump every time we run poke. But no problem, just set the variable in a file called .pokerc in your home directory, like this:

pk_dump_size = 160#B

Every time poke starts, it reads ~/.pokerc and executes the commands contained in it. See pokerc.

The dump command is very flexible, and accepts a lot of options and customization variables that we won’t be covering in this chapter. For a complete description of the command, see dump.

2.4 Poking Bytes

Let’s look again at the first bytes of the file foo.o:

(poke) dump :size 64#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 f700 0102 0000 0000 0000 0000 0000  ................
00000020: 0102 0000 0000 0000 9801 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0800 0700  ..............

At this point we know how to use the ruler to localize specific bytes just by looking at the displayed data. If we wanted to operate on the values of some given bytes, we could look at the dump and type the values in the REPL. For example, if we wanted to add the values of the bytes at offsets 0x2 and 0x4, we could look at the dump and then type:

(poke) 0x4c + 0x02
0x4e

GNU poke supports many operators that take integers as arguments, to perform arithmetic, relational, logical and bit-wise operations on them (see Integers). Since bytes are no more (and no less) than little unsigned integers, we can use these operators to perform calculations on bytes.

For example, this is how we would calculate whether the highest bit in the second byte in foo.o is set:

(poke) 0x45 & 0x80
0

Note how booleans are encoded in Poke as integers, 0 meaning false, any other value meaning true.

Looking at the output of dump and writing the desired byte value in the prompt is cumbersome. Fortunately, there is a much more convenient way to access the value of a byte, given its offset in the file: it is called mapping a byte value. This operation is implemented by a binary operator, called the map operator.

This is how it works. Assuming we were interested in the byte at 64 bytes from the beginning of the file, this is how we would refer to it (or “map” it):

(poke) byte @ 64#B
37UB

This application of the map operator tells poke to map a byte at the offset 64 bytes. It can be read as “byte at 64 bytes”. Note how poke replies with the value 37UB. The suffix UB means “unsigned byte”, and is an indication for the user about the nature of the preceding number: it is unsigned, and it occupies a byte when stored.

As we can see in this example, poke uses decimal by default when showing values in the REPL. We already noted how it is usually better to work in hexadecimal when dealing with byte values. Fortunately, we can change the numeration base used by poke when printing numbers, using the .set obase (“set output base”) dot-command as this:

(poke) .set obase 16

After this, we can map the byte again, this time getting the result expressed in hexadecimal:

(poke) byte @ 64#B
0x25UB

Again, you may find it useful to add the .set obase 16 command to your .pokerc file, if you want the customization to be persistent between poke invocations.

Going back to the example of calculating whether the highest bit in the second byte in foo.o is set, this is how we would do it with a map:

(poke) (byte @ 2#B) & 0x80
0

Turns out the answer is no.

The map operator can also be used at the left side of an assignment operator:

(poke) byte @ 0x28#B = 0xff

Which reads “assign 0xff to the byte at offset 0x4a bytes”. Dumping again, we can verify that the byte actually changed:

76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 f700 0102 0000 0000 0000 0000 0000  ................
00000020: 0102 0000 0000 0000 ff01 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0800 0700  ..............

Does this mean that foo.o changed accordingly, in disk? The answer is yes. poke always commits changes immediately to the file being edited. This, that is an useful feature, can also be a bit tricky if you forget about it, leading to data corruption, so please be careful.

Incidentally, altering the byte at offset 0x28 most probably have caused foo.o to stop being a valid ELF file, but since we are just editing bytes (and not ELF structures) we actually don’t care much.

2.5 Values and Variables

Up to now we have worked with byte values, either writing them in the REPL or mapping them at the current IO space. Often it is useful to save values under meaningful names, and access to them by name. In poke we do that by storing the values in variables.

Before being used, variables shall be defined using the var construction. Let’s get the byte at offset 64 bytes and save it in a variable called foo:

(poke) var foo = byte @ 64#B

This defines a new variable (foo) and initializes it to the value of the byte at offset 64 bytes. This results on foo to hold the value 37.

Once defined, we can get the value of a variable by just giving its name to poke:

(poke) foo
37UB

Several variables can be defined in a single var declaration. Each definition is separated by a comma. This is equivalent of issuing several vars:

(poke) var a = 10, b = 20

In general, a variable containing a byte value can be used in any context where the contained value would be expected. If we wanted to check the highest bit in the byte value stored in foo we would do:

(poke) foo & 0x80
0x0

Assigning a value to a variable makes the value the new contents of the variable. For example, we can increase the value of foo by one like this:

(poke) foo = foo + 1

At this point, an important question is: when we change the value of the variable foo, are we also changing the value of the byte stored in foo.o at offset 64 bytes? The answer is no. This is because when we do the mapping:

(poke) var foo = byte @ 64#B

The value stored in foo is a copy of the value returned by the map operator @. You can imagine the variable as a storage cell located somewhere in poke’s memory. After the assignment above is executed there are two copies of the byte value 0x25: one in foo.o at offset 64 bytes, and the other in the variable foo.

It follows that if we wanted to increase the byte in the file, we would need to do something like:

(poke) var foo = byte @ 64#B
(poke) foo = foo + 1
(poke) byte @ 64#B = foo

Or, more succinctly, omitting the usage of a variable:

(poke) byte @ 64#B = (byte @ 64#B) + 1

Or, even more succinctly:

(poke) (byte @ 64#B) += 1

Note how we have to use parenthesis around the map at the right hand side, because the map operator @ has less precedence than the plus operator +.

2.6 From Bytes to Integers

The bytes we have been working with are unsigned whole numbers (or integers) in the range 0..255. We saw how poke sees the contents of the files as a sequence of bytes, and how each byte can be addressed using an offset. Mapping bytes using the map operator @ gives us these values, which are denoted in poke with literals like 10UB or 0x0aUB.

This very limited range of values have consequences when it comes to do arithmetic with bytes. Suppose for example we wanted to calculate the average of the first byte values stored in foo.o. We could do something like:

(poke) a0 = byte @ 0#B
(poke) a1 = byte @ 1#B
(poke) a2 = byte @ 2#B
(poke) a0
0x7fUB
(poke) a1
0x45UB
(poke) a2
0x4cUB
(poke) (a0 + a1 + a2) / 3UB
5UB

That is obviously the wrong answer. What happened? Let’s do it step by step. First, we add the first two bytes:

(poke) a0 + a1
0xc4UB

Which is all right. 0xc4 is 0x7f plus 0x45. But, let’s add now the third byte:

(poke) a0 + a1 + a2
0x10UB

That’s no good. Adding the value of the third byte (0x4c) we overflow the range of valid values for a byte value. The calculation went banana at this point.

Another obvious problem is that we surely will want to store integers bigger than 255 in our files. Clearly we need a way to encode them somehow, and since all we have in a file are bytes, the integers will have to be composed of them.

Integers bigger than 255 can be encoded by interpreting consecutive byte values in a certain way. First, let’s consider a single byte. If we print a byte value using binary rather than decimal or hexadecimal, we will observe that eight bits are what it takes to encode the numbers between 0 and 0xff (255) using a natural binary encoding:

(poke) .set obase 2
(poke) 0UB
0b00000000UB
(poke) 0xFFUB
0b11111111UB

This is the reason why people say bytes are “composed” of eight bits, or that the width of a byte is eight bits. But this way of talking doesn’t really reflect the view that the operating system has of devices like files or memory buffers: both disk and memory controllers provide and consume bytes, i.e. little unsigned numbers in the range 0..255. At that level, bytes are indivisible. We will see later that poke provides ways to work on the “sub-byte” level, but that is just really an artifact to make our life easier: underneath, all that goes in and out are bytes.

Anyhow, if we were to “concatenate” the binary representation of two consecutive bytes, we would end with a much bigger range of possible numbers, in the range 0b00000000_00000000..0b11111111_11111111³, or 0x0000..0xffff in hexadecimal. poke provides a bit-concatenation operator ::: that does exactly that:

(poke) 0x1UB
0b00000001UB
(poke) 0x1UB ::: 0x1UB
0b0000000100000001UH

Note how the suffix of the resulting number is now UH. This indicates that the number is no longer a byte value: it is too big for that. The H in this new suffix means “half”, and it is a traditional way to call an integer that is encoded using two bytes, or 16 bits.

So, using our method of encoding bigger numbers concatenating bytes, what would be the “half” integer composed of two bytes at the beginning of foo.o?

(poke) .set obase 16
(poke) (byte @ 0#B):::(byte @ 1#B)
0x7f45UH

Now, let’s go back to the syntax we used to map a byte value. In the invocation of the map operator byte @ 0#B the operand at the left (in this case byte) tells the operator what kind of value to map. This is called a type specifier; byte is the type specifier for a single byte value, and byte[3] is the type specifier for a group of three byte values arranged in an array.

As it happens, byte is a synonym for another slightly more interesting type specifier: uint<8>. You can probably infer the meaning already: a byte is an unsigned integer which is 8 bits big. We can of course use this alternate specifier in a mapping operation, achieving exactly the same result than if we were using byte:

(poke) uint<8> @ 0#B
0x7fUB

You may be wondering: is it possible to use a similar type specifier for mapping bigger integers, like these “halves” that are composed of two bytes? Yeah, it is indeed possible:

(poke) uint<16> @ 0#B
0x7f45UH

Mapping an unsigned integer of 16-bits at the offset 0 gives us an unsigned “half” value, as expected.

You can easily build bigger and bigger numbers concatenating more and more bytes. Three bytes? sure:

(poke) uint<24> @ 0#B
(uint<24>) 0x7f454c

Note that in this case poke uses a prefix instead of a suffix to indicate that the given value is 24-bits long. Four bytes?

(poke) uint<32> @ 0#B
0x7f454c46U

Certain integer widths are so often used that easier-to-type synonyms for their type specifiers are provided. We already know byte for uint<8>. Similarly, ushort is a synonym for uint<16>, uint is a synonym for uint<32> and ulong is a synonym for uint<64>. Try them!

GNU poke supports integers up to eight bytes, i.e. up to 64-bits. This may change in the future, as we are planning to support arbitrarily large integers.

2.7 Big and Little Endians

When talking about whole numbers (integers) we should distinguish between their value (such as 123) and their written form that we would use when writing the number on a piece of paper, such as 123.

The written form of a number is composed of digits, arranged in certain order. We all know that the ordering of the digits in the written form of a number is important: if we write 123 we are referring to a different value than if we write 321. The mathematical reason for this is that depending on the position they occupy in the written form, each digit contributes with a different “weight” to the total value of the number. This is always the case, regardless of the numerical base used to denote the number.

For example, the value of the number 123 (whose written form is 123) is calculated as 1*10^2+2*10^1+3*10^0. If we swap the last two digits in the written form of the number, we have 1*10^2+3*10^1+2*10^0, which results in a different value: 132. When we consider other numerical bases, the bases in the polynomial change accordingly, but the correspondence between written form and value stands: for example, the value of 0x123 is calculated as 1*16^2+2*16^1+3*16^0.

The “higher” a digit is in the polynomial, the more significant it is, i.e. the more weight it has on the value of the number where it appears. In the written number 123, for example, the digit 1 is the most significant digit of the number, and the digit 3 is the least significant digit.

This distinction between the written form of a number and its value is very important. Just like in certain languages letters are read right-to-left (Arabic) or even down-to-up (Japanese) we could certainly conceive a language in which the digits of numbers were arranged from right-to-left instead of left-to-right. In such a language the written representation of 123 would be 321, not 123. In other words: the least significant digit would come first, not last, in the written form of the number.

Now when it comes to store numbers in computers, rather than writing them on a paper, the role of the paper is played by the computer’s memory, be it ephemeral (like RAM) or persistent (like a spinning hard disk or a Flash memory), which is organized as a sequence of bytes. Since we are composing numbers with bytes, it makes sense to have each byte to play the role of a digit in the written form of the bigger number. Since bytes can have values from 0 to 255, the base is 256. But what is the “written form” for our byte-composed numbers?

In the last section we tried to compose bigger integers by concatenating bytes together and interpreting the result. In doing so, we assumed (quite naturally) that in the written form of the resulting integer the bytes are ordered in the same order than they appear in the file, i.e. we assume that the written form of the number b1*256^2+b2*256^1+b3*256^0 would be b1b2b3, where b1, b2 and b3 are bytes. In other words, given a written form b1b2b3, b1 would be the most significant byte (digit) and b3 would be the least significant byte (digit). In our world of IO spaces, the “written form” is the disposition of the bytes in the IO space (file, memory buffer, etc) being edited.

That interpretation of the written form is exactly what the bit-concatenation operator implements:

(poke) dump :from 0#B :size 3#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c                                  .EL
(poke) var b1 = byte @ 0#B
(poke) var b2 = byte @ 1#B
(poke) var b3 = byte @ 2#B
(poke) b1:::b2:::b3
(uint<24>) 0x7f454c

However, much like in certain human languages the written form is read from right to left, some computers also read numbers from right to left in their “written form”. Actually, turns out that most modern computers do it like that. This means that, in these computers, given the written form b1b2b3 (i.e. given a file where b1 comes first, followed by b2 and then b3) the most significant byte is b3 and the least significant byte is b1. Therefore, the value of the number would be b3*256^2+b2*256^1+b3*256^0.

So, given the written form of a bigger number b1b2b3 (i.e. some ordering of bytes implied by the file they are stored in) there are at least two ways to interpret them to calculate the value of the number. When the written form is read from left to right, we talk about a big endian interpretation. When the written form is read from right to left, we talk about a little endian interpretation.

Given the first three bytes in foo.o, we can determine the value of the integer composed of these three bytes in both interpretations:

(poke) b1:::b2:::b3
(uint<24>) 0x7f454c
(poke) b3:::b2:::b1
(uint<24>) 0x4c457f

Remember how the type specifier byte is just a synonym of uint<8>, and how we can use type specifiers like uint<24> and uint<32> to map bigger integers? When we do that, like in:

(poke) uint<24> @ 0#B
(uint<24>) 0x7f454c

Poke should somehow decide what kind of interpretation to use, i.e. how to read the “written form” of the number. As you can see from the example, poke uses the left-to-right interpretation, or big-endian, by default. But you can change it using a new dot-command: .set endian:

(poke) .set endian little
(poke) uint<24> @ 0#B
(uint<24>) 0x4c457f

The currently used interpretation (also called endianness) is shown if you invoke the dot-command without an argument⁴:

(poke) .set endian
little

Different systems use different endianness. Into a given system, it is to be expected that most files will be encoded following the same conventions. Therefore poke provides you a way to set the endianness to whatever endianness is in the system. You do it this way:

(poke) .set endian host

2.8 Negative Integers

2.8.1 Negative Encodings

Up to this point we have worked with unsigned integers, i.e. whole numbers which are zero or bigger than zero. Much like it happened with endianness, the interpretation of the value of several bytes as a negative number depends on the specific interpretation.

In computing there are two main ways to interpret the values of a group of bytes as a negative number: one’s complement and two’s complement.

At the moment GNU poke supports the two complement interpretation, which is really ubiquitous and is the negative encoding used by the vast majority of modern computers and operating systems.

We may consider adding support for one’s complement in the future, but only if there are real needs that would justify the effort (which wouldn’t be a small one ;)).

2.8.2 Signed Integers

Unsigned values are never negative. For example:

(poke) 0UB - 1UB
0xffUB

Instead of getting a -1, we get the result of an unsigned underflow, which is the biggest possible value for an unsigned integer of size 8 bits: 0xff.

When using type specifiers like uint<8> or uint<16> in a map, we get unsigned values such as 0UB. It follows that we need other type specifiers to map signed values. These look like int<8> and int<16>.

For example, let’s map a signed 16-bit value from foo.o:

(poke) .set obase 10
(poke) int<16> @ 0#B
28515H

Note how the suffix of the value is now H and not UH. This means that the value is signed! But in this case it is still positive, so let’s try to get an actual negative value:

(poke) var h = int<16> @ 0#B
(poke) h - h - 1H
-1H

2.8.3 Mixing Signed and Unsigned Integers

Adding two signed integers gives you a signed integer:

(poke) 1 + 2
3

Likewise, adding two unsigned integers results in an unsigned integer:

(poke) 1U + 2U
3U

But, what happens if we mix signed and unsigned values in an expression? Is the result signed, or unsigned? Let’s find out:

(poke) 1U + 2
3U

Looks like combining an unsigned value with a signed value gives us an unsigned value. This actually applies to all the operators that work on integer values: multiplication, division, exponentiation, etc.

What actually happens is that the signed operand is converted to an unsigned value before executing the expression. You can also convert signed values into unsigned values (and vice-versa) using cast constructions:

(poke) 2 as uint<32>
2U

Therefore, the expression 1U + 2 is equivalent to 1U + 2 as uint<32>:

(poke) 1U + 2 as uint<32>
3U

You may be wondering: why not doing it the other way around? Why not converting the unsigned operand into a signed value and then operate? The reason is that, given an integer of some particular size, the positive range that you can store in it is bigger when interpreted as an unsigned integer than when interpreted as a signed integer. Therefore, converting signed into unsigned before operating reduces the risk of positive overflow. This of course assumes that we, as users, will be working with positive numbers more often than with negative numbers, but that is a reasonable assumption to do, as it is often the case!

2.9 Weird Integers

Up to this point we have been playing with integers that are built using a whole number of bytes. However, we have seen that the type specifier for an integer has the form int<N> or uint<N> for signed and unsigned variants, where N is the width of the integer, in bits. We have used bit-sizes that are multiple of 8, which is the size of a byte. So, why is this so? Why is N not measured in bytes instead?

The reason is that poke is not limited to integers composed of a whole number of bytes. You can actually have integers having any number of bits, between 1 and 64. So yes, int<3> is a type specifier for signed 3-bit integers, and uint<17> is a type specifier for unsigned 7-bit integers.

We call integers like this weird integers.

The vast majority of programming languages do not provide any support for weird integers. In the few cases they do, it is often in a very limited and specific way, like bitmap fields in C structs. Such constructions are often vague, obscure, and often their semantics depend on the specific implementation of the language, and/or the characteristics of the system where you run your program.

In poke, on the contrary, weird numbers are first class citizens, and they don’t differ in any way from “normal” integers composed of a whole number of bytes. Their interpretation is also well defined, and they keep the same semantics regardless of the characteristics of the computer on which poke is running.

2.9.1 Incomplete Bytes

Let’s consider first weird numbers that span for more than one byte. For example, an unsigned integer of 12 bits. Let’s visualize the written form of this number, i.e. the sequence of its constituent bytes as they appear in the underlying IO space:

  byte 0  |  byte 1
+---------+----+----+
|::::::::::::::|    |
+---------+----+----+
|   uint<12>   |

All right, the first byte is used in its entirely, but only half of the second byte is used to conform the value of the number. The other half of the second byte has no influence of the value of the 12 bits number.

Now, we talk about the “second half of the byte”, but what do that means exactly? We know that bytes in memory and files (bytes in IO spaces) are indivisible at the system level: bytes are read and written one at a time, as little integers in the range 0..255. However, we can create the useful fiction that each byte is composed by bits, which are the digits in the binary representation of the byte value.

So, we can look at a byte as composed of a group of eight bits, like this:

           byte
+-------------------------+
| b7 b6 b5 b4 b3 b2 b1 b0 |
+-------------------------+

Note how we decided to number the bits in descending order from left to right. This is because these bits correspond to the base of the polynomial equivalent to the binary value of the byte, i.e. the value of the byte is b7*2^7+b6*2^6+b5*2^5+b4*2^4+b3*2^3+b2*2^2+b1*2^1+b0*2^0. In other words: at the bit level poke always uses a big endian interpretation, and the bit that “comes first” in this imaginary stream of bits is the most significant bit in the binary representation of the number. Please note that this is just a convention, set by the poke authors: the opposite could have been chosen, but it would have been a bit confusing, as we would have to picture binary numbers in reverse order!

With this new way of looking at bytes, we can now visualize what we mean exactly with the “first half” and “second half” of the trailing byte, in our 12 bits unsigned number:

           byte 0         |           byte 1
+-------------------------+-------------+-------------+
| a7 a6 a5 a4 a3 a2 a1 a0   b7 b6 b5 b4 :             |
+-------------------------+-------------+-------------+
|                uint<12>               |

Thus the first half of byte 1 is the sequence of bits b7 b6 b5 b4. The second half, which is not pictured since it doesn’t contribute to the value of the number, would be b3 b2 b1 b0.

So what would be the value of the 12-bit integer? Exactly like with non-weird numbers, this depends on the current selected endianness, which determines the ordering of bytes.

If the current endianness is big, then byte 0 provides the most significant bits of the result number, and the used portion of byte 1 provides the least significant bits of the result number:

0b a7 a6 a5 a4 a3 a2 a1 a0 b7 b6 b5 b4

However, if the current selected endianness is little, then the used portion of byte 1 provides the most significant bits of the result number, and byte 0 provides the least significant bits of the result number:

0b b7 b6 b5 b4 a7 a6 a5 a4 a3 a2 a1 a0

Let’s see this in action. Let’s take a look to the value of the first two bytes in foo.o, in binary:

(poke) .set obase 2
(poke) byte @ 0#B
0b01111111UB
(poke) byte @ 1#B
0b01000101UB

Looking at these bytes as sequences of bits, we have:

        byte @ 0#B       |        byte @ 1#B
+-------------------------+-------------+-------------+
|  0  1  1  1  1  1  1  1    0  1  0  0 :  0  1  0  1
+-------------------------+-------------+-------------+
|                uint<12>               |

Let’s map our weird number at offset 0 bytes, using big endian:

(poke) .set endian big
(poke) uint<12> @ 0#B
(uint<12>) 0b011111110100

That matches what we explained before: the most significant bits of the unsigned 12 bits number come from the byte at offset 0, i.e. 01111111, whereas the least significant bits come from the byte at offset 1, i.e. 0100.

Now let’s map it using little endian:

(poke) uint<12> @ 0#B
(uint<12>) 0b010001111111

This time the most significant bits of the unsigned 12 bits number come from the byte at offset 1, i.e. 0100, whereas the least significant bits come from the byte at offset 0, i.e. 01111111.

An important thing to note is that non-weird numbers, i.e. numbers built with a whole number of bytes, are basically a particular case of weird numbers where the last byte in the written form (in the IO space) provides all its bits. The rules are exactly the same in all cases, which makes it easy to obtain predictable and natural results when building integers using poke.

2.9.2 Quantum Bytenics

The second kind of weird numbers are integers using less than 8 bits. These “sub-byte” numbers do not use all the bytes of their containing byte. Consider for example the written form of an unsigned integer of size 5 bits:

    byte
+-----+----+
|:::::|    |
+-----+----+
uint<5>

Now let’s view the byte as a sequence of bits:

             byte
+----------------+----------+
| b7 b6 b5 b4 b3 |          |
+----------------+----------+
|    uint<5>     |

What is the value of this number? Applying the general rules for building integers from bytes, we can easily see that regardless of the current endianness the value, in binary, is:

0b b7 b6 b5 b4 b3

Let’s see this in poke:

(poke) .set obase 2
(poke) .set endian big
(poke) byte @ 0#B
0b01111111UB
(poke) uint<5> @ 0#B
(uint<5>) 0b01111
(poke) .set endian little
(poke) uint<5> @ 0#B
(uint<5>) 0b01111

2.9.3 Signed Weird Numbers

In the section discussing negative integers, we saw how the difference between a signed number and an unsigned number is basically a different interpretation of the most significant byte. Exactly the same applies to weird numbers.

Let’s summon our unsigned 12-bit integer at the beginning of the file foo.o:

(poke) .set endian big
(poke) uint<12> @ 0#B
(uint<12>) 0b011111110100

The most significant byte of the resulting value (not of its written form) indicates that this number would be positive if we were mapping the corresponding signed value. Let’s see:

(poke) int<12> @ 0#B
(int<12>) 0b010001111111
(poke) .set obase 10
(poke) int<12> @ 0#B
(int<12>) 1151

Let’s make it a bit more interesting, and change the value of the first byte in the file so we get a negative number:

(poke) .set obase 2
(poke) byte @ 0#B = 0b1111_1111
(poke) int<12> @ 0#B
(int<12>) 0b111111110100
(poke) .set obase 10
(poke) int<12> @ 0#B
(int<12>) -12

Now, let’s switch to little endian:

(poke) .set endian little
(poke) .set obase 2
(poke) int<12> @ 0#B
(int<12>) 0b010011111111
(poke) .set obase 10
(poke) int<12>  0#B
(int<12>) 1279

2.10 Unaligned Integers

We have mentioned above that the data stored in computers, that we edit with poke, is arranged as a sequence of bytes. The entities we edit with poke (that we call IO devices) are presented to us as IO spaces. Up to now, we have accessed this IO space in terms of bytes, in commands like dump :from 32#B and in expressions like 2UB + byte @ 0#B. We said that mapped integers are built from bytes read from the IO space.

However, the IO space that poke offers to us is actually a space of bits, not a space of bytes, and the poke values are mapped on this space of bits. The following figure shows this:

poke values      |        uint<16> @ 2#b         |
-----------      |                               |
IO space     |b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|b|
-----------  |               |               |               |
IO device    |     byte0     |     byte1     |     byte2     |

The main consequence of this, that you can see in the figure above, is that we can use offsets in mapping operations that are not aligned to bytes. You can specify an offset in bits, instead of bytes, using the #b suffix instead of #B. Little b means bits, and big B means bytes.

Let’s map an unaligned 16 bit unsigned integer in foo.o:

(poke) dump :from 0#B :size 3#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c                                  .EL
(poke) .set obase 2
(poke) byte @ 0#B
0b01111111UB
(poke) byte @ 1#B
0b01000101UB
(poke) byte @ 2#B
0b01001100UB
(poke) .set endian big
(poke) uint<16> @ 2#b
0b1111110100010101UH

Graphically:

poke values      |        uint<16> @ 2#b         |
-----------      |                               |
IO space     |0|1|1|1|1|1|1|1|0|1|0|0|0|1|0|1|0|1|0|0|1|1|0|0|
-----------  |               |               |               |
IO device    |     0x7f      |      0x45     |      0x4c     |

These three levels of abstractions make it very easy and natural to work with unaligned data. Imagine for example that you are poking packages in a network protocol that is bit-oriented. This means that the packages will generally not be aligned to byte boundaries, but still the payload stored in the packages contains integers of several sizes. poke allows you to directly map these integers as if they were aligned to byte boundaries, and work with them.

However, when one tries to determine the correspondence between a given poke value and the underlying bytes in the IO device, things can get complicated. This is particularly true when we map what we called “weird numbers”, i.e. numbers with partial bytes. As we saw, the rules to build these numbers were expressed in terms of bytes.

In order to ease the visualization of the process used to build integer values (especially if they are weird numbers, i.e. integers with partial bytes) one can imagine an additional layer of “virtual bytes” above the space of bits provided by the IO space. Graphically:

poke values       |        uint<16> @ 2#b         |
-----------       |                               |
Virtual bytes     | virt. byte1   |  virt. byte2  |
-----------       |               |               |
IO space      |0|1|1|1|1|1|1|1|0|1|0|0|0|1|0|1|0|1|0|0|1|1|0|0|
-----------   |               |               |               |
IO device     |     0x7f      |      0x45     |      0x4c     |

It is very important to understand that the IO space is an abstraction provided by poke. The underlying file, or memory buffer, or whatever, is actually a sequence of bytes; poke translates the operations on integers, bits, bytes, etc into the corresponding byte operations, which are far from trivial. Fortunately, you can let poke to do that dirty job for you, and abstract yourself from that complexity.

2.11 Integers of Different Sizes

When integer values of different sizes are passed to an arithmetic or relational operator, the “smaller” operand gets converted into the size of the “bigger” operand. For example:

(poke) 1H + 2
3

The operands are of size 16-bit and 32-bit respectively, and the result is a 32-bit integer. This is equivalent to:

(poke) 1H as int<32> + 2
3

2.12 Offsets and Sizes

Early in the design of what is becoming GNU poke I was struck by a problem that, to my surprise, would prove not easy to fix in a satisfactory way: would I make a byte-oriented program, or a bit-oriented program? Considering that the program in question was nothing less than an editor for binary data, this was no petty dilemma.

Since the very beginning I had a pretty clear idea of what I wanted to achieve: a binary editor that would be capable of editing user defined data structures, besides bytes and bits. I also knew I needed some sort of domain specific language to describe these structures and operate on them. How that language would look like, and what kind of abstractions it would provide, however, was not clear to me. Not at all.

So once I sketched an initial language design, barely something very similar to C structs, I was determined to not continue with the poke implementation until I had described as many as binary formats in my language as possible. That, I reckoned, was the only way to make sure the implemented language would be expressive, complete and useful enough to fulfil my requirements.

The first formats I implemented using my immature little language included ELF, FLV, MP3, BSON… of them describing structures based on whole bytes. Even when they try to be compact, it is always by packing bit-fields in elements that are, invariably, sized as a multiple of bytes. Consequently, the language I was evolving became byte oriented as well. No doubt also influenced by my C inheritance, I would think of bit-fields either as a sort of second class citizen, or as mere results of shifting and masking.

This worked well. The language evolved to be able to express many different aspects of these formats in a very nice way, like variable-length data and holes in structures. Consider the following definition, which is not valid in today’s Poke:

type Data =
  struct
  {
    byte magic;
    byte count;
    byte dstart;

    byte[count] data @ dstart;
  };

The data starts with a byte that is a magic number. Then the size of the data stored, in bytes, and then the data itself. This data, however, doesn’t start right after dstart: it starts at dstart, which is expressed as an offset, in bytes, since the beginning of the Data. I conceived struct field labels to be any expression evaluating to an integer, which would be…, bytes obviously.

Then, one day, it was the turn for IETF RFC1951, which is the specification of the DEFLATE algorithm and associated file format. Oh dear. Near the beginning of the spec document it can be read:

This document does not address the issue of the order in which bits of a byte are transmitted on a bit-sequential medium, since the final data format described here is byte- rather than bit-oriented. However, we describe the compressed block format in below, as a sequence of data elements of various bit lengths, not a sequence of bytes.

Then it goes on describing rules to pack the DEFLATE elements into bytes. I was appalled, and certainly sort of deflated as well. The purpose of my program was precisely to edit binary in terms of the data elements described by a format. And in this case, these data elements came in all sort of bit lengths and alignments. This can be seen in the following RFC1951 excerpt, that describes the header of a compressed block:

Each block of compressed data begins with 3 header bits containing the following data:

first bit       BFINAL
next 2 bits     BTYPE

Note that the header bits do not necessarily begin on a byte boundary, since a block does not necessarily occupy an integral number of bytes.

At this point I understood that my little language on the works would never be capable to describe the DEFLATE structures naturally: C-like bit-fields, masking and shifting, all based on byte-oriented containers and boundaries, would never provide the slickness I wanted for my editor. I mean, just use C and get done with it.

This pissed me off. Undoubtedly other formats and protocols would be impacted in a similar way. Even when most formats are byte oriented, what am I supposed to tell to the people hacking bit-oriented stuff? “Sorry pal, this is not supported, this program is not for you”? No way, I thought, not on my watch.

The obvious solution for the problem, is to be general. In this case, to express every offset and every memory size in bits instead of bytes. While this obviously gives the language maximum expressiveness power, and is ideal for expressing the few bit-oriented formats, it has the disadvantage of being very inconvenient for most situations.

To see how annoying this is, let’s revisit the little Data element we saw above. In a bit-oriented description language, we would need to write something like:

type BitData =
  struct
  {
    byte magic;
    byte count;
    byte dstart;

    byte[count] data @ dstart * 8;
  };

Yeah… exactly. The * and 8 keys in the keyboards of the poke users would wear out very fast, not to mention their patience as well. Also, should I provide both sizeof and bitsizeof operators? Nasty.

I am very fond of the maxim “Never write a program you would never use yourself”⁵, so I resigned myself to make GNU poke byte oriented, and to provide as many facilities for operating on bit-fields as possible.

Fortunately, I have smart friends…

During one of the Rabbit Herd’s Hacking Weekends⁶ I shared my frustration and struggle with the other rabbits, and we came to realize that offsets and data sizes in Poke should not be pure magnitudes or mere integer values: they should be united. They should have units.

It makes full sense when you come to think about it. For a program like poke, it is only natural to talk about different memory units, like bits, bytes, kilobytes, megabits, and so on. Bits and bytes are just two common units. Apart from allowing me to express values in different units, this approach also has other benefits as we will see shortly.

I’m really grateful to Bruno Haible, Luca Saiu and Nacho Gonzalez for putting me on the right track.

2.13 Buffers as IO Spaces

We have mentioned already that files are not the only entities that can be edited using poke. Remember the dot-command .file that opened a file as an IO space?

(poke) .file foo.o
The current IOS is now `./foo.o'.
(poke) .info ios
  Id	Type	Mode	Size		Name
* #0	FILE	rw	0x000004c8#B	./foo.o

Memory buffers can be created using a similar dot-command, .mem:

(poke) .mem foo
The current IOS is now `*foo*'.
(poke) .info ios
  Id	Type	Mode    Size            Name
* #1	MEMORY		0x00001000#B    *foo*
  #0	FILE	rw	0x000004c8#B    ./foo.o

Note how the name of the buffer is built by prepending and appending asterisks. Therefore, the name of the buffer created by the command .mem foo is *foo*. Note also that the new buffer is created with a default size of 0x1000 bytes, or 4096 bytes. The contents of the buffer are zeroed:

(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000030: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000040: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000050: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000060: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000070: 0000 0000 0000 0000 0000 0000 0000 0000  ................

Memory buffer IO spaces grow automatically when a value is mapped beyond their current size. This is very useful when populating newly created buffers. However, for security reasons, there is a limit: the IO spaces are only allow to grow 4096 bytes at a time.

When it comes to map values, there is absolutely no difference between an IO space backed by a file and an IO space backed by a memory buffer. Exactly the same rules apply in both cases.

2.14 Copying Bytes

Memory buffer IO spaces are cheap, and they are often used as “scratch” areas.

Suppose for example we want to experiment with the ELF header of foo.o. We could open it in poke:

(poke) .file foo.o
The current IOS is now `./foo.o'.

The header of an ELF file comprises the first 64 bytes in the file:

(poke) dump :size 64#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 3e00 0100 0000 0000 0000 0000 0000  ..>.............
00000020: 0000 0000 0000 0000 0802 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0b00 0a00  ..............

We know that as soon as we poke something on an IO space, the underlying file is immediately modified. So if we start playing with foo.o’s header, we may corrupt the file. We could of course make a copy of foo.o and work on the copy, but it is much better to create a memory IO space and copy the ELF header there:

(poke) .mem scratch
The current IOS is now `*scratch*'.
(poke) copy :from_ios 0 :from 0#B :size 64#B
(poke) dump :size 64#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 7f45 4c46 0201 0100 0000 0000 0000 0000  .ELF............
00000010: 0100 3e00 0100 0000 0000 0000 0000 0000  ..>.............
00000020: 0000 0000 0000 0000 0802 0000 0000 0000  ................
00000030: 0000 0000 4000 0000 0000 4000 0b00 0a00  ..............

The command copy above copies 64 bytes starting at byte 0 from the IO with id 0 (the file foo.o) to the current IO space (the buffer *scratch*).

Once we are done working with the copy of the ELF header, and satisfied, we can copy it back to the file and close the memory IO space:

(poke) copy :from 0#B :size 64#B :to_ios 0
(poke) .close
The current IOS is now `./foo.o'.

Note how the command arguments :from_ios and :to_ios are assumed to be the current IO space if they are not explicitly specified in the command invocation. For detailed information on the copy command, see copy.

2.15 Saving Buffers in Files

Another useful command when working with buffer IO spaces (and in general, with any IO space) is save. Let’s say we want to save a copy of the header of an ELF file in another file. We could do it the pokeish way:

$ poke foo.o
[…]
(poke) save :from 0#B :size 64#B :file "header.dat"

The command above saves the first 64 bytes in the current IO space (which is backed by the file foo.o) into a new file header.dat in the current working directory.

Let’s now consider again the scenario where we are using a memory IO space as a scratch area. It is late in the night and we are tired, so we would like to save the contents of the scratch buffer to a file, so we can continue our work tomorrow. This is how we would do that:

(poke) .info ios
  Id    Type	Mode    Size            Name
* #1	MEMORY		0x00001000#B    *scratch*
  #0	FILE	rw	0x000f4241#B    ./foo.o
(poke) save :from 0#B :size iosize (1) :file "scratch.dat"

Here we used the built-in function iosize, that given an IO space identifier returns its size.

2.16 Character Sets

Computers understand text as a sequence of codes, which are numbers identifying some particular character. A character can represent things like letters, digits, ideograms, word separators, religious symbols, etc. Collections of character codes are called character sets.

Some character sets try to cover one or a few similar written languages. This is the case of ASCII and ISO Latin-1, for example. These character sets are small, i.e. just a few hundred codes.

Other character sets are much more ambitious. This is the case of Unicode, that tries to cover the entire totality of human languages in the globe, including the fictitious ones, like klingon. Unicode is a really big character set.

In order to store character codes in a computer’s memory, or a file, we need to encode each character code in one or more bytes. The number of bytes needed to encode a given character code depends on the range of codes in the containing set.

ASCII, for example, defines 128 character codes: a single byte is enough to encode every possible ASCII character. It is very easy to encode ASCII.

Unicode, on the contrary, defines many thousand of character codes, and has room for many more: we would need 31 bits in order to encode any conceivable Unicode character code. However, it would be wasteful to use that many bits per character: most used character codes tend to be in lower regions of the code space. For example, the code corresponding to the Latin letter 'a' is a fairly small number, whereas the codes corresponding to the Klingon alphabet are really big numbers. Consequently, some systems opt to just encode a subset of Unicode, like the first 16 bits of the Unicode space, which is called the Basic Multilingual Plane, and contains all the characters that most people will ever need. There are also variable-length encodings of Unicode, that try to use as less bytes as possible to encode any given code. A particularly clever encoding of Unicode, designed by Rob Pike, is backwards compatible with the ASCII encoding, i.e. it encodes all the ASCII codes in one byte each, and the values of these byte are the same than in ASCII. This clever encoding is called UTF-8.

2.17 From Bytes to Characters

2.17.1 Character Literals

poke has built-in support for ASCII, and its simple encoding: each ASCII code is encoded using one byte. Let’s see:

(poke) 'a'
0x61UB

We presented poke with the character a, and it answered with its corresponding code in the ASCII character set, which is 0x61. In fact, ’a’ and 0x61UB are just two ways to write exactly the same byte value in poke:

(poke) 'a' == 0x61UB
1
(poke) 'a' + 1
0x62U

In order to make this more explicit, poke provides yet another synonym for the type specifier uint<8>: char.

2.17.2 Classifying Characters

When working with characters it is very useful to have some acquaintance of the ASCII character set, which is designed in a very clever way with the goal of easing certain code calculations. See Table of ASCII Codes for a table of ASCII codes in different numeration bases.

Consider for example the problem of determining whether a byte we map from an IO space is a digit. Looking at the ASCI table, we observe that digits all have consecutive codes, so we can do:

(poke) var b = byte @ 23#B
(poke) b >= '0' && b <= '9'
1

Now that we know that b is a digit, how could we calculate its digit value? If we look at the ASCII table again, we will find that the character codes for digits are not only consecutive: they are also ordered in ascending order 0, 1, … Therefore, we can do:

(poke) b
0x37UB
(poke) '0' - b
7UB

b contains the ASCII code 0x37UB, which corresponds to the character 7, which is a digit.

How would we check whether b is a letter? Looking at the ASCII table, we find that lower-case letters are encoded consecutively, and the same applies to upper-case letters. This leads to repeat the trick again:

(poke) (b >= 'a' && b <= 'z') || (b >= 'A' && b <= 'Z')
0

2.17.3 Non-printable Characters

Not all ASCII code are printable using the glyph that are usually supported in terminals. If you look at the table in Table of ASCII Codes, you will find codes for characters described as “start of text”, “vertical tab”, and so on.

These character codes, which are commonly known as non-printable characters, can be represented in poke using its octal code:

(poke) '\002'
0x2UB

This is of course no different than using 2UB directly, but in some contexts the “character like” notation may be desirable, to stress the fact that the byte value is used as an ASCII character.

Some of the non-printable characters also have alternative notations. This includes new-line and horizontal tab:

(poke) '\n'
0xaUB
(poke) '\t'
0x9UB

These \ constructions in character literals are called escape sequences. See Characters for a complete list of allowed escapes in character literals.

2.18 ASCII Strings

2.18.1 String Values

Now that we know how to manipulate ASCII codes in poke, we may wonder how can we combine them to conform words or, more generally, strings of ASCII characters.

GNU poke has support for native ASCII string values. The characters conforming the string are written between " and " characters, like in:

(poke) "word"
"word"

Note, and this is important, that string values are as atomic as integer values: they are not really composite values. The fact that "word" contains an r at position 3 is like the fact that the value 123 contains a digit 2 at position 2.

Like in character literals, poke strings support several escape sequences that help to denote non-printed characters, such as new lines and horizontal tabs. See String Literals.

2.18.2 Poking Strings

Let’s start with a fresh memory buffer IOS *scratch*:

(poke) .mem scratch
The current IOS is now `*scratch*'.
(poke) dump :size 48#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................

If we wanted to, somehow, store the word word in this IO space, encoded in ASCII, we could proceed as:

(poke) char @ 0x12#B = 'w'
(poke) char @ 0x13#B = 'o'
(poke) char @ 0x14#B = 'r'
(poke) char @ 0x15#B = 'd'
(poke) dump :size 48#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000010: 0000 776f 7264 0000 0000 0000 0000 0000  ..word..........
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................

This worked. The ASCII part of the dump output, which interprets the bytes as ASCII, clearly shows the word word at the offset where we poked the character values. However, we can do better: string values can be mapped themselves.

String values use the type specifier string. As any other kind of value in poke, they can be mapped from an IO space:

(poke) string @ 0x12#B
"word"

Clearly that is the string resulting from the concatenation of the character values that we poked before.

The question now is: how did poke know that the last character of the string was the d at offset 0x15#B? The fact the character code 0 (also known as the NULL character) at offset 0x16#B is non-printable, doesn’t imply it is not part of the ASCII character set. Clearly, we have to pick an ASCII code and reserve it to mark the end of strings. Like the C programming language, and countless formats and systems do, poke uses the NULL character to delimit strings.

Now consider this:

(poke) "word"'length
4UL
(poke) "word"'size
40UL#b

Using the length and size attributes, poke tells us that the length of the string "word" is 4 characters, but the size of the string value is 40 bits, or 5 bytes. Why this discrepancy? The size value attribute tells how much storage space a given value required once mapped to an IO space, and in the case of strings it should count the space occupied by the terminating NULL character.

Poking string values on the IO space is as straightforward as poking integers:

(poke) string @ 0x22#B = "WORD"
(poke) dump :size 48#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000010: 0000 776f 7264 0000 0000 0000 0000 0000  ..word..........
00000020: 0000 574f 5244 0000 0000 0000 0000 0000  ..WORD..........

2.18.3 From Characters to Strings

Strings can be concatenated using the string-concatenation + operators:

(poke) "foo" + "bar"
"foobar"

The string resulting from the operation above has length 6 characters, and size 7 bytes. The terminating NULL character of "foo" is lost in the operation. This is easily seen:

(poke) "foo"'size + "bar"'size
0x40UL#b
(poke) ("foo" + "bar")'size
0x38UL#b

The string concatenation operator requires two strings. Therefore, if we wanted to append a character to a string, we would get an error:

(poke) "Putin" + 'a'
<stdin>:1:1: error: invalid operands in expression
"Putin" + 'a';
^~~~~~~~~~~~~

It is possible to transform a character value (i.e. a byte value) into a string composed of that character using a cast:

(poke) 'a' as string
"a"

Using that cast, we can now append:

(poke) "Putin" + 'a' as string
"Putina"

2.19 From Strings to Characters

Despite being atomic values, poke strings can be indexed in order to retrieve individual characters:

(poke) "word"[2]
0x72UB

Note how the indexing is zero-based, i.e. the first position in the string is referred as [0], the second position with [1], and so on.

If you specify a negative index, or an index that is too big, you will get an error:

(poke) "word"[-1]
<stdin>:1:8: error: index is out of bounds of string
"word"[-1];
       ^
(poke) "word"[5]
<stdin>:1:8: error: index is out of bounds of string
"word"[5];
       ^

3 Structuring Data

In the previous chapter we learned how to manipulate pre-defined entities like bytes, integers and strings. One of the big points of poke, however, is that it allows you to define your own data abstractions, and to operate in term of them. This is achieved by defining data structures.

3.1 The SBM Format

3.1.1 Images as Maps of Pixels

There are two main ways to store two-dimensional images in computers.

One is to explicitly store the different pixels that compose the image. In these bitmaps, the pixels are arranged sequentially and implicitly organized into lines. A header typically provides information to determine how many pixels fit in each line:

 <--- line_width  --->
| pixel | pixel | … | pixel | pixel | … | …
|        line 1       |         line 2      |

Several properties have to be encoded for each pixel, depending on the sophistication of the image: for monochrome images each pixel can be just either switched on or off, so a single bit could be used to encode each pixel (this is the origin of the term “bitmap”). When color is added, a bit is no longer sufficient: the color of the pixel shall be encoded in some way, typically using a color schema such as RGB, that requires triplet of little integers to be encoded. If transparency is to be supported, the degree of transparency of the pixel shall also be encoded somehow.

The other way to store an image is to store a functional description of the “painted” parts of the image. This functional description usually contains instructions like “paint a line of thickness 1 and color red from the coordinate 10,20 to coordinate 10,40”. Once executed with certain parameters (like the desired resolution) the functional description generates a bitmap. Image formats using this approach are commonly known as vectorial formats.

When it comes to bitmaps, there are a plethora of different formats out there (bmp, jpg, png) competing in terms of capabilities such as higher color depths, better resolution, support for transparency (alpha channels), higher compression level, and the like. These capabilities greatly complicate these formats, but ultimately any bitmap can be reduced to a sequence of pixels, which can be further structured in terms of lines.

We don’t know enough poke yet to handle the complications of these real-life bitmap formats, so in subsequent sections we introduce a very simple format for bitmaps, the Stupid BitMap format, or simply SBM, that we will use for learning purposes.

But please do not feel disappointed: later in this book, when we become more proficient pokers, we will be messing with these shiny complex formats as well :)

3.1.2 SBM header

A SBM file starts with a header that contains a “magic number” composed of the three bytes ’S’ (0x53UB), ’B’ (0x42UB) and ’M’ (0x4dUB). The next two bytes indicate the number of pixels per line, and the number of lines, respectively. In summary:

                SBM header
+-------+-------+-------+-------+-------+
|  'S'  |  'B'  |  'M'  |  ppl  | lines |
+-------+-------+-------+-------+-------+
  byte0   byte1   byte2   byte3   byte4

Here ppl stands for pixels-per-line. The encoding of these fields implies that the bigger image that can be encoded in SBM has dimensions 255x255.

3.1.3 SBM data

Albeit stupid, SBM is a colorful format. It supports more than a million colors, encoding each color in what is known as RGB24. In RGB24, each color is encoded using three little integers, specifying the amount of red, green and blue that, added together, compose the color.

The name RGB24 comes from the fact that each color is encoded using three bytes, or 24 bits. Therefore, each pixel in a SBM image is encoded using three consecutive bytes:

        SBM pixel
+-------+-------+-------+
|  Red  | Green | Blue  |
+-------+-------+-------+
  byte0   byte1   byte2

Each byte in this encoding determines the amount of the base color (red, green or blue) that compose the color. We will talk more about these components later.

3.2 Poking a SBM Image

3.2.1 P is for poke

Let’s compose our first SBM image, using poke. The image we want to encode is the very simple rendering of the letter P shown in the figure below.

  | 0 | 1 | 2 | 3 | 4 |
  +---+---+---+---+---+
0 |   | * | * |   |   |
1 |   | * |   | * |   |
2 |   | * |   | * |   |
3 |   | * | * |   |   |
4 |   | * |   |   |   |
5 |   | * |   |   |   |
6 |   | * |   |   |   |

The image has seven lines, and there are five pixels per line, i.e. the dimension of the image in pixels is 5x7. Also, the pixels denoted by asterisks are red, whereas the pixels denoted with empty spaces are white. In other words, our image uses a red foreground over a write background. The “painted” pixels are called foreground pixels, the non painted pixels are called background pixels.

3.2.2 Preparing the Canvas

The first thing we need is some suitable IO space where to encode the image. Let’s fire up poke and create a memory buffer:

$ poke
[…]
(poke) .mem image
The current IOS is now `*image*'.
(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000030: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000040: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000050: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000060: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000070: 0000 0000 0000 0000 0000 0000 0000 0000  ................

Freshly created memory IO spaces are 4096 bytes long, and that’s big enough for our little image. If we wanted to work with more data, remember that memory IO spaces will grow automagically when poked past their size.

3.2.3 Poking the Header

The first three bytes of the header of a SBM file contains the magic number that identifies the file as a SBM bitmap. We can poke these bytes very easily:

(poke) byte @ 0#B = 'S'
(poke) byte @ 1#B = 'B'
(poke) byte @ 2#B = 'M'

The next couple of bytes encode the dimensions of the bitmap, in this case 5x7:

(poke) byte @ 3#B = 5
(poke) byte @ 4#B = 7

There is something worth noting in this last mapping. Even tough we were poking bytes (passing the byte type specifier to the map operators) we specified the 32-bit signed integers 5 and 7 instead of 5UB and 7UB. When poke finds a situation like this, where certain kind of integers are expected but other kind are provided, it converts the value from the provided type to the expected type. This conversion may result in truncation (think about converting, say 0xfff to an unsigned byte, whose maximum possible value is 0xff) but certainly not in the case at hands.

The final header looks like:

(poke) dump :size 16#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 5342 4d05 0800 0000 0000 0000 0000 0000  SBM.............

3.2.4 Poking the Pixels

Now that we have written a SBM header, we have to encode the sequence of pixels composing the image.

Recall that every pixel is encoded using three bytes, that conform a RGB24 color. We have two kinds of pixels in our image: white pixels, and red pixels. In RGB24 white is encoded as (255,255,255). Pure red is encoded as (255,0,0), but to make things more interesting we will be using a nicer tomato-like red (255,99,71).

Therefore, poking a white pixel at some offset offset would involve the following operations:

(poke) byte @ offset = 255
(poke) byte @ offset+1#B = 255
(poke) byte @ offset+2#B = 255

Likewise, the operations to poke a tomato pixel would look like:

(poke) byte @ offset = 255
(poke) byte @ offset+1#B = 99
(poke) byte @ offset+2#B = 71

To ease things a bit, we can define variables with the color components for both foreground and background pixels:

(poke) var bg1 = 255
(poke) var bg2 = 255
(poke) var bg3 = 255
(poke) var fg1 = 255
(poke) var fg2 = 99
(poke) var fg3 = 71

Then to poke a foreground pixel would involve doing:

(poke) byte @ offset = fg1
(poke) byte @ offset+1#B = fg2
(poke) byte @ offset+2#B = fg3

At this point, you may feel that the perspective of mapping the pixels of our image is not very appealing, considering we have 5x7 = 35 pixels in our image. We will need to poke 35 * 3 = 105 bytes. We may feel tempted to, somehow, use a bigger integer to “encapsulate” the bytes. Using the bit-concatenation operator, we could do something like:

(poke) var bg = 255UB::255UB::255UB
(poke) var fg = 255UB::99UB::71UB
(poke) bg
(uint<24>) 0xffffff
(poke) fg
(uint<24>) 0xff6347

This encodes each color with a 24-bit unsigned integer. When looking at the hexadecimal values of bg and fg above, note that 0xff = 255, 0x63 = 99 and 0x47 = 71. Each byte seems to be in the right position in the 24-bit containing number. Now, poking a pixel at some given offset should be as easy as issuing just one map operation, right? Let’s see, using some arbitrary offset 10#B:

(poke) uint<24> @ 10#B = fg
(poke) dump :from 10#B :size 4#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
0000000a: 4763 ff00                                Gc..

If your current endianness is little (i.e. you are running on a x86 system or similar) you will get the dump above. The bytes are reversed, and consequently the resulting pixel has the wrong color. Our little trick didn’t work :(

So are we doomed to poke three bytes for each pixel we want to poke in our image? No, not really. The Poke language provides a construction oriented to alleviate cases like this, where several similar elements are to be “encapsulated” in a container. These constructions are called arrays.

Using array values, we can define the foreground and background colors like this:

(poke) var bga = [255UB, 255UB, 255UB]
(poke) var fga = [255UB, 99UB, 71UB]

All the elements on an array should be of the same kind, i.e. of the same type. Therefore, this is not allowed:

(poke) [1,"foo"]
<stdin>:1:1: error: array initializers should be of the same type
[1,"foo"];
^~~~~~~~~

Given an array value, it is possible to query for the number of values contained in it (called elements) by using the 'length value attribute. For example:

(poke) bga'length
3UL

Tells us that the array value stored in the variable bga has three elements.

How can we poke an array value? We know that the map operator accepts two operands: a type specifier and the value to map. The type specifier of an array of three bytes is denoted as byte[3]. Therefore, we can again try to poke a foreground pixel at offset 10#B, this time using fga:

(poke) byte[3] @ 10#B = fga
(poke) dump :from 10#B :size 4#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
0000000a: ff63 4700                                 .cG.

This time, the bytes were written in the right order. This is because array elements are always written using their “written” ordering, with no mind to endianness. We can also map a pixel from a given offset:

(poke) byte[3] @ 10#B
[255UB,99UB,71UB]

3.2.5 Poking Lines

At this point, we could encode the 40 pixels composing the image, by issuing the same number of pokes of byte[3] arrays. However, we can simplify the task even further.

Our pixels are arrays of bytes, denoted by the type specifier byte[3]. Similarly, we could conceive arrays of 32-bit signed integers, denoted by int[3], or arrays of bits, denoted by uint<1>[3]. But, is it possible to have arrays of other arrays? Yes, it is:

(poke) [[1,2],[3,4]]

The value above is an array of two arrays of two integers each. If we wanted to map such an array, what would be the type specifier we would need to use? It would be int[2][2], which should be read from right-to-left as “array of two arrays of two integers”. Let’s map one from an arbitrary offset in our IO space:

(poke) int[2][2] @ 100#B
[[0,0],[0,0]]

Consider again the sequence of pixels composing the image. Using the information we have in the SBM header, we can group the pixels in the sequence into “lines”. In our example image, each line contains 5 pixels. It would be natural to express each line as a sequence of pixels. The first line in our image would be:

(poke) var l0 = [bga,fga,fga,bga,bga]
(poke) l0
[[255UB,255UB,255UB],[255UB,99UB,71UB],…]

Let’s complete the image lines:

(poke) var l0 = [bga,fga,bga,fga,bga]
(poke) var l1 = [bga,fga,bga,fga,bga]
(poke) var l2 = [bga,fga,fga,bga,bga]
(poke) var l3 = [bga,fga,bga,bga,bga]
(poke) var l4 = l3
(poke) var l5 = l4

Note how we exploited the fact that the three last lines of our image are identical, to avoid to write the same array thrice. Array values can be assigned, and in general manipulated, like any other kind of value, such as integers or strings.

At this point, we could poke the pixels line-by-line. What would be the type specifier for a line? A line is an array of five arrays of 3 bytes each, so the type specifier would be byte[3][5]. Let’s do that:

(poke) byte[3][5] @ 5#B = l0
(poke) byte[3][5] @ 10#B = l1
(poke) byte[3][5] @ 15#B = l2
(poke) byte[3][5] @ 20#B = l3
(poke) byte[3][5] @ 25#B = l4
(poke) byte[3][5] @ 30#B = l5
(poke) byte[3][5] @ 35#B = l6

Not bad, we went from poking 105 bytes in the IO space to poking six lines. But we can still do better…

3.2.6 Poking Images

When we poked the lines at the end of the previous section, we had to increase the offset in every map operation. This is inconvenient.

In the same way that a sequence of bytes can be abstracted in a line, a sequence of lines can be abstracted in an image. It follows that we can look at the image data as an array of lines. But lines are themselves arrays of arrays… no matter, there is no limit on the number of arrays-of levels that you can nest.

So, let’s define our image as an array of the lines defined above:

(poke) var image_data = [l0,l1,l2,l3,l4,l5]
(poke) image_data
[[[255UB,255UB,255UB],[255UB,99UB,71UB],[255UB,99UB,71UB]…]…]

What would be the type specifier for an image? It would be an array of seven arrays of five arrays of three bytes each, in other words byte[3][5][7]. Let’s poke the pixels:

(poke) byte[3][5][6] @ 5#B = image_data

This is an example of how abstraction can simplify the handling of binary data: we switched from manipulating bytes to manipulate higher abstractions such as colors, lines and images. We achieved that by structuring the data in a way that reflects these abstractions. That’s the way of the Poker.

3.2.7 Saving the Image

Now that we have completed the SBM image in our buffer *image*, it is time to save it to disk. For that, we can use the save command we are already familiar with.

We know that the SBM image starts at offset 0#B, but what is the size of its entire binary representation? The header is easy: it spans for 5 bytes. The size of the sequence of pixels can be derived from the pixels per line byte, and the number of lines byte. We know that each pixel occupies 3 bytes, so calculating…

(poke) var ppl = byte @ 3#B
(poke) var lines = byte @ 4#B
(poke) save :from 0#B :size 5#B + ppl#B * lines#B :file "p.sbm"

Note how we expressed “ppl bytes” as ppl#B, and “lines bytes” as lines#B. This is the same than expressing “10 bytes” as 10#B. We will talk more about these united values later.

There is another way of getting the size of the stream of pixels. Recall that we have the entire set of pixels, structured as lines, stored in the variable image_data. Given an array, it is possible to query for its size using the 'size attribute:

(poke) .set obase 10
(poke) [1,2,3]'size
96UL#b

The above indicates that the size of the array of the three integers 1, 2 and 3 is 96 bits. Using that attribute, we can also obtain the size of the pixels in the image:

(poke) image_data'size
720UL#b

And we can use it in the save command:

(poke) save :from 0#B :size 5#B + image_data'size :file "p.sbm"

Using either strategy, at this point a file named p.sbm should have been written in the current working directory, containing our “P is for poke” image. Keep that file around, because we will be poking it further!

3.3 Modifying SBM Images

3.3.1 Reading a SBM File

Let’s open with poke the cute image we created in the last section, p.sbm:

$ poke p.sbm
[…]
(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 5342 4d05 07ff ffff ff63 47ff 6347 ffff  SBM......cG.cG..
00000010: ffff ffff ffff ffff 6347 ffff ffff 6347  ........cG....cG
00000020: ffff ffff ffff ff63 47ff ffff ff63 47ff  .......cG....cG.
00000030: ffff ffff ffff 6347 ff63 47ff ffff ffff  ......cG.cG.....
00000040: ffff ffff ff63 47ff ffff ffff ffff ffff  .....cG.........
00000050: ffff ffff 6347 ffff ffff ffff ffff ffff  ....cG..........
00000060: ffff ff63 47ff ffff ffff ffff ffff       ...cG.........

You can see the P in the ASCII column, right? If it wasn’t for the header, it would be pictured almost straight. This is because dump shows 16 bytes per row, and our image has lines that are 15 bytes long. This is a happy coincidence: you definitely shouldn’t expect to see ASCII art in the dump output of SBM files in general! :)

Now let’s read the image’s metadata from the header: pixels per line and how many lines are contained in the image:

(poke) var ppl = byte @ 3#B
(poke) ppl
5UB
(poke) var lines = byte @ 4#B
(poke) lines
7UB

All right, our image is 7x5. Knowing that each pixel occupies three bytes, and that each line contains ppl pixels, and that we have lines lines, we can map the entire image data using an array type specifier:

(poke) var image_data = byte[3][ppl][lines] @ 5#B
(poke) image_data
[[[255UB,255UB,255UB],[255UB,99UB,71UB], …]…]

3.3.2 Painting Pixels

The “P is for poke” slogan was so successful and widely appraised that the recutils⁷ chaps wanted to do a similar campaign “R is for recutils”. For that purpose, they asked us for a tomato-colored SBM image with an R in it.

Our creative department got at it, and after a lot of work they came with the following design:

  | 0 | 1 | 2 | 3 | 4 |
  +---+---+---+---+---+
0 |   | * | * |   |   |
1 |   | * |   | * |   |
2 |   | * |   | * |   |
3 |   | * | * |   |   |
4 |   | * | * |   |   |
5 |   | * |   | * |   |
6 |   | * |   | * |   |

Observe that this design really looks like our P (so much for a creative department). The bitmap has exactly the same dimensions, and difference are just three pixels, that pass from being background pixels to foreground pixels.

Therefore, it makes sense to read our p.sbm and use it as a base, completing the missing pixels. We saw in the last section how to read a SBM image. This time, however, we will copy the image first to a memory IO space to avoid overwriting p.sbm:

$ poke p.sbm
[…]
(poke) .mem scratch
The current IOS is now `*scratch*'.
(poke) .info ios
  Id	Type	Mode	Size		Name
* #1	MEMORY		0x00001000#B	*scratch*
  #0	FILE	rw	0x0000006e#B	./p.sbm
(poke) copy :from_ios 0 :from 0#B :to 0#B :size iosize (0)
(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 5342 4d05 07ff ffff ff63 47ff 6347 ffff  SBM......cG.cG..
00000010: ffff ffff ffff ffff 6347 ffff ffff 6347  ........cG....cG
00000020: ffff ffff ffff ff63 47ff ffff ff63 47ff  .......cG....cG.
00000030: ffff ffff ffff 6347 ff63 47ff ffff ffff  ......cG.cG.....
00000040: ffff ffff ff63 47ff ffff ffff ffff ffff  .....cG.........
00000050: ffff ffff 6347 ffff ffff ffff ffff ffff  ....cG..........
00000060: ffff ff63 47ff ffff ffff ffff ffff 0000  ...cG...........
00000070: 0000 0000 0000 0000 0000 0000 0000 0000  ................

Good. Now let’s map the contents of the image, both header information and the sequence of pixels:

(poke) var ppl = byte @ 3#B
(poke) var lines = byte @ 4#B
(poke) var image_data = byte[3][ppl][lines] @ 5#B

Let’s modify the image. Since the dimensions of the new image are exactly the same, the header remains the same. It is the pixel sequence that is different. We basically need to turn the pixels at coordinates (4,2), (5,3) and (6,3) from background pixels to foreground pixels.

Remember how we would change the value of some integer in the IO space? First, we would map it into a variable, change the value, and then poke it back to the IO space. Something like this:

(poke) var n = int @ offset
(poke) n = n + 1
(poke) int @ offset = n

This three-steps process is necessary because in the n = n + 1 above we are modifying the value of the variable n, not the integer actually stored at offset offset in the current IO space. Therefore we have to explicitly poke it back if we want the IO space to be updated as well.

Array values (and, as we shall see, other “composited” values) are different: when they are the result of the application of the map operator, the resulting values are mapped themselves.

When a Poke value is mapped, updating their elements have a side effect: the area corresponding to the updated element, in whatever IO space it is mapped on, is updated as well!

Why is this? The map operator, regardless of the kind of value it is mapping, always returns a copy of the value found stored in the IO space. We already saw how this worked with integers. However, in Poke values are copied around using a mechanism called “shared value”. This means that when a composite value like an array is copied, its elements are shared by both the original value and the new value.

We can depict this graphically for better understanding. A Poke array like:

(poke) var a = [1,2,3]

is stored in memory like this:

   +---+---+---+
 a | 1 | 2 | 3 |
   +---+---+---+

If we make a copy of the array in another variable b:

(poke) var b = a

we get

   +---+---+---+      +---+---+---+
 a | 1 | 2 | 3 |    b | 1 | 2 | 3 |
   +---+---+---+      +---+---+---+

Note how each of the integer elements has been copied to the new value. The resulting two arrays can then be modified independently:

(poke) a[1] = 5

resulting in:

   +---+---+---+      +---+---+---+
 a | 1 | 5 | 3 |    b | 1 | 2 | 3 |
   +---+---+---+      +---+---+---+

However, consider the following array whose elements are also arrays:

(poke) var a = [[1,2],[3,4],[5,6]]

This array is stored in memory like this:

   +---+---+---+
 a |   |   |   |
   +---+---+---+
     |   |   |   +---+---+
     |   |   +---| 5 | 6 |
     |   |       +---+---+
     |   |
     |   |       +---+---+
     |   +-------| 3 | 4 |
     |           +---+---+
     |
     |           +---+---+
     +-----------| 1 | 2 |
                 +---+---+

If now we make a copy of the same array into another variable b:

(poke) var b = a

The elements of the array are copied “by shared value”, i.e.

   +---+---+---+           +---+---+---+
 a |   |   |   |         b |   |   |   |
   +---+---+---+           +---+---+---+
     |   |   |   +---+---+   |   |   |
     |   |   +---| 5 | 6 |---+   |   |
     |   |       +---+---+       |   |
     |   |                       |   |
     |   |       +---+---+       |   |
     |   +-------| 3 | 4 |-------+   |
     |           +---+---+           |
     |                               |
     |           +---+---+           |
     +-----------| 1 | 2 |-----------+
                 +---+---+

The elements are indeed shared between the original array and the copy! If now we modify any of the shared values, this will be reflected in both containing values:

(poke) a[1][1] = 9
(poke) a
[[1,2],[3,9],[5,6]]
(poke) b
[[1,2],[3,9],[5,6]]

or graphically:

   +---+---+---+           +---+---+---+
 a |   |   |   |         b |   |   |   |
   +---+---+---+           +---+---+---+
     |   |   |   +---+---+   |   |   |
     |   |   +---| 5 | 6 |---+   |   |
     |   |       +---+---+       |   |
     |   |                       |   |
     |   |       +---+---+       |   |
     |   +-------| 3 | 9 |-------+   |
     |           +---+---+           |
     |                               |
     |           +---+---+           |
     +-----------| 1 | 2 |-----------+
                 +---+---+

Back to our image, it follows that if we wanted to change the color of some SBM pixel stored at offset offset, we would do this:

(poke) var a = byte[3] @ offset
(poke) a[1] = 10

There is no need to poke the array back explicitly: the side effect of assigning 10 to a[1] is that the byte at offset offset+1 is poked.

Generally speaking, mapped values can be handled in exactly the same way than non-mapped values. This is actually a very central concept in poke. However, it is possible to check whether a given value is mapped or not using the 'mapped attribute.

As we said, simple values such as integers and strings are never mapped, regardless of where they come from. Both ppl and lines are integers, therefore:

(poke) ppl'mapped
0
(poke) lines'mapped
0

However, image_data is an array that was the result of the application of a map operator, so:

(poke) image_data'mapped
1

When a value is mapped, you can ask for the offset where it is mapped, and the IO space where it is mapped, using the attributes 'offset and 'ios respectively. Therefore:

(poke) image_data'ios
1
(poke) image_data'offset
40UL#b

In other words, image_data is mapped in the IO space with id 1 (the *scratch* buffer) at offset 40 bits, or 5 bytes. We already knew that, because we mapped the image data ourselves, but in other situations these attributes are most useful. We shall see how later.

Well, at this point it should be clear how to paint pixels. First, let’s define our background and foreground pixels:

(poke) var bga = [255UB, 255UB, 255UB]
(poke) var fga = [255UB, 99UB, 71UB]

Then, we just update the pixels in the image data using the right coordinates:

(poke) image_data[4][2] = fga
(poke) image_data[5][3] = fga
(poke) image_data[6][3] = fga
(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 5342 4d05 07ff ffff ff63 47ff 6347 ffff  SBM......cG.cG..
00000010: ffff ffff ffff ffff 6347 ffff ffff 6347  ........cG....cG
00000020: ffff ffff ffff ff63 47ff ffff ff63 47ff  .......cG....cG.
00000030: ffff ffff ffff 6347 ff63 47ff ffff ffff  ......cG.cG.....
00000040: ffff ffff ff63 47ff 6347 ffff ffff ffff  .....cG.cG......
00000050: ffff ffff 6347 ffff ffff 6347 ffff ffff  ....cG....cG....
00000060: ffff ff63 47ff ffff ff63 47ff ffff 0000  ...cG....cG.....
00000070: 0000 0000 0000 0000 0000 0000 0000 0000  ................

3.3.3 Cropping the R

Looking at our new image, we realize that the first and the last column are all background pixels. We are aware that the recutils project is always short of resources, so we would like to modify the image to remove these columns, cropping it so it looks like this:

  | 0 | 1 | 2 |
  +---+---+---+
0 | * | * |   |
1 | * |   | * |
2 | * |   | * |
3 | * | * |   |
4 | * | * |   |
5 | * |   | * |
6 | * |   | * |

In order to perform this operation we need to rework the stream of pixels to reflect the desired result, and then to update the header metadata accordingly.

3.3.4 Shortening and Shifting Lines

Let’s think in term of lines. In the original image, each line has 5 pixels, that we can enumerate as:

     +----+----+----+----+----+
line | p0 | p1 | p2 | p3 | p4 |
     +----+----+----+----+----+

What we want is to crop out the first and the last column, so the resulting line would look like:

     +----+----+----+
line | p1 | p2 | p3 |
     +----+----+----+

Let’s get the first line from the original image_data:

(poke) var l0 = image_data[0]
(poke) l0
[[255UB,255UB,255UB],[255UB,99UB,71UB],…]

We could create the corresponding cropped line, by doing something like this:

(poke) var cl0 = [l0[1],l0[2],l0[3]]

And the same for the other lines. However, Poke provides a better way to easily obtain sub portions of arrays. It is called trimming. Given an array like the line l0, we can obtain the desired portion of it by issuing:

(poke) l0[1:4]
[[255UB,99UB,71UB],[255UB,99UB,71UB]]

Note how the limits of the semi-open interval specified in the trim reflect array indexes, and hence they are 0 based. Also, the left limit is closed and the right limit is open. The result of an array trimming is always another array, which may be empty:

(poke) l0[1:3]
[[255UB,99UB,71UB]]

Armed with this new operation, we can very easily mount the sequence of pixels for our cropped image:

(poke) var l0 = image_data[0]
(poke) var l1 = image_data[1]
(poke) var l2 = image_data[2]
(poke) var l3 = image_data[3]
(poke) var l4 = image_data[4]
(poke) var l5 = image_data[5]

And then update the lines in the mapped image data:

(poke) image_data[0] = l0[1:4]
(poke) image_data[0] = l1[1:4]
(poke) image_data[1] = l2[1:4]
(poke) image_data[2] = l3[1:4]
(poke) image_data[3] = l4[1:4]
(poke) image_data[4] = l5[1:4]

3.3.5 Updating the Header

The last step is to update the header to reflect the new dimensions of the image:

(poke) byte[] @ 0#B = ['S','B','M']
(poke) byte @ 3#B = 3
(poke) byte @ 4#B = 7

And we are done. Note how this time we wrote the magic bytes as an array, to save some typing and silly manual offset arithmetic. You may have noticed that the type specifier we used this time in the map is byte[] instead of byte[3]. This type specifier denotes an array of any number of bytes, which certainly includes arrays of three bytes, like in the example.

3.3.6 Saving the Result

And finally, let’s write out the new file as r.sbm:

(poke) save :from 0#B :size 5#B + image_data'size :file "r.sbm"

3.4 Defining Types

3.4.1 Naming your Own Abstractions

During the process of creating and manipulating SBM files we soon started talking about things like lines, pixel sequences, pixels, colors, and so on. What is more, very naturally we started thinking in terms of these entities: let’s drop this or that line, or let’s change the green component of this pixel.

Consider for example RGB colors. We know that each color is defined by three levels of light: red, green and blue. These components are also called color beams. Since each color beam has a range of 0 to 255, many formats, like SBM, use bytes to encode them.

Therefore, in the previous sections we used the type specifier byte when we needed to map RGB color beams, like in:

(poke) byte[3] @ 5#B

Recall that the mapping operation above means “map three bytes at the offset 5 bytes in the current IO space”. But what we really want is to map color beams, not bytes!

Poke provides a way to assign names to type specifiers:

(poke) type RGB_Color_Beam = byte

The definition above tells poke that a RGB color beam is composed of a byte, i.e. an unsigned 8-bit integer. Any type specifier can be used at the right side of the assignment sign, and also names of already defined types. From this point on, we can map in terms of color beams:

(poke) RGB_Color_Beam[3] @ 5#B

Meaning “map three RGB color beams at the offset 5 in the current IO space”.

Once a type is defined, the name can be used anywhere where a type specifier is expected.

By the way, we mentioned many times how byte is a synonym for uint<8>, int is a synonym for int<32> and so on. These synonyms are actually result of type definitions that are in the poke standard library. This library is loaded by poke at startup time.

3.4.2 Abstracting the Structure of Entities

Since we didn’t know better, during our work with SBM images we had to remember how these entities were constructed from more simple entities such as bits and bytes, every time we needed to map them, or to poke them. For example, if we wanted to map a pixel at some particular offset, we had to issue the following command:

(poke) var pixel = byte[3] @ 5#B

Now that we made poke aware of what a RGB color beam is, we can rewrite the above as:

(poke) var pixel = RGB_Color_Beam[3] @ 5#B

This is better, but still adoleces from a big problem: what if at some point the SBM pixels get expanded to also have a transparency index, stored in a fourth byte? If that ever happens (and will happen later in this book) then we would need to remember it before issuing commands like:

(poke) var image_data = RGB_Color_Beam[4][ppl][lines] @ 5#B

To avoid this problem, we define yet another type, this time describing the structure of a SBM pixel:

(poke) type SBM_Pixel = RGB_Color_Beam[3]

And then we can define image_data as a table of SBM pixels, instead of as a table of triplets of RGB color beams:

(poke) var image_data = SBM_Pixel[ppl][lines] @ 5#B

This later definition doesn’t need to be changed if we change the definition of what a SBM pixel is. This is called encapsulation and is a very useful abstraction in computing.

3.5 Pickles

3.5.1 poke Commands versus Poke Constructions

In Nomenclature we mentioned that poke, the program, implements a domain specific programming language called Poke, with a big p. In the examples so far we have already used the Poke language, somewhat extensively, while interacting with the program using the REPL.

For example, in:

(poke) 10 + 2
12

We are giving poke a Poke expression 10 + 2 to be evaluated. Once the expression is evaluated, the REPL prints the resulting value for us.

Similarly, when we define a variable or a type with var and type respectively, we are providing poke definitions to be evaluated. When we assign a value to a variable we are actually providing a Poke statement to be executed.

So the REPL accepts poke commands, some of which happen to be Poke expressions, definitions or statements. But we also have used dot-commands like .file or .info. These dot-commands are not part of the Poke programming language.

Every time we insert a line in the REPL and hit enter, poke recognizes the nature of the line, and then does the right thing. If the line is recognized as a Poke expression, for example, the Poke compiler is used to compile the statement into a routine, that is executed by the Poke Virtual Machine. The resulting value is then printed for the benefit of the user.

3.5.2 Poke Files

Poke programs are basically a collection of definitions and statements, which most often than not are stored in files, which avoids the need to type them again and again. By convention, we use the .pk file extension when naming files containing Poke programs.

Remember how we defined the foreground and background pixels for p.sbm in the REPL?

(poke) var bga = [255UB, 255UB, 255UB]
(poke) var fga = [255UB, 99UB, 71UB]

Where bga is a white pixel and fga is a tomato colored pixel. We could write these definitions in a file colors.pk like this:

var white = [255UB, 255UB, 255UB];
var tomato = [255UB, 99UB, 71UB];

Note that variable definitions in Poke are terminated by a semicolon (;) character, but we didn’t need to specify them when we issued the definitions in the REPL. This is because poke adds the trailing semicolon for us when it detects a Poke construction requiring it is introduced in the REPL.

Another difference is that Poke constructions can span for multiple lines, like in most programming languages. For example, we could have the following variable definition in a file:

var matrix = [[10, 20, 30],
              [40, 50, 60],
              [70, 80, 90]];

Once we have written our colors.pk file, how can we make poke aware of it? A possibility is to use the load construction:

(poke) load colors

Assuming a file named colors.pk exists in the current working directory, poke will load it and evaluate its contents. After this, we can use the colors:

(poke) tomato
[0xffUB,0x63UB,0x47UB]

Before, we would need to define these colors every time we would like to poke SBM files or, in fact, any RGB24 data. Now we just have to load the file colors.pk and use the variables defined there.

If you try to load a file whose name contains a dash character (-) you will get an error message:

(poke) load my-colors
<stdin>:1:8: error: syntax error, unexpected '-', expecting ';'
load my-colors;
       ^

This is because the argument to load is interpreted as a Poke identifier, and dash characters are not allowed in identifiers. To alleviate this problem, you can also specify the name of the file to load encoded in a string, line in:

(poke) load "my-colors.pk"

If you use this form of load, however, you are required to specify the complete name of the file, including the .pk file extension.

Since loading files is such a common operation, poke provides a dot-command .load that does auto-completion:

(poke) .load my-colors.pk

Which is equivalent to load "my-colors.pk".

Since load is part of the Poke language, it can also be used in Poke programs stored in files. We will explore this later.

3.5.3 Pickling Abstractions

In the last section we defined a couple of RGB colors white and tomato in a file called colors.pk. If we keep adding colors to the file, we may end with a nice collection of colors that are available to us at any time, by just loading the file.

Since there are many ways to understand the notion of “color”, and also many ways to implement these many notions, it would be better to be more precise and call our file rgb24.pk, since the notion of color we are using is of that RGB24. While are at it, let’s also rename the variables to reflect the fact they denote not just any kind of colors, but RGB24 colors:

var rgb24_white = [255UB, 255UB, 255UB];
var rgb24_tomato = [255UB, 99UB, 71UB];

At this point, we can also add the definitions of a couple of types to our rgb24.pk:

type RGB_Color_Beam = byte;
type RGB24_Color = RGB_Color_Beam[3];

var rgb24_white = [255UB, 255UB, 255UB];
var rgb24_tomato = [255UB, 99UB, 71UB];

Any time we want to manipulate RGB24 colors we can just load the file rgb24.pk and use these types and variables.

In poke parlance we call files like the above, that contain definitions of conceptually related entities, pickles. Pickles can be very simple, like the rgb24.pk sketched above, or fairly complicated like dwarf.pk.

It is common for pickles to load other pickles. For example, if we were to write a SBM pickle, we would load the RGB24 pickle from it:

load rgb24;

[… SBM definitions …]

This way, when we load sbm from the repl, the dependencies get loaded as well.

GNU poke includes several already written pickles for commonly used file formats, and other domains. The load construction knows where these pickles are installed, so in order to load the pickle to manipulate ELF files, for example, all you have to do is to:

(poke) load elf

3.5.4 Startup

When poke starts it loads the file .pokerc located in our home directory, if it exists. This initialization file contains poke commands, one per line.

If we wanted to get some Poke file loaded at startup, we could do it by adding a load command to our .pokerc. For example:

# My poke configuration - jemarch
[…]
.load ~/.poke.d/mydefs.pk

3.6 Poking Structs

3.6.1 Heterogeneous Related Data

Let’s recap the structure of the header of a Stupid BitMap:

                SBM header
+-------+-------+-------+-------+-------+
|  'S'  |  'B'  |  'M'  |  ppl  | lines |
+-------+-------+-------+-------+-------+
  byte0   byte1   byte2   byte3   byte4

The header is composed of five fields, which actually compose three different logical fields: a magic number, the number of pixels per line, and the number of lines.

We could of course abstract the header using an array of five bytes, like this:

type SBM_Header = byte[5];

However, this would not capture the properties of the fields themselves, which would need to be remembered by the user: which of these five bytes correspond to the magic number? Is the pixels per line number signed or unsigned? etc.

Poke provides a much better way to abstract collections of heterogeneous data: struct types. Using a struct type we can abstract the SBM header like this:

type SBM_Header =
 struct
 {
   byte[3] magic;
   uint<8> ppl;
   uint<8> lines;
 }

Note how the struct has three named fields: magic, ppl and lines. magic is an array of three bytes, while ppl and lines are both unsigned integers.

3.6.2 Mapping Structs

Once defined, struct types can be referred by name. For example, we can map the SBM header at the beginning of our file p.sbm like this:

(poke) SBM_Header @ 0#B
SBM_Header {
  magic=[0x53UB,0x42UB,0x4dUB],
  ppl=0x5UB,
  lines=0x7UB
}

The value resulting from the mapping is a struct value. The fields of struct values are accessed using the familiar dot-notation:

(poke) var header = SBM_Header @ 0#B
(poke) header.ppl * header.lines
35UB

The total number of pixels in the image is 35. Note how both header.ppl and header.lines are indeed unsigned byte values, and thus the result of the multiplication is also an unsigned byte. This could be problematic if the image contained more than 255 pixels, but this can be prevented by using a cast:

(poke) header.ppl as uint * header.lines
35U

Now the second operand header.lines is promoted to a 32-bit unsigned value before the multiplication is performed. Consequently, the result of the operation is also 32-bit wide (note the suffix of the result.)

3.6.3 Modifying Mapped Structs

Remember when we wanted to crop a SBM image by removing the first and last row? We updated the header in a byte by byte manner, like this:

(poke) byte @ 3#B = 3
(poke) byte @ 4#B = 7

Now that we have the header mapped in a variable, updating it is much more easy and convenient. The dot-notation is used to update the contents of a struct field, by placing it at the left hand side of an assignment:

(poke) header.ppl = 3
(poke) header.lines = 7

This updates the pixel per line and the number of lines, in the IO space:

(poke) dump :size 5#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 5342 4d03 07                             SBM..

3.7 How Structs are Built

First we need to define some structure to use as an example. Let’s say we are interesting in poking Packets, as defined by the Packet Specification 1.2 published by the Packet Foundation (none less).

In a nutshell, each Packet starts with a byte whose value is always 0xab, followed by a byte that defines the size of the payload. A stream of bytes conforming the payload follows, themselves followed by another stream of the same number of bytes with “control” values.

We could translate this description into the following Poke struct type definition:

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;
    byte[size] payload;
    byte[size] control;
  };

There are some details described by the fictitious Packet Specification 1.2 that are not covered in this simple definition, but we will be attending to that later in this manual.

So, given the definition of a struct type like Packet, there are only two ways to build a struct value in Poke.

One is to map it from some IO space. This is achieved using the map operator:

(poke) Packet @ 12#B
Packet {
  magic = 0xab,
  size = 2,
  payload = [0x12UB,0x30UB],
  control = [0x1UB,0x1UB]
}

The expression above maps a Packet starting at offset 12 bytes, in the current IO space. See the Poke manual for more details on using the map operator.

The second way to build a struct value is to _construct_ one, specifying the value to some, all or none of its fields. It looks like this:

(poke) Packet {size = 2, payload = [1UB,2UB]}
Packet {
  magic = 0xab,
  size = 2,
  payload = [0x1UB,0x2UB],
  control = [0x0UB,0x0UB]
}

In either case, building a struct involves to determine the value of all the fields of the struct, one by one. The order in which the struct fields are built is determined by the order of appearance of the fields in the type description.

In our example, the value of magic is determined first, then size, payload and finally control. This is the reason why we can refer to the values of previous fields when defining fields, such as in the size of the payload array above, but not the other way around: by the time payload is mapped or constructed, the value of size, has already been mapped or constructed.

What happens behind the curtains is that when poke finds the definition of a struct type, like Packet, it compiles two functions from it: a mapper function, and a constructor function. The mapper function gets as arguments the IO space and the offset from which to map the struct value, whereas the constructor function gets the template specifying the initial values for some, or all of the fields; reasonable default values (like zeroes) are used for fields for which no initial values have been specified.

These functions, mapper and constructor, are invoked to create fresh values when a map operator @ or a struct constructor is used in a Poke program, or at the poke prompt.

3.8 Variables in Structs

Fields are not the only entity that can appear in the definition of a struct type.

Suppose that after reading more carefully the Packet Specification 1.2 (that spans for several thousand of pages) we realize that the field size doesn’t really store the number of bytes of the payload and control arrays, like we thought initially. Or not exactly: the Packet Foundation says that if size has the special value 0xff, then the size is zero.

We could of course do something like this:

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    byte[size == 0xff ? 0 : size] payload;
    byte[size == 0xff ? 0 : size] control;
  };

However, we can avoid replicating code by using a variable instead:

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    var real_size = (size == 0xff ? 0 : size);

    byte[real_size] payload;
    byte[real_size] control;
  };

Note how the variable can be used after it gets defined. In the underlying process of mapping or constructing the struct, the variable is incorporated into the lexical environment. Once defined, it can be used in constraint expressions, array sizes, and the like. We will see more about this later.

Incidentally, it is of course possible to use global variables as well. For example:

var packet_special = 0xff;
type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    var real_size = (size == packet_special ? 0 : size);

    byte[real_size] payload;
    byte[real_size] control;
  };

In this case, the global packet_special gets captured in the lexical environment of the struct type (in reality in the lexical environment of the implicitly created mapper and constructor functions) in a way that if you later modify packet_special the new value will be used when mapping/constructing new values of type Packet. Which is really cool, but let’s not get distracted from the main topic... :)

3.9 Functions in Structs

Further reading of the Packet Specification 1.2 reveals that each Packet has an additional crc field. The content of this field is derived from both the payload bytes and the control bytes.

But this is no vulgar CRC we are talking about. On the contrary, it is a special function developed by the CRC Foundation in partnership with the Packet Foundation, called superCRC (patented, TM).

Fortunately, the CRC Foundation distributes a pickle supercrc.pk, that provides a calculate_crc function with the following spec:

fun calculate_crc = (byte[] data, byte[] control) int:

So let’s use the function like this in our type, after loading the supercrc pickle:

load supercrc;

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    var real_size = (size == 0xff ? 0 : size);

    byte[real_size] payload;
    byte[real_size] control;

    int crc = calculate_crc (payload, control);
  };

However, there is a caveat: it happens that the calculation of the CRC may involve arithmetic and division, so the CRC Foundation warns us that the calculate_crc function may raise E_div_by_zero. However, the Packet 1.2 Specification tells us that in these situations, the crc field of the packet should contain zero. If we used the type above, any exception raised by calculate_crc would be propagated by the mapper/constructor:

(poke) Packet @ 12#B
unhandled division by zero exception

A solution is to use a function that takes care of the extra needed logic, wrapping calculate_crc:

load supercrc;

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    var real_size = (size == 0xff ? 0 : size);

    byte[real_size] payload;
    byte[real_size] control;

    fun corrected_crc = int:
    {
      try return calculate_crc (payload, control);
      catch if E_div_by_zero { return 0; }
    }

    int crc = corrected_crc;
  };

Again, note how the function is accessible after its definition. Note as well how both fields and variables and other functions can be used in the function body. There is no difference to define variables and functions in struct types than to define them inside other functions or in the top-level environment: the same lexical rules apply.

3.10 Struct Methods

At this point you may be thinking something on the line of “hey, since variables and functions are also members of the struct, I should be able to access them the same way than fields, right?”.

So you will want to do:

(poke) var p = Packet  12#B
(poke) p.real_size
(poke) p.corrected_crc

But sorry, this won’t work.

To understand why, think about the struct building process we sketched above. The mapper and constructor functions are derived/compiled from the struct type. You can imagine them to have prototypes like:

Packet_mapper (IOspace, offset) -> Packet value
Packet_constructor (template)   -> Packet value

You can also picture the fields, variables and functions in the struct type specification as being defined inside the bodies of Packet_mapper and Packet_constructor, as their contents get mapped/constructed. For example, let’s see what the mapper does:

Packet_mapper:

  . Map a byte, put it in a local `magic'.
  . Map a byte, put it in a local `size'.
  . Calculate the real size, put it in a local `real_size'.
  . Map an array of real_size bytes, put it in a local `payload'.
  . Map an array of real_size bytes, put it in a local `control'.
  . Compile a function, put it in a local `corrected_crc'.
  . map a byte, call the function in the local `corrected_crc',
    complain if the values are not the same, otherwise put the
    mapped byte in a local `crc'.
  . Build a struct value with the values from the locals `magic',
    `size', `payload', `control' and `crc', and return it.

The pseudo-code for the constructor would be almost identical. Just replace "map a byte" with “construct a byte”.

So you see, both the values for the mapped fields and the values for the variables and functions defined inside the struct type end as locals of the mapping process, but only the values of the fields are actually put in the struct value that is returned in the last step.

This is where methods come in the picture. A method looks very similar to a function, but it is not quite the same thing. Let me show you an example:

load supercrc;

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;

    var real_size = (size == 0xff ? 0 : size);

    byte[real_size] payload;
    byte[real_size] control;

    fun corrected_crc = int:
    {
      try return calculate_crc (payload, control);
      catch if E_div_by_zero { return 0; }
    }

    int crc = corrected_crc;

    method c_crc = int:
    {
      return corrected_crc;
    }
  };

We have added a method c_crc to our Packet struct type, that just returns the corrected superCRC (patented, TM) of a packet. This can be invoked using dot-notation, once a Packet value is mapped/constructed:

(poke) var p = Packet  12#B
(poke) p.c_crc
0xdeadbeef

Now, the important bit here is that the method returns the corrected crc of a Packet. That’s it, it actually operates on a Packet value. This Packet value gets implicitly passed as an argument whenever a method is invoked.

We can visualize this with the following “pseudo Poke”:

method c_crc = (Packet SELF) int:
{
   return SELF.corrected_crc;
}

Fortunately, poke takes care to recognize when you are referring to fields of this implicit struct value, and does The Right Thing(TM) for you. This includes calling other methods:

method foo = void: { ... }
method bar = void:
{
 [...]
 foo;
}

The corresponding “pseudo-poke” being:

method bar = (Packet SELF) void:
{
 [...]
 SELF.foo ();
}

It is also possible to define methods that modify the contents of struct fields, no problem:

var packet_special = 0xff;

type Packet =
  struct
  {
    byte magic = 0xab;
    byte size;
    [...]

    method set_size = (byte s) void:
    {
      if (s == 0)
        size = packet_special;
      else
        size = s;
    }
  };

This is what is commonly known as a setter. Note, incidentally, how a method can also use regular variables. The Poke compiler knows when to generate a store in a normal variable such as packet_special, and when to generate a set to a SELF field.

Given the different nature of the variables, functions and methods, there are a couple of restrictions:

• Functions can’t set fields defined in the struct type.

  This will be rejected by the compiler:

  type Foo =
    struct
    {
       int field;
       fun wrong = void: { field = 10; }
    };
  

  Remember the construction/mapping process. When a function accesses a field of the struct type like in the example above, it is not doing one of these pseudo SELF.field = 10. Instead, it is simply updating the value of the local created in this step in Foo_mapper:

  Foo_mapper:
  
   . Map an int, put it in a local `field'.
   . [...]
  

  Setting that local would impact the mapping of the subsequent fields if they refer to field (for example, in their constraint expression) but it wouldn’t actually alter the value of the field field in the struct value that is created and returned from the mapper!

  This is very confusing, so we just disallow this with a compiler error “invalid assignment to struct field”, for your own sanity 8-)

• Methods can’t be used in field constraint expressions, nor in variables or functions defined in a struct type.

  How could they be? The field constraint expressions, the initialization expressions of variables, and the functions defined in struct types are all executed as part of the mapper/constructor and, at that time, there is no struct value yet to pass to the method.

  If you try to do this, the compiler will greet you with an “invalid reference to struct method” message.

3.11 Dealing with Alternatives

3.11.1 BSON

BSON ⁸ is a binary encoding for JSON documents. The top-level entity described by the spec is the document, which contains the total size of the document, a sequence of elements, and a trailing null byte.

We can describe the above in Poke with the following type definition:

type BSON_Doc =
  struct
  {
    offset<int<32>,B> size;
    BSON_Elem[size - size'size - 1#B] elements;
    byte endmark : endmark == 0x0;
  };

BSON elements come in different kinds, which correspond to the different types of JSON entities: 32-bit integers, 64-bit integers, strings, arrays, timestamps, and so on. Every element starts with a tag, which is a 8-bit unsigned integer that identifies it’s kind, and a name encoded as a NULL-terminated string. What comes next depends on the kind of element.

The following Poke type definition describes a subset of BSON elements, namely integers, big integers and strings:

type BSON_Elem =
  struct
  {
    byte tag;
    string name;

    union
    {
      int32 integer32 : tag == 0x10;
      int64 integer64 : tag == 0x12;
      BSON_String str : tag == 0x02;
    } value;
  };

The union in BSON_Elem corresponds to the variable part. When poke decodes an union, it tries to decode each alternative (union field) in turn. The first alternative that is successfully decoded without raising a constraint violation exception is the chosen one. If no alternative can be decoded, a constraint violation exception is raised.

To see this process in action, let’s use the BSON corresponding to the following little JSON document:

{
    "name" : "Jose E. Marchesi",
    "age" : 40,
    "big" : 1076543210012345
}

Let’s take a look to the different elements:

(poke) .load bson.pk
(poke) var d = BSON_Doc  0#B
(poke) d.elements'length
0x3UL
(poke) d.elements[0]
BSON_Elem {
  tag=0x2UB,
  name="name",
  value=struct {
    str=BSON_String {
      size=0x11,
      value="Jose E. Marchesi",
      chars=[0x4aUB,0x6fUB,0x73UB,0x65UB...]
    }
  }
}
(poke) d.elements[1]
BSON_Elem {
  tag=0x10UB,
  name="age",
  value=struct {
    integer32=0x27
  }
}
(poke) d.elements[2]
BSON_Elem {
  tag=0x12UB,
  name="big",
  value=struct {
    integer64=0x3d31c3f9e3eb9L
  }
}

Note how unions decode into structs featuring different fields. What field is available depends on the alternative chosen while decoding.

In the session above, d.elements[1].value contains an integer32 field, whereas d.elements[2].value contains an integer64 field. Let’s see what happens if we try to access the wrong field:

(poke) d.elements[1].value.integer64
unhandled invalid element exception

We get a run-time exception. This kind of errors cannot be catched at compile time, since both integer32 and integer64 are potentially valid fields in the union value.

3.11.2 Unions are Tagged

Unlike in C, Poke unions are tagged. Unions have their own lexical closures, and it is the capured values that determine what field is chosen at every time. Wherever the union goes, its tag accompanies it.

To see this more clearly, consider the following alternative Poke description of the BSON elements:

type BSON_Elem =
  union
  {
    struct
    {
      byte tag : tag == 0x10;
      string name;
      int32 value;
    } integer32;

    struct
    {
      byte tag : tag == 0x12;
      string name;
      int64 value;
    } integer64;

    struct
    {
      byte tag : tag == 0x12;
      string name;
      BSON_String value;
    } str;
  };

This description is way more verbose than the one used in previous sections, but it shows a few important properties of Poke unions.

First, the constraints guiding the decoding are not required to appear in the union itself: it is a recursive process. In this example, BSON_String could have constraints on it’s own, and these constraints will also impact the decoding of the union.

Second, there are generally many different ways to express the same plain binary using different type structures. This is no different than getting different parse trees from the same sequence of tokens using different grammars denoting the same language.

See how different a BSON element looks (and feels) using this alternative description:

(poke) d.elements[0]
BSON_Elem {
  str=struct {
    tag=0x2UB,
    name="name",
    value=BSON_String {
      size=0x11,
      value="Jose E. Marchesi",
      chars=[0x4aUB,0x6fUB,0x73UB,0x65UB...]
    }
  }
}
(poke) d.elements[1]
BSON_Elem {
  integer32=struct {
    tag=0x10UB,
    name="age",
    value=0x27
  }
}
(poke) d.elements[2]
BSON_Elem {
  integer64=struct {
    tag=0x12UB,
    name="big",
    value=0x3d31c3f9e3eb9L
  }
}

What is the best way? It certainly depends on the kind of data you want to manipulate, and the level of abstraction you want to achieve. Ultimately, it is up to you.

Generally speaking, the best structuring is the one that allows you to manipulate the data in terms of the structured abstractions as naturally as possible. That’s the art and craft of writing good pickles.

3.12 Structured Integers

When we structure data using Poke structs, arrays and the like, we often use the same structure than a C programmer would use. For example, to model ELF RELA structures, which are defined in C like:

typedef struct
{
  Elf64_Addr   r_offset;  /* Address */
  Elf64_Xword  r_info;    /* Relocation type and symbol index */
  Elf64_Sxword r_addend;  /* Addend */
} Elf64_Rela;

we could use something like this in Poke:

type Elf64_Rela =
  struct
  {
    Elf64_Addr r_offset;
    Elf64_Xword r_info;
    Elf64_Sxword r_addend;
  };

Here the Poke struct type is pretty equivalent to the C incarnation. In both cases the fields are always stored in the given order, regardless of endianness or any other consideration.

However, there are situations where stored integral values are to be interpreted as composite data. This is the case of the r_info field above, which is a 64-bit unsigned integer (Elf64_Xword) which is itself composed by several fields, depicted here:

 63                                          0
+----------------------+----------------------+
|       r_sym          |      r_type          |
+----------------------+----------------------+
MSB                                         LSB

In order to support this kind of composition of integers, C programmers usually resort to either bit masking (most often) or to the often obscure and undefined behaviour-prone C bit fields. In the case of ELF, the GNU implementations define a few macros to access these “sub-fields”:

#define ELF64_R_SYM(i)         ((i) >> 32)
#define ELF64_R_TYPE(i)        ((i) & 0xffffffff)
#define ELF64_R_INFO(sym,type) ((((Elf64_Xword) (sym)) << 32) + (type))

Where ELF64_R_SYM and ELF64_R_TYPE are used to extract the fields from an r_info, and ELF64_R_INFO is used to compose it. This is typical of C data structures.

We could of course mimic the C implementation in Poke:

fun Elf64_R_Sym = (Elf64_Xword i) uint<32>:
   { return i .>> 32; }
fun Elf64_R_Type = (Elf64_Xword i) uint<32>:
   { return i & 0xffff_ffff; }
fun Elf64_R_Info = (uint<32> sym, uint<32> type) Elf64_Xword:
   { return sym as Elf64_Xword <<. 32 + type; }

However, this approach has a huge disadvantage: since we are not able to encode the logic of these “sub-fields” in proper Poke fields, they become second class citizens, with all that implies: no constraints on their own, can’t be auto-completed, can’t be assigned individually, etc.

But we can use the so-called integral structs! These are structs that are defined exactly like your garden variety Poke structs, with a small addition:

type Elf64_RelInfo =
  struct uint<64>
  {
    uint<32> r_sym;
    uint<32> r_type;
  };

Note the uint<64> addition after struct. This can be any integer type (signed or unsigned). The fields of an integral struct should be integral themselves (this includes both integers and offsets) and the total size occupied by the fields should be the same size than the one declared in the struct’s integer type. This is checked and enforced by the compiler.

The Elf64 RELA in Poke can then be encoded like:

type Elf64_Rela =
  struct
  {
    Elf64_Addr r_offset;
    struct Elf64_Xword
    {
      uint<32> r_sym;
      uint<32> r_type;
    } r_info;
    Elf64_Sxword r_addend;
  };

When an integral struct is mapped from some IO space, the total number of bytes occupied by the struct is read as a single integer value, and then the values of the fields are extracted from it. A similar process is using when writing. That is what makes it different with respect a normal Poke struct.

It is possible to obtain the integral value corresponding to an integral struct using a cast to an integral type:

(poke) type Foo = struct int<32> { int<16> hi; uint<16> lo; };
(poke) Foo { hi = 1 } as int<32>
0x10000

An useful idiom, that doesn’t require to specify an explicit integral type, is this:

(poke) type Foo = struct int<32> { int<16> hi; uint<16> lo; };
(poke) var x = Foo @ 0#B;
(poke) +x
0xfe00aa

These casts allow to “integrate” struct values explicitly, but the compiler also implicitly promotes integral struct values to integers in all contexts where an integer is expected:

• Operator to an addition, subtraction, multiplication, division, ceil division, bit left shift, bit right shift, bit-wise not, bit-wise and, bit-wise or, bit-wise xor, logical and, logical or, logical not, positive, negative, bit-concatenation, condition in conditional expression.
• Argument to a function whose formal expects an integer.
• Value returned from a function that returns an integer.
• Condition in loop statement.
• Constraint expression in a struct field.
• Expression in a conditional struct field.

Note that the above list doesn’t include relational operators, since these operators work with struct values.

4 Maps and Map-files

Editing data with GNU poke mainly involves creating mapped values and storing them in Poke variables. However, this may not be that convenient when poking several files simultaneously, and when the complexity of the data increases. poke provides a convenient mechanism for this: maps and map files.

4.1 Editing using Variables

Editing data with GNU poke mainly involves creating mapped values and storing them in Poke variables. For example, if we were interested in altering the fields of the header in an ELF file, we would map an Elf64_Ehdr struct at the beginning of the underlying IO space (the file), like in:

(poke) .file foo.o
(poke) load elf
(poke) var ehdr = Elf64_Ehdr @ 0#B

At this point the variable ehdr holds an Elf64_Ehdr structure, which is mapped. As such, altering any of the fields of the struct will update the corresponding bytes in foo.o. For example:

(poke) ehdr.e_entry = 0#B

A Poke value has three mapping related attributes: whether it is mapped, the offset at which it is mapped in an IO space, and in which IO space. This information is accessible for both the user and Poke programs using the following attributes:

(poke) ehdr'mapped
1
(poke) ehdr'offset
0UL#b
(poke) ehdr'ios
0

Thats it, ehdr is mapped at offset zero byte in the IO space #0, which corresponds to foo.o:

(poke) .info ios
  Id   Type   Mode   Size           Name
* #0   FILE   rw     0x000004c8#B   ./foo.o

Now that we have the ELF header, we may use it to get access to the ELF section header table in the file, that we will reference using another variable shdr:

(poke) var shdr = Elf64_Shdr[ehdr.e_shnum] @ ehdr.e_shoff
(poke) shdr[1]
Elf64_Shdr {
  sh_name=0x1bU#B,
  sh_type=0x1U,
  sh_flags=#<ALLOC,EXECINSTR>,
  sh_addr=0x0UL#B,
  sh_offset=0x40UL#B,
  sh_size=0xbUL#B,
  sh_link=0x0U,
  sh_info=0x0U,
  sh_addralign=0x1UL,
  sh_entsize=0x0UL#b
}

Variables are convenient entities to manipulate in Poke. Let’s suppose that the file has a lot of sections and we want to do some transformation in every section. It is a time consuming operation, and we may forget which sections we have already processed and which not. We could create an empty array to hold the sections already processed:

(poke) var processed = Elf64_Shdr[] ()

And then, once we have processed some given section, add it to the array:

... edit shdr[23] ...
(poke) processed += [shdr[23]]

Note how the array =processed= is not mapped, but the sections contained in it are mapped: Poke uses copy by shared value. So, after we spend the day carefully poking our ELF file, we can ask poke, are we done with all the sections in the file?

(poke) shdr'length == processed'length
1

Yes, we are. This can be made as sophisticated as desired. We could easily write a function that saves the contents of processed in files, so we can continue hacking tomorrow, for example.

We can then concluding that using mapped variables to edit data structures stored in IO spaces works well in common and simple cases like the above: we make our ways mapping here and there, defining variables to hold data that interests us, and it is easy to remember that the variables ehdr and shdr are mapped, where are they mapped, and that they are mapped in the file foo.o.

However, GNU poke allows to edit more than one IO space simultaneously. Let’s say we now want to poke the sections of another ELF file: bar.o. We would start by opening the file:

(poke) .file bar.o
(poke) .info ios
  Id   Type   Mode   Size           Name
* #1   FILE   rw     0x000004c8#B   ./bar.o
  #0   FILE   rw     0x000004c8#B   ./foo.o

Now that bar.o is the current IO space, we can map its header. But now, what variable to use? We would rather not redefine ehdr, because that is already holding the header of foo.o. We could adapt our naming schema on the fly:

(poke) var foo_ehdr = ehdr
(poke) var bar_ehdr = Elf64_Ehdr @ 0#B

But then we would need to do the same for the other variables too:

(poke) var foo_shdr = shdr
(poke) var bar_shdr = Elf64_Shdr[bar_ehdr.e_shnum] @ bar_ehdr.e_shoff

However, we can easily see how this can degenerate quickly: what about processed, for example? In general, as the number of IO spaces being edited increases it becomes more and more difficult to manage our mapped variables, which are associated to each IO space.

4.2 poke Maps

As we have seen mapping variables is a very powerful, general and flexible mean to edit stored binary data in one or more IO spaces. However it is easy to lose track of where the variables are mapped and, ideally speaking, we would want to have a mean to refer to, say, the “ELF header”, and get the header as a mapped value regardless of what specific file we are editing. Sort of a “meta variable”. GNU poke provides a way to do this: maps.

A map can be conceived as a sort of “view” that can be applied to a given IO space. Maps have entries, which are values mapped at some given offset, under certain conditions. For example, we have seen an ELF file contains, among other things, a header at the beginning of the file and a table of section headers of certain size and located at certain location determined by the header. These would be two entries of a so-called ELF map.

poke maps are defined in map files. These files use the .map extension. A map file self.map (for sectioned/simple elf) defining the view of an ELF file as a header and a table of section header would look like this:

/* self.map - map file for a simplified view of an ELF file.  */

load elf;

%%

%entry
%name ehdr
%type Elf64_Ehdr
%offset 0#B

%entry
%name shdr
%type Elf64_Shdr[(Elf64_Ehdr @ 0#B).e_shnum]
%condition (Elf64_Ehdr @ 0#B).e_shnum > 0
%offset (Elf64_Ehdr @ 0#B).e_shoff

This map file defines a view of an ELF file as a header entry ehdr and an entry with a table of section headers shdr.

The first section of the file, which spans until the separator line containing %%, is arbitrary Poke code which as we shall see, gets evaluated before the map entries are processed. This is called the map "prologue". In this case, the prologue contains a comment explaining the purpose of the file, and a single statement load that loads the elf.pk pickle, since the entries below use definitions like Elf64_Ehdr that are defined by that pickle. The prologue is useful to define Poke functions and other entities that are then used in the definitions of the entries.

A separator line containing only %% separates the prologue from the next section, which is a list of entries definitions. Each entry definition starts with a line %entry, and has the following attributes:

%name

Like ehdr and shdr. These names should follow the same rules than Poke variables, but as we shall see later, map entries are not Poke variables. This attribute is mandatory.

%type

This can be any Poke expression denoting a type, like int, Elf64_Ehdr or Elf64_Shdr[(Elf64_Ehdr @ 0#B).e_shnum]. This attribute is mandatory.

%condition

If specified, will determine whether to include the entry in the map. In the example above, the map will have an entry shdr only if the ELF file has one or more sections. Any Poke expression evaluating to a boolean can be used as conditions. This attribute is optional: entries not having a condition will always be included in the map.

%offset

Offset in the IO space where the entry will be mapped. Any Poke expression evaluating to an offset can be used as entry offset. This attribute is mandatory.

4.3 Loading Maps

So we have written our =self.map=, which denotes a view or structure of ELF files we are interested on, and that resides in the current working directory. How to use it?

The first step is to fire up poke and open some object file. Let’s start with foo.o:

(poke) .file foo.o

Now, we can load the map using the .map load dot-command:

(poke) .map load self
[self](poke)

The .map load self command makes poke to look in certain directories for a file called self.map, and to load it. The list of directories where poke looks for map files is encoded in the variable map_load_path as a string containing a maybe empty list of directories separated by : characters. Each directory is tried in turn. This variable is initialized with suitable defaults:

(poke) map_load_path
"/home/jemarch/.poke.d:.:/home/jemarch/.local/share/poke:[...]"

Once a map is loaded, observe how the prompt changed to contain a prefix [self]. This means that the map self is loaded for the current IO space. You can choose to not see this information in the prompt by setting the prompt-maps option either at the prompt or in your .pokerc:

poke) .set prompt-maps no

By default prompt-maps is yes. This prompt aid is intended to provide a cursory look of the "views" or maps loaded for the current IO space. If we load another IO space and switch to it, the prompt changes accordingly:

(poke) [self](poke) .mem foo
The current IOS is now `*foo*'.
(poke) .ios #0
The current IOS is now `./foo.o'.
[self](poke)

At any time the .info maps dot-command can be used to obtain a full list of loaded maps, with more information about them:

(poke) .info maps
IOS   Name   Source
#0    self   ./self.map

In this case, there is a map self loaded in the IO space #0, which corresponds to foo.o.

Once we make foo.o our current IO space, we can ask poke to show us the entries corresponding to this map using another dot-command:

    : (poke) .map show self
    : Offset     Entry
    : 0x0UL#B    $self::ehdr
    : 0x208UL#B  $self::shdr

This tells us there are two entries for self in foo.o: $self::ehdr and $self::shdr= Note how map entries use names that start with the $ character, then contain the name of the map an the name of the entry we defined in the map file, separated by ::.

We can now use these entries at the prompt like if they were regular mapped variables:

[self](poke) $self::ehdr
Elf64_Ehdr {
  e_ident=struct {
    ei_mag=[0x7fUB,0x45UB,0x4cUB,0x46UB],
    [...]
  },
 e_type=0x1UH,
 e_machine=0x3eUH,
 [...]
}
(poke) $self::shdr'length
11UL

It is important to note, however, that map entries like $foo::bar are not part of the Poke language, and are only available when using poke interactively. Poke programs and scripts can’t use them.

Let’s now open another ELF file, and the =self= map in it:

(poke) .file /usr/local/lib/libpoke.so.0.0.0
(poke) .map load self
[self](poke)

So now we have two ELF files loaded in poke: foo.o and libpoke.so.0.0.0, and in both IO spaces we have the self map loaded. We can easily see that the map entries are different depending on the current IO space:

[self](poke) .map show self
Offset       Entry
0UL#B        $self::ehdr
3158952UL#B  $self::shdr
[self](poke) .ios #0
The current IOS is now `./foo.o'.
[self](poke) .map show self
Offset   Entry
0UL#B    $self::ehdr
520UL#B  $self::shdr

foo.o is an object file, whereas libpoke.so.0.0.0 is a DSO:

(poke) .ios #0
The current IOS is now `./foo.o'.
[self](poke) $self::ehdr.e_type
1UH
[self](poke) .ios #2
The current IOS is now `/usr/local/lib/libpoke.so.0.0.0'.
[self](poke) $self::ehdr.e_type
3UH

The interpretation of the map entry $self::ehdr is different depending on the current IO space. This makes it possible to refer to the “ELF header” of the current file.

Underneath, poke implements this by defining mapped variables and “redirecting” the entry names $foo::bar to the right variable depending on the IO space that is currently selected. It hides all that complexity from us.

4.4 Multiple Maps

It is perfectly possible (and useful!) to load more than one map in the same IO space. It is very natural for a single file, for example, to contain data that can be interpreted in several ways, or of different nature.

Let’s for example open again an ELF file, this time compiled with -g:

(poke) .file foo.o

We now load our self map, to get a view of the file as a collection of sections:

(poke) .map load self
[self](poke)

And now we load the dwarf map that comes with poke, to get a view of the file as having debugging information encoded in DWARF:

[self(poke) .map load dwarf
[dwarf,self](poke)

See how the prompt now reflects the fact that the current IO space contains DWARF info! Let’s take a look:

[dwarf,self](poke) .info maps
IOS   Name    Source
#0    dwarf   /home/jemarch/gnu/hacks/poke/maps/dwarf.map
#0    self    ./self.map
[dwarf,self](poke) .map show dwarf
Offset    Entry
0x5bUL#B  $dwarf::info

Now we can access entries from any of the loaded maps, i.e. access the file in terms of different perspectives. As an ELF file:

[dwarf,self](poke) $self::shdr[1]
Elf64_Shdr {
  sh_name=0xb5U#B,
  sh_type=0x11U,
  sh_flags=#<>,
  sh_addr=0x0UL#B,
  sh_offset=0x40UL#B,
  sh_size=0x8UL#B,
  sh_link=0x18U,
  sh_info=0xfU,
  sh_addralign=0x4UL,
  sh_entsize=0x4UL#b
}

And as a file containing DWARF info:

[dwarf,self](poke) $dwarf::info
Dwarf_CU_Header {
  unit_length=#<0x0000004eU#B>,
  version=0x4UH,
  debug_abbrev_offset=#<0x00000000U#B>,
  address_size=0x8UB#B
}

If you are curious about how the DWARF entries are defined, look at maps/dwarf.map in the poke source distribution, or in your installed poke (.info maps will tell you the file the map got loaded from.)

It is possible to unload or remove a map from a given IO space using the .map remove dot-command. Say we are done looking at the DWARF in foo.o, and we are no longer interested in it as a file containing debugging info. We can do:

[dwarf,self](poke) .map remove dwarf
[self](poke)

Note how the prompt was updated accordingly: only self remains as a loaded map on this file.

4.5 Auto-map

Certain maps make sense when editing certain types of data. For example, dwarf.map is intended to be used in ELF files. In order to ease using maps, poke provides a feature called auto mapping, which is disabled by default.

You can set auto mapping like this:

(poke) .set auto-map yes

When auto mapping is enabled, poke will look to the value of the pre-defined variable auto_map, which must contain an array of pairs of strings, associating a regular expression with a map name.

For example, you may want to initialize auto_map like this in your .pokerc file:

auto_map = [[".*\\.mp3$", "mp3"],
            [".*\\.o$", "elf"],
            ["a\\.out$", "elf"]];

This will make poke to load =mp3.map= for every file whose name ends with .mp3, and elf.map for files having names like foo.o and a.out.

Following the usual pokeish philosophy of being as less as intrusive by default as possible, the default value of auto_map is the empty string.

4.6 Constructing Maps

As we have seen, we can define our own maps using map files like self.map, which contain a prologue and a set of map entries. However, sometimes it is useful to create maps “on the fly” while we explore some data with poke.

To make this possible, poke provides a suitable set of dot-commands. Let’s say we are poking some data, and we want to create a map for it. We can do that like this:

(poke) .map create mymap

This creates an empty map named mymap, with no entries:

[mymap](poke) .map show mymap
Offset   Entry

Adding entries is easy. First, we have to map some variable, and then use it as the base for the new entry:

[mymap](poke) var foo = int[3]  0#B
[mymap](poke) .map entry add mymap, foo
[mymap](poke) .map show mymap
Offset   Entry
0x0UL#B  $mymap::foo

Note how the entry $mymap::foo gets created, associated to the current IO space and mapped at the same offset than the variable foo.

We can remove entries from existing maps using the .map entry remove dot-command:

[mymap](poke) .map entry remove mymap, foo
[mymap](poke) .map show mymap
Offset   Entry
[mymap](poke)

We plan to add an additional command to save maps to map files. The idea is that you can create your maps on the fly, save them, and then load them back some other day when you are ready to continue poking. This is not implemented yet though.

4.7 Predefined Maps

GNU poke comes with a set of useful pre-written maps, which get installed in a system location. We want to expand this collection, so please send us your map files!

5 Writing Pickles

GNU poke encourages the user to write little pieces of code in order to face spontaneous needs and fix situations. Is that name in the file encoded in a fixed array of characters padded with white spaces? No problem, just write a three lines function so you can update the file using a comfortable NULL-terminated string. Better than waiting for some poke maintainer to add that function for you, isn’t it?

Just save your functions in some personal .pk file that you load at startup, and your magic tricks bag will increase in time, making your poking more and more efficient.

However, when it comes to share the code with other people, it is important to follow certain conventions in order to achieve certain uniformity. This makes it easier for other people to discover what your hack provides, and how it works. It is this consistency, and these conventions, that makes some random .pk file a pickle.

This guide contains guidelines and recommendations for the pickle’s writer.

5.1 Pretty-printers

5.1.1 Convention for pretty-printed Output

Very often the structure of the data encoded in binary is not very intelligible. This is because it is usual for binary formats to be designed with goals in mind other than being readable by humans: compactness, detail etc.

In our pickle we of course want to provide access to the very finer detail of the data structures. However, we also want for the user to be able to peruse the data visually, and only look at the fine detail on demand.

Consider for example an ID3V1 tag data from some MP3 file. This is the result of mapping a ID3V1_Tag:

ID3V1_Tag {
  id=[0x54UB,0x41UB,0x47UB],
  title=[0x30UB,0x31UB,0x20UB,0x2dUB,0x20UB,...],
  artist=[0x4aUB,0x6fUB,0x61UB,0x71UB,0x75UB,...],
  album=[0x4dUB,0x65UB,0x6eUB,0x74UB,0x69UB,...],
  year=[0x20UB,0x20UB,0x20UB,0x20UB],
  data=struct {
    extended=struct {
      comment=[0x20UB,0x20UB,0x20UB,0x20UB,0x20UB,...],
      zero=0x0UB,
      track=0x1UB
    }
  },
  genre=0xffUB
}

Not very revealing. Fortunately, poke supports pretty printers. If a struct type has a method called _print, it will be used by poke as a pretty printer if the pretty-print option is set:

(poke) .set pretty-print yes

By convention, the output of pretty-printers should always start with #< and end with >. The convention makes it explicit for the user that everything she sees between #< and > is pretty-printed, and do not necessarily reflect the physical structure of the data. Also some information may be missing. In order to get an exact and complete description of the data, the user should .set pretty-print no and evaluate the value again at the prompt.

For example, in the following BPF instructions it is obvious at first sight that the shown register values are pretty-printed:

BPF_Insn = {
  ...
  regs=BPF_Regs {
     src=#<%r3>,
     dst=#<%r0>
  }
  ...
}

If the pretty-printed representation spans for more than one line, please place the opening #< in its own line, then the lines with the data, and finally > in its own line, starting at column 0.

Example of the MP3 tag above, this time pretty-printed:

#<
  genre: 255
  title: 01 - Eclipse De Mar
  artist: Joaquin Sabina
  album: Mentiras Piadosas
  year:
  comment:
  track: 1
>

5.1.2 Pretty Printing Optional Fields

Let’s say we are writing a pretty-printer method for a struct type that has an optional field. Like for example:

type Packet =
  struct
  {
    byte magic : magic in [MAGIC1,MAGIC2];
    byte n;
    byte[n] payload;
    PacketTrailer trailer if magic == MAGIC2;
  }

In this case, the struct value will have a trailer conditionally, which has to be tackled on the pretty-printer somehow.

An approach that often works good is to replicate the logic in the optional field condition expression, like this:

  struct
  {
    byte magic : magic in [MAGIC1,MAGIC2];
    [...]
    PacketTrailer trailer if packet_magic2_p (magic);

    method _print = void:
    {
      [...]
      if (magic == MAGIC2)
        pretty_print_trailer;
    }
  }

This works well in this simple example. In case the expression is big and complicated, we can avoid rewriting the same expression by encapsulating the logic in a function:

fun packet_magic2_p = int: { return magic == MAGIC2; }
type Packet =
  struct
  {
    byte magic : magic in [MAGIC1,MAGIC2];
    [...]
    PacketTrailer trailer if packet_magic2_p (magic);

    method _print = void:
    {
      [...]
      if (packet_magic2_p (magic))
        pretty_print_trailer;
    }
  }

However, this may feel weird, as the internal logic of the type somehow leaks its natural boundary into the external function packet_magic2_p.

An alternative is to use the following idiom, that checks whether the field actually exists in the struct:

type Packet =
  struct
  {
    byte magic : magic in [MAGIC1,MAGIC2];
    [...]
    PacketTrailer trailer if packet_magic2_p (magic);

    method _print = void:
    {
      [...]
      try pretty_print_trailer;
      catch if E_elem {}
    }
  }

This approach also works with unions:

type ID3V1_Tag =
  struct
  {
    [...]
    union
    {
      /* ID3v1.1  */
      struct
      {
        char[28] comment;
        byte zero = 0;
        byte track : track != 0;
      } extended;
      /* ID3v1  */
      char[30] comment;
    } data;
    [...]

    method _print = void:
    {
      [...]
      try print "  comment: " + catos (data.comment) + "\n";
      catch if E_elem
      {
        print "  comment: " + catos (data.extended.comment) + "\n";
        printf "  track: %u8d", data.extended.track;
      }
    }
  }

5.2 Setters and Getters

Given a struct value, the obvious way to access the value of a field is to just refer to it using dot-notation.

For example, for the following struct type:

type ID3V1_Tag =
  struct
  {
    [...]
    char[30] title;
    char[30] artist;
    char[30] album;
    char[4] year;
    [...]
  }

Suppose the find out the year in the tag is wrong, off by two years: the song was release in 1980, not in 1978!. Unfortunately, due to the bizarre way the year is stored in the file (as a sequence of digits encoded in ASCII, non-NULL terminated) we cannot just write:

(poke) tag.year = tag.year + 2
error

Instead, we can use facilities from the standard library and a bit of programming:

(poke) stoca (format ("%d", atoi (catos (tag.year)) + 2), tag.year)

The above line basically transforms the data we want to operate on (the tag year) from the stored representation into a more useful representation (from an array of character digits to an integer) then operates with it (adds two) then converts back to the stored representation. Let’s call this “more useful” representation the “preferred representation”.

A well written pickle should provide getter and setter methods for fields in struct types for which the stored representation is not the preferred representation. By convention, getter and setter methods have the following form:

method get_field preferred_type: { ... }
method set_field (preferred_type val) void: { ... }

Using the get_ and set_ prefixes consistently is very important, because the pokist using your pickle can easily find out the available methods for some given value using tab-completion in the REPL.

For example, let’s add setter and getter methods for the field year in the ID3V1 tag struct above:

type ID3V1_Tag =
  struct
  {
    [...]
    char[4] year;

    method get_year = int: { return atoi (catos (year)); }
    method set_year = (int val) void:
    {
      var str = format "%d", val;
      stoca (str, year);
    }
    [...]
  }

What constitutes the preferred representation of a field is up to the criteria of the pickle writer. For the tag above, I would say the preferred representations for the title, artist, album and year are string for title, artist and album, and an integer for the year.

6 Configuration

6.1 .pokerc

Upon invocation poke will read and execute the commands of an initialization file, if it exists: the pokerc file. There are two ways to keep an initialization file.

The simple, traditional way is to have a file named .pokerc in your home directory. GNU poke will look for a file like that first.

If .pokerc is not found your home directory, poke looks in the locations specified by the XDG Base Directory Specification⁹ for a file named poke/pokerc.conf. For example, you could use ~/.config/poke/pokerc.conf.

Which way is better depends on your specific requirements and taste.

GNU poke can be insturcted to not read the initialization file by passing -q or --no-init-file in the command line.

Example of initialization file:

# My poke configuration.
.set endian host
.set obase 16
.set pretty-print yes
pk_dump_cluster_by = 4
.load ~/.poke.d/mypickles.pk

6.2 Load Path

The load_path Poke variable contains a list of directories separated by the colon character (:). When a module is loaded using the load construction, these directories are searched in sequential order for the file corresponding to the requested module.

Empty directory names and entries that do not name existing directories are ignored.

Some entries have special meanings:

%DATADIR%

This is interpreted as the system-wide datadir directory, that depends on the prefix where poke is installed.

For example, say you want to maintain .pk files in your ~/.poke.d directory. You will probably want to add that directory to the load_path, when poke initializes. A way to do that is to add a command like this to your pokerc file:

load_path = getenv ("HOME") + "/.poke.d:" + load_path

If the environment variable POKE_LOAD_PATH is defined in the environment, its value is added to load_path at poke startup time.

6.3 Styling

XXX

7 Time

Systems encode time-stamps or dates in several ways. GNU poke provides support for some of them.

7.1 POSIX Time

In POSIX systems time is encoded as a certain number of seconds elapsed since some origin date, namely the first of January 1970. The time pickle provides two types for 32-bit and 64-bit POSIX time encodings:

POSIX_Time32
POSIX_Time64

A function ptime is also provided, that prints a human-readable representation of a POSIX date, given a number of seconds.

8 Colors

Colors are often found in binary data, encoded in several different ways. GNU poke provides several pickles that makes it easier to work with these colors.

8.1 The Color Registry

The color pickle provides a registry of standard colors, organized as a space of integers. Each integer identifies a standard color, which are accessible as variables named color_* after the pickle is loaded. Example:

(poke) load color
(poke) color_tomato
114

The purpose of having this register is to have a global namespace for colors that can be used in different pickles. The position of each color in the registry is totally unrelated to how the color may be encoded. Other pickles, as we shall see below, contain tables associating standard colors with their encoding, such as RGB.

If you want to add a new color that is not in the standard collection, you can use the color_register function, which gets no arguments:

(poke) var mycolor = color_register
(poke) mycolor
490

The total number of registered colors is recorded in the variable color_num_colors. You can use it to iterate on all the colors in the register, from 0 to color_num_colors - 1.

The index in the registry of the first user-defined color is hold in the variable color_LAST. For example, if you wanted to store a frob per standard color, you would do it like:

(poke) type Frobs = Frob[color_LAST]

The pickle also provides a function color_name that, given a color code, returns a printable name for the color, i.e. a string describing it. For user defined colors, this string is fixed:

(poke) color_name (23)
"lavender blush"
(poke) color_name (color_register)
"user-defined color"

If a non-existent color code is passed to color_name the function raises E_out_of_bounds:

(poke) color_name (color_num_colors)
unhandled out of bounds exception

8.2 RGB24 Encoding

The RGB24 encoding encodes each color as a triplet of color beans, each beam indicating a level of red, green and blue respectively.

Types are provided for both the color beams, and the triplets:

type RGB24_Color_Beam = uint<8>;
type RGB24_Color = RGB24_Color_Beam[3];

The indexes RGB24_RED, RGB24_GREEN and RGB24_BLUE can be used to access to a specific beam of a given color:

(poke) rgb24_color[color_tomato][RGB24_GREEN]
99UB

The rgb24 pickle also provides a table associating poke standard colors (see The Color Registry) with their RGB24 encodings. Once the pickle is loaded, this table is available in the variable rgb24_color. The table is to be indexed by color codes:

(poke) load rgb24
(poke) rgb24_color[color_tomato]
[255UB,99UB,71UB]

9 Audio

9.1 MP3

9.1.1 ID3V1 Tags

The id3v1 pickle provides abstractions in order to edit the metadata stored in a MP3 file.

9.1.1.1 Song Genres

The ID3V1 tags support the notion of song genre. The space for genres is from 0 to 254. The genre code 255 is reserved to mean “no genre”.

The table id3v1_genres can be indexed with a code in order to get the corresponding genre name:

(poke) id3v1_genres[14]
"rhythm and blues"

Conversely, the function id3v1_search_genre gives us the code of a genre, given its name. The prototype of this function is:

fun id3v1_search_genre = (string name) uint<8>:

For example:

(poke) id3v1_search_genre ("rock")
17UB

If id3v1_search_genre is given a name that doesn’t correspond with any genre in the genres table, then E_inval is raised.

9.1.1.2 The ID3V1_Tag Type

The main data structure defined in the id3v1 pickle is ID3V1_Tag, which corresponds to a ID3V1 tag (surprise!).

Tags comprise the following information:

• - The title of the song, which is limited to 30 bytes.
• - The artist, which is limited to 30 bytes.
• - The album, which is limited to 30 bytes.
• - The year, which is encoded as text in 4 bytes, each byte containing the ASCII code for the corresponding digit.
• - A comment, which is limited to either 28 or 30 bytes, depending whether the tag contains track information or not.
• - An optional track, which is an unsigned 8-bit number.

The specification does not mention any specific encoding for the entries that store text (such as title or artist), but it is safe to assume some ASCII-compatible encoding is used.

The text entries are stored as arrays of characters, and they are not finished by NULL characters. Instead the arrays of characters are filled with whitespaces (ASCII code 0x20) at the right. For example, the artist name Picasso encoded in ID3V1 would be:

['P','i','c','a','s','s','o',' ',' ', ..., ' ']

In order to ease the manipulation of the text fields, setters and getters are provided in order to handle these values as strings and not as whitespace-filled arrays of characters. Example:

(poke) tag.title
[48UB,49UB,32UB,45UB,32UB,...]
(poke) tag.get_title
"01 - Eclipse De Mar"

Also for setters:

(poke) tag.set_title ("Join us Now")
(poke) tag.title
[74UB,111UB,105UB,110UB,32UB,...]

Setters and getters are also provided in order to manipulate the year as an integer value:

(poke) tag.year
[0x31UB,0x39UB,0x38UB,0x30UB]
(poke) tag.get_year
1980
(poke) tag.set_year (1988)

10 Object Formats

This chapter is of special interest to toolchain developers, people doing reverse-engineering, and similar low-level development activities.

10.1 ELF

XXX

10.2 Dwarf

XXX

11 Programs

This chapter describes pickles whose purpose is to facilitate the writing of binary utilities in Poke.

11.1 argp

Like any other program, Poke scripts often need to handle options passed in the command line. The argp pickle provides a convenient and easy way to handle these arguments.

The main entry point of the pickle is the function argp_parse, with the following prototype:

fun argp_parse = (string program,
                  string version = "",
                  string summary = "",
                  Argp_Option[] opts = Argp_Option[](),
                  string[] argv = string[](),
                  int allow_unknown = 0) string[]

Where program is the name of the program providing the command-line options. version is either a string identifying the version of the program, or the empty string. summary is a short summary of what the program does. opts is an array of Argp_Option structs, each of which describe a command line option and, in particular, a handler for that option. argv is an array of string containing the command line elements to process. Finally, allow_unknown specifies whether unknown options constitute an error or not. This last option is useful in order to support sub-parsers.

Once invoked, argp_parse analyzes the command line provided in argv, recognizes options, performs checks making sure that all specified options are well formed, and that every option requiring an argument gets one, invokes the handlers registered for each option, and finally returns an array of non-option arguments in argv, for further processing by the user.

Each Argp_Option struct describes an option. They have this form:

type Argp_Option =
  struct
  {
    string name;
    string long_name;
    string summary;
    string arg_name;
    int arg_required;
    Argp_Option_Handler handler;
  };

Where name is a string of size one character specifying the short name for the option. long_name is a non-empty string of any size specifying the long name for the option. summary is a short paragraph with a summary of that the option does. arg_name, if specified, is a string to be used to describe the argument of the option in the --help output. If arg_name is not specified then "ARG" is used. arg_required is a boolean specifying whether the option accepts and expects an argument. Defaults to 0. Finally, handler is a function that will be invoked to process the option. The function has the following prototype:

type Argp_Option_Handler = (string)void

The argp_parse function provides default handlers for several standard options:

--help

The default handler for this option prints out an usage message in the standard GNU way, and exits. In particular, this output is compatible with the help2man utility.

--version

The default handler for this option prints out the name and the version of the program, and exits.

If the user provides handlers for the options above, these take precedence wrt the default handlers.

Several short options can be accumulated this way:

foo -qyz

The special option --, if found in the command line, stops the processing of options. Everything found about it is considered non-option arguments.

12 Dot-Commands

12.1 .load

The .load command loads a file containing Poke code and compiles and executes it. These files usually have the extension .pk.

If a relative path is provided, then prefix/share/poke is tried first as a base directory to find the specified file. If it is not found, then the current directory is tried next.

If the environment variable POKEDATADIR is defined, it replaces prefix/share/poke. This is mainly intended to test a poke program before it gets installed in its final location.

If an absolute path is provided, it is used as-is.

12.2 .source

The .source command executes command lines from some given file. The syntax is:

.source path

where path is a path to the file containing the commands. See Command Files.

12.3 .file

The .file command opens a new IO space backed by a file, or switches to a previously opened file. The syntax is:

.file path

where path is a path to a file to open, which can be relative to poke’s current working directory or absolute.

Tilde expansion is performed in path, much like it’s done in the shell. This means you can include special characters like ~ (which will expand to your home directory) delimit the file name with " in case it includes leading or trailing blank characters, etc.

When a new file is opened it becomes the current IO space. From that point on, every map executed in the REPL or while loading a Poke program will operate on that IO space:

(poke) .file foo.o
The current file is now `foo.o'.

12.4 .mem

The .mem command opens a new IO space backed by a memory buffer. The syntax is:

.mem name

where name is the name of the buffer to create. Note that poke adds prefix and trailing asterisk characters, to differentiate file names from buffer names.

When a new memory buffer IOS is opened it becomes the current IO space. See file command.

12.5 .nbd

The .nbd command opens a new IO space backed by an external NBD server. The syntax is:

.nbd uri

where uri is the name of the newly created buffer, matching the NBD URI specification.

When a new NBD IOS is opened, it becomes the current IO space. See file command.

NBD support in GNU poke is optional, depending on whether poke was compiled against libnbd.

For an example of connecting to the guest-visible content of a qcow2 image, with the default export name as exposed by using qemu as an NBD server:

$ qemu-nbd --socket=/tmp/mysock -f qcow2 image.qcow2
$ poke
(poke) .nbd nbd+unix:///socket=?/tmp/mysock
The current file is now `nbd+unix:///socket=?/tmp/mysock'.

12.6 .ios

A list of open files, and their corresponding tags, can be obtained using the .info ios command. Once a tag is known, you can use the .ios command to switch back to that file:

(poke) .ios #1
The current IOS is now `foo.o'.

12.7 .close

The .close command closes the selected IO space. The syntax is:

.close #tag

where #tag is a tag identifying an open IO stream.

12.8 .doc

The .doc command is used to display this manual in poke’s REPL. The syntax is:

.doc [node]

where node is an optional parameter which indicates the chapter or section at which the manual should be opened.

This command uses whatever documentation viewer configured using .set doc-viewer whose valid options are either info and less.

Unless doc-viewer is set to less, this command uses the info program (The GNU Texinfo Manual) to interactively present the manual. If info is not installed, then less is tried next.

When using less to display the documentation, the entry in the Table of Contents corresponding to the requested entry is highlighted. Just press / and then RET to jump to the corresponding section in the manual.

If neither info nor less are installed, the .doc command will fail. If poke is not running interactively then .doc does nothing.

12.9 .editor

The .editor command (usually abbreviated as .edit) invokes an external text editor on a temporary file. You can then put contents on that file, save it and exit the editor. At that point poke will read the file contents, turn them into a single line and execute them in the repl. If poke is not running interactively, then .editor does nothing.

The editor used is identified by the EDITOR environment variable.

12.10 .info

The .info command provides information about several kinds of entities. The recognized sub commands are:

.info ios

Display a list of open files.

(poke) .info ios
Id	Type	Mode	Size		Name
  #1	FILE	r	0x0000df78#B	foo.o
* #0	FILE	rw	0x00000022#B	foo.bson

The file acting as the current IO space is marked with an asterisk character * at the beginning of the file. The mode in which the file is open is also specified. The Id field is the tag of the file that can be passed to the .file command in order to switch to it as the new current IO space:

(poke) .ios #1
The current file is now `foo.o'.
(poke) .info ios
  Id	Type	Mode	Size		Name
* #1	FILE	r	0x0000df78#B	foo.o
  #0	FILE	rw	0x00000022#B	foo.bson

.info variables

Shows a list of defined variables along with their current values and the location where the variables were defined.

.info functions

Shows a list of defined functions along with their prototypes and the location where the functions were defined.

.info types

Shows a list of defined types along with the locations where the types were defined.

12.11 .set

The .set command allows you to inspect and set the value of global settings. The syntax is:

.set setting [value]

where setting is an identifier identifying the setting to inspect or modify. If value is specified, then it is the new value for the setting. If value is not specified the current value of the setting is displayed.

The following settings can be handled with .set:

endian

Byte endianness that will be used when mapping the IO space. Valid values are big, little and host. The default endianness is big endian.

obase

Numeric base to be used when displaying values in the REPL and in printf statements using the %v format tag. Valid values are 2, 8, 10 and 16. Default value is 10.

pretty-print

Flag indicating whether pretty-printers shall be used when printing values in the REPL and in printf statements using the %v format tag. Valid values are yes and no. Default value is no.

error-on-warning

Flag indicating whether handling compilation warnings as errors. Default value is no.

omode

It defines the way the binary struct data is displayed. In flat mode data is not formatted in any special way. In tree mode the struct data is displayed in a hierarchical (tree) mode.

odepth

In tree and flat mode the struct fields are recursively displayed up to the depth-th level. The default value 0 means no limit.

oindent

Number defining the number of spaces used for indentation for each level. Only values >=1 and <= 10 are valid. Default value is ’2’.

oacutoff

When displaying an array as struct field, display only the elements up to the cutoff index and display … after that. Value of 0 means no limit. This cutoff value is not used when directly displaying arrays content.

omaps

Flag indicating whether including mapping information when printing out mapped values.

12.12 .vm

The Poke Virtual Machine (PVM) executes the programs that are the result of the compilation of what you write in the REPL or the pickles you load. The .vm command provides sub-commands to interact with the PVM.

12.12.1 .vm disassemble

The .vm disassemble command provides access to the PVM disassembler. It supports the following subcommands:

.vm disassemble expression expr

Dumps the assembler corresponding to the Poke expression expr.

.vm disassemble function function

Dumps the assembler corresponding to the Poke function called function. The function shall be reachable from the top-level.

.vm disassemble mapper expr

If expr is a mapped value, dumps the assembler corresponding to its mapper function.

.vm disassemble writer expr

If expr is a mapped value, dumps the assembler corresponding to its writer function.

The disassembler will provide a PVM disassembly by default, but it can be passed the flag /n to do a native disassembly instead in whatever architecture running poke.

12.12.2 .vm profile

The .vm profile command provides access to the PVM profiler. This profiler is only available if poke has been configured with PVM profiling support. This command supports the following subcommands:

.vm profile reset

Resets the profiling counts in the virtual machine.

.vm profile show

Outputs a summary with both counts and sample information.

12.13 .exit

The .exit command exits poke. The syntax is:

.exit [status]

Poke will terminate, returning the exit status status. If status is omitted, then the exit status zero will be returned.

12.14 .quit

The .quit command behaves exactly like .exit. See exit command.

13 Commands

13.1 dump

At the most basic level, memory can be examined byte by byte. To do this, use the dump command.

This command has the following synopsis.

dump [:from offset] [:size offset] [:ruler bool] [:ascii bool]
     [:group_by int] [:cluster_by int]

All arguments are optional, which means the simplest use of the command is to simply type dump:

(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000000: 9b07 5a61 4783 f306 4897 f37c fe39 4cd3  ..ZaG...H..|.9L.
00000010: b6a2 a578 8d82 7b7f 2076 374c 3eab 7150  ...x..{. v7L>.qP
00000020: 31df 8ecb 3d33 ee12 429b 2e13 670d 948e  1...=3..B...g...
00000030: 86f1 2228 ae07 d95c 9884 cf0a d1a8 072e  .."(...\........
00000040: f93c 5368 9617 6c96 3d61 7b92 9038 a93b  .<Sh..l.=a{..8.;
00000050: 3b0d f8c9 efbd a959 88d0 e523 fd3b b029  ;......Y...#.;.)
00000060: e2eb 51d5 cb5b 5ba9 b890 9d7a 2746 72ad  ..Q..[[....z'Fr.
00000070: 6cbd 6e27 1c7f a554 8d2e 77f9 315a 4415  l.n'...T..w.1ZD.

The first row is the ruler which serves as a heading for each subsequent row. On the left hand side is the offset of the io space under examination. The centre block displays the hexadecimal representation of each byte, and on the right hand side is their ascii representation. If a byte is not representable in ascii, then the byte will be displayed as a dot.

By default the dump command reads from the currently selected IO space. However, it is possible to specify an explicit IO space using the ios option:

(poke) var myfile = open ("/path/to/file")
(poke) dump :ios myfile

13.1.1 Information dump shows

By default dump displays 128 bytes of memory starting at offset 0#B. You can change the quantity and starting offset by using the size and from arguments. For example:

(poke) dump :from 0x10#B :size 0x40#B
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff  0123456789ABCDEF
00000010: b6a2 a578 8d82 7b7f 2076 374c 3eab 7150  ...x..{. v7L>.qP
00000020: 31df 8ecb 3d33 ee12 429b 2e13 670d 948e  1...=3..B...g...
00000030: 86f1 2228 ae07 d95c 9884 cf0a d1a8 072e  .."(...\........
00000040: f93c 5368 9617 6c96 3d61 7b92 9038 a93b  .<Sh..l.=a{..8.;

Note that both the size and from arguments are offsets. As such, both must be specified using # and an appropriate unit. (see Offset Literals).

The other arguments change the appearance of the dump. If the ruler argument is zero, then the ruler will be omitted:

(poke) dump :ruler 0 :size 0x40#B
00000000: b1fd 1608 2346 759c 46a6 aa94 6fcd 846a  ....#Fu.F...o..j
00000010: e39f 473f 3247 415f 174d a32b ed89 a435  ..G?2GA_.M.+...5
00000020: d2c6 2c52 bc82 e0a7 e767 31ea 84de 41e5  ..,R.....g1...A.
00000030: 2add 2869 e9c2 226b e222 8c74 4b94 af24  *.(i.."k.".tK..$

To omit the ascii representation of the memory, call dump with the ascii argument set to zero:

(poke) dump :ruler 0 :size 0x40#B :ascii 0
00000000: 4393 85e7 0b0c 3921 5a26 39ec 2f5f 5f15
00000010: cc46 e6f3 d50f 6ae6 8988 d50e f8c4 d1c6
00000020: 5a2f 7c3e 490b 18d8 d867 4b6f 2549 1f6c
00000030: 34a9 a0d7 24d2 e9ac 9240 8247 10cb 4ba1

13.1.2 Presentation options for dump

By default, the hexadecimal display shows two bytes grouped together, and then a space. You can alter this behaviour using the group_by parameter.

(poke) dump :ascii 0 :size 0x40#B :group_by 4#B
00000000: 68f19a63 df2a8886 c466631c a7fdd5c7  h..c.*...fc.....
00000010: 3075746a 0adb03ca f5b1ff14 6166fa07  0utj........af..
00000020: 0dd3cfbd 8eff46a2 4152a81d 471beddf  ......F.AR..G...
00000030: a0501cae 8bfcec6f 7a4f5701 45ba9fc3  .P.....ozOW.E...

Another parameter is the cluster_by argument. By setting cluster_by to n, this causes dump to display an additional space after the nth group has been displayed, and also in the corresponding position in the ascii display:

(poke) dump :size 0x40#B :group_by 2#B :cluster_by 4
76543210  0011 2233 4455 6677  8899 aabb ccdd eeff  01234567 89ABCDEF
00000000: 91b8 540d d4dc 49ae  3320 ba7d efd1 16ab  ..T...I. 3 .}....
00000010: b1a8 5ea0 5846 8bea  f741 3f80 42bc 201f  ..^.XF.. .A?.B. .
00000020: 6e5e fa50 23fb f16a  d380 be8c fc98 d195  n^.P#..j ........
00000030: 7bbf fa3e 3fc2 43a4  2a1e 9763 2bd6 5d24  {..>?.C. *..c+.]$

Another variable that the user can customize is pk_dump_nonprintable_char. It defaults to '.', and contains the character to use in the ASCII dump to represent a non-printable character.

If you have a personal preference on how memory dumps should appear, you can set the relevant pk_dump_* variables. These global variables serve as the defaults for dump, so this way, you will not need to explicitly pass them when you call the function.

13.2 copy

The command copy allows to copy regions of data inside an IO space, or between different IO spaces.

This command has the following synopsis.

copy [:from offset] [:to offset] [:size offset]
     [:from_ios ios] [:to_ios ios]

All arguments are optional. When invoked with no arguments, copy does nothing.

The arguments from and size determine the region to copy. By default, this region is taken from the current IO space, but this can be overwritten using the optional from_ios argument.

The argument to tells copy where to copy the stuff. It defaults to from. Again, this offset is applied to the current IO space by default, but this can be overwritten using the to_ios argument.

Note that it is allowed for the source and destination ranges to overlap.

13.3 save

Use the save command in order to write a region from an IO space into a file in your file system.

This command has the following synopsis.

save [:ios ios] [:from offset] [:size offset] [:file string]
     [:append bool] [:verbose bool]

All arguments are optional. When invoked with no arguments, save does nothing.

The arguments from and size are used to determine the region of the IO space to save. This is how you would extract and save the contents of an ELF section to a file out.text:

(poke) var s = elf_section_by_name (Elf64_Ehdr @ 0#B, ".text")
(poke) save :file "out.text" :from s.sh_offset :size s.sh_size

Both from and size are arbitrary offsets. You should keep in mind however that files are byte oriented. Therefore saving, say, nine bits to a file will actually write two bytes.

By default save will extract the data from the current IO space. However, it is possible to specify an alternative IOS by using the ios argument.

By default save will truncate the output file, if it exists, before starting writing. If the argument append is set as true, however, it will append to the existing contents of the file. In this case, the file should exist.

13.4 extract

Use the extract command in order to create a temporary memory IO space with the contents of a mapped value.

This command the has the following synopsis.

extract :val val :to ios_name

Where :val is a mapped value, and :to is the name of the memory IOS to create (or use). The contents of value are copied to the beginning of the memory IOS *ios_name*.

13.5 scrabble

The scrabble command rearranges portions of an IO space based on a transformation defined by a pair of patterns.

This command has the following synopsis:

scrabble [:from offset] [:size offset]
         [:from_pattern string] [:to_pattern string]
         [:ent_size offset]
         [:from_ios ios] [:to_ios ios]

The optional arguments are documented below.

:from offset

Beginning of the range in the origin IO space from where to start rearranging units. Defaults to 0#B.

:size offset

Size of the range to scrabble in the origin IO space. Defaults to 0#B.

:from_pattern string

String specifying a sequence of units. Each character is an unit. Defaults to "".

:to_pattern string

String specifying the scrabbled units. Each character is an unit. If a character that doesn’t appear in the from_pattern is found in this string, it is ignored. Defaults to from_pattern.

:ent_size offset

Size of the entities to scrabble around. Defaults to 1#B.

:from_ios ios

IO space from where to obtain the data to scrabble. Defaults to the current IO space.

:to_ios ios

The IO space to where write the scrabbled data. Defaults to from_ios.

14 The Poke Language

14.1 Integers

Most of the values manipulated in Poke programs are whole numbers, also typically known as integers in computing parlance. This is because integers are pervasive in binary formats, often featuring unusual characteristics in terms of size and/or alignment. Single bits denoting flags or packed small integers are good examples of this. In order to ease the manipulation of such entities, and unlike most programming languages, Poke provides integer types of any number of bits and a rich set of accompanying operators.

14.1.1 Integer Literals

Integers literals can be expressed in several numeration bases.

Decimal numbers use the usual syntax [1-9][0-9]*. For example, 2345.

Octal numbers are expressed using a prefix 0o (or 0O) followed by one or more digits in the range [0-7]. Examples are 0o0, 0o100 and 0o777.

Hexadecimal numbers are expressed using a prefix 0x (or 0X) followed by one or more hexadecimal digits in the range [0-f]. Examples are 0x0 and 0xfe00ffff. Note that both the x in the prefix and the letters in the hexadecimal number are case insensitive. Thus, 0XdeadBEEF is a valid (but ugly as hell) literal.

Binary numbers are expressed using a prefix 0b (or 0B) followed by one or more binary digits in the range [0-1]. Examples of binary literals are 0b0 and 0B010.

Negative numbers, of any numeration base, are constructed using the minus operator as explained below. Therefore the minus symbol - in negative numbers is not part of the literal themselves.

14.1.1.1 The digits separator _

The character _ can appear anywhere in a numeric literal except as the first character. It is ignored, and its purpose is to make it easier for programmers to read them:

0xf000_0000_0000_0000
0b0000_0001_0000_0001

14.1.1.2 Types of integer literals

The type of a numeric literal is the smallest signed integer capable of holding it, starting with 32 bits, in steps of powers of two and up to 64 bits.¹⁰

So, for example, the value 2 has type int<32>, but the value 0xffff_ffff has type int<64>, because it is out of the range of signed 32-bit numbers.

A set of suffixes can be used to construct integer literals of certain types explicitly. L or l is for 64-bit integers. H or h is for 16-bit integers (also known as halves), B or b is for 8-bit integers (also known as bytes) and n or N is for 4-bit integers (also known as nibbles).

Thus, 10L is a 64-bit integer with value 0x0000_0000_0000_000A, 10H is a 16-bit integer with value 0x000A and 10b is a 8-bit integer with value 0x0A.

Similarly, the signed or unsigned attribute of an integer can be explicitly specified using the suffix u or U (the default are signed types). For example 0xffff_ffffU has type uint<32> and 0ub has type uint<8>. It is possible to combine width-indicating suffixes with signedness suffixes: 10UL denotes the same literal as 10LU.

The above rules guarantee that it is always possible to determine the width and signedness of an integer constant just by looking at it, with no ambiguity.

14.1.2 Characters

8-bit unsigned integers can use an alternative literal notation that is useful when working with ASCII character codes. Printable character codes can be denoted with 'c'.

Non-printable characters can be expressed using escape-sequences. The allowed sequences are:

\n

New-line character (ASCII 012).

\t

Tab character (ASCII 011).

\\

The backslash character.

\[0-9][0-9]?[0-9]?

Character whose ASCII code is the specified number, in octal.

Examples:

'o'
'\n'
'\t'
'\\'
'\0'

The type of a character literal is always char, aka uint<8>.

14.1.3 Booleans

Like in C, truth values in Poke are encoded using integers. Zero (0) denotes the logical value “false”, and any integer other than zero denotes the logical value “true”.

14.1.4 Integer Types

Most general-purpose programming languages provide a small set of integer types, each featuring a range corresponding to strategic storage sizes: basically, signed and unsigned variants of 8, 16, 32, 64 bits. As we have seen in the previous sections, suffixes like H or L are used in Poke for that purpose.

However, in conventional programming languages when integers having an “odd” width (like 13 bits, for example) get into play for whatever reason, the programmer is required to use the integer arithmetic operators (and sometimes bit-wise operators) herself, in a clever way, in order to achieve the desired results.

Poke, on the contrary, provides a rich set of integer types featuring different widths, in both signed and unsigned variants. The language operators are aware of these types, and will do the right thing when operating on integer values having different widths.

Unsigned integer types are specified using the type constructor uint<n>, where n is the number of bits. n should be an integer literal in the range [1,64]. Examples:

uint<1>
uint<7>
uint<64>

Similarly, signed integer types are created using the type constructor int<n>, where n is the number of bits. n should be an integer literal in the range [1,64]. Examples:

int<1>
int<8>
int<64>

Note that expressions are not allowed in the type integral constructor parameters. Not even constant expressions. Thus, things like int<foo> and uint<2+3> are not allowed.

14.1.5 Casting Integers

The right-associative unary operator cast as can be used to derive a new integer value having a different type from an existing value.

For example, this is how we would create a signed 12-bit integer value holding the value 666:

(poke) 666 as int<12>
(int<12>) 666

Note that the 666 literal is originally a 32-bit signed integer. The cast performs the conversion.

Casts between integer types are always allowed, and never fail. If the new type is narrower than the existing type, truncation may be performed to accommodate the value in its new type. For example, the expression 0x8765_4321 as uint<16> evaluates to 0x4321. If the new type is wider than the existing type, either zero-extension or sign-extension is performed depending on the signedness of the operand.

The semantics of the sign-extension operation depends on the signedness of the value being converted, and on the currently selected encoding for negative numbers.

When using two’s complement encoding, converting a signed value will always sign-extend regardless of the signedness of the target type. Thus:

(poke) -2H as uint<32>
0xfffffffeU
(poke) -2H as int<32>
0xfffffffe

Likewise, converting an unsigned value will always zero-extend regardless of the signedness of the target type. Thus:

(poke) 0xffffUH as uint<32>
0xffffU
(poke) 0xffffUH as int<32>
0xffff

14.1.6 Relational Operators

The following binary relational operators are supported on integer values, in descending precedence order:

• Equality == and inequality !=.
• Less than < and less or equal than <=.
• Greater than > and greater or equal than >=.

When applied to integer and character values, these operators implement an arithmetic ordering.

These operators resolve in boolean values encoded as 32-bit integers: 0 meaning false and 1 meaning true.

14.1.7 Arithmetic Operators

The following left-associative binary arithmetic operators are supported, in descending precedence order:

• Exponentiation **, multiplication *, integer division /, integer ceil-division /^ and modulus %.
• Addition + and subtraction -.

In all the binary arithmetic operations automatic promotions (coercions) are performed in the operands as needed. The rules are:

• If one of the operands is unsigned and the other operand is signed, the second is converted to an unsigned value.
• If the size in bits of one of the operands is bigger than the size of the other operand, the second is converted to the same number of bits.

The following right-associative unary arithmetic operators are supported:

• Unary minus - and unary plus +.

Finally, pre-increment, pre-decrement, post-increment and post-decrement operators ++ and -- are supported.

14.1.8 Bitwise Operators

The following left-associative bitwise binary operators are supported, in descending precedence order:

• Bitwise shift left <<. and bitwise shift right .>>.
• Bitwise AND &.
• Bitwise exclusive OR ^.
• Bitwise inclusive OR |.
• Bitwise concatenation :::.

Both <<. and .>> operators perform logical shifting. Unlike in many other programming languages, arithmetic right-shifting operators are not provided. This means that right shifting always inserts zeroes at the most-significant side of the value operand, whereas left shifting always inserts zeroes at the least-significant side of the value operand. Left shifting by a number of bits equal or bigger than the value operand is an error, and will trigger either a compile-time error or a run-time E_out_of_bounds exception.

Bitwise concatenation works with any integral type, of any bit length.

The following right-associative unary bitwise operators are supported:

• Bitwise complement ~.

14.1.9 Boolean Operators

The following left-associative, short-circuited binary logical operators are supported, in descending precedence order:

• Logical AND: &&.
• Inclusive OR: ||.

The following right-associative unary logical operators are supported:

• Logical negation !.

14.1.10 Integer Attributes

The following attributes are defined for integer values.

size

Gives an offset with the storage occupied by the string. This includes the terminating null. Examples:

(poke) 10'size
0x20UL#b
(poke) 10N'size
0x4UL#b
(poke) (10 as int<1>)'size
0x1UL#b

signed

Gives 1 if the value is a signed integer, 0 otherwise. Examples:

(poke) 10'signed
1
(poke) 10UL'signed
0

mapped

Always 0 for integers. (see Mapping).

14.2 Offsets

Poke uses united values to handle offsets and data sizes. This is a very central concept in poke.

14.2.1 Offset Literals

Poke provides a convenient syntax to provide united values, which are called offsets (because in a binary editor you mostly use them to denote offsets in the file you are editing):

12#B
7#b
1024#KB

The offsets above denote twelve bytes, seven bits and one thousand twenty four kilobytes, respectively. The unit can be separated from the magnitude by blank characters, so you can write the following instead if you are so inclined:

12 #B
7 #b
(1024 * 1024) #Kb

Note how the magnitude part of an offset doesn’t need to be constant. If the variable a contains an integer, this is how you would denote “a bytes”:

a#B

In the offset syntax units are specified as #unit, where unit is the specification of an unit. See the next section for details.

14.2.2 Offset Units

There are several ways to express the unit of an offset, which is always interpreted as a multiple of the basic unit, which is the bit (one bit).

14.2.2.1 Named Units

The first way is to refer to an unit by name. For example, 2#B for two bytes. Units are defined using the unit construction:

unit name = constant_expression [, name = constant_expression]...;

where name is the name of the unit, and constant_expression is a constant expression that should evaluate to an integral value. The resulting value is always coerced into an unsigned 64-bit integer.

Note that unit names live in a different namespace than variables and types. However, when a given name is both a type name and an unit name in an unit context, the named unit takes precedence:

(poke) type xx = int
(poke) unit xx = 2
(poke) 1#xx
1#2

Many useful units are defined in the standard library. See Standard Units.

14.2.2.2 Arbitrary Units

It is also possible to express units in multiples of the base unit, which is the bit. Using this syntax, it becomes possible to express offsets in any arbitrary unit, as disparate as it may seem:

17#3
0#12
8#1

That’s it: 17 units of 3 bits each, zero units of 12 bits each, and eight units of 1 bit each. Note that the unit should be greater than 0.

14.2.2.3 Types as Units

But then, why stop there? Poking is all about defining data structures and operating on them… so why not use these structures as units as well? Consider the following struct:

type Packet = struct { int i; long j; };

The size of a Packet is known at compile time (which is not generally true for Poke structs). Wouldn’t it be nice to use it as a unit in offsets? Sure it is:

23#Packet

The above is the size occupied by 23 packets. Any type whose size is known at compile time can be specified as an offset unit.

Expressing offsets as united values also relieves the programmer from doing many explicit unit conversions: poke can do them for you. Consider for example an ELF section header. One of its fields is the size of the described section, in bytes:

type Elf64_Shdr =
  struct
  {
   …
   offset<Elf64_Xword,B> sh_size;
   …
  };

If a given section is to contain, say, relocations with addends, we can set its size doing something like this:

shdr.sh_size = 10#Elf64_Rela;

Instead of doing the conversion to bytes explicitly.

If the magnitude of an offset is 1 then it is allowed to omit it entirely. To denote one kilobyte, for example, we can write #KB.

14.2.3 Offset Types

Offset types are denoted as offset<base_type,unit>, where base_type is an integer type and unit the specification of an unit.

The offset base type is the type of the magnitude part of the united value. It can be any integer type, signed or unsigned, of any size.

The unit specification should be one of the unit identifiers that are allowed in offset literals (see above), a constant positive integer or the name of a Poke type whose size is known at compile time.

Let’s see some examples. A signed 32-bit offset expressed in bytes has type offset<int<32>,B>. An unsigned 12-bit offset expressed in kilobits has type offset<uint<12>,Kb>. The latter type can also be written using an explicit integer unit like in offset<uint<12>,1024>. Finally, a signed 64-bit offset in units of “packets”, where a packet is denoted with a Poke type Packet has type offset<uint<64>,Packet>.

14.2.4 Casting Offsets

The right-associative unary operator cast as can be used to derive a new offset value having a different type from an existing value.

For example, this is how we would create a signed 12-bit offset in units bytes:

(poke) 1024#b as offset<int<12>,B>
(int<12>) 128#B

The same rules governing conversion of integers apply for the magnitude part. Depending on the unit, there can be truncation, like in:

(poke) 9#b as offset<int,B>
1#B

14.2.5 Offset Operations

Poke supports a little algebra for offsets.

14.2.5.1 Addition and subtraction

The addition or subtraction of two offsets results in another offset. Examples:

(poke) 1#B + 1#b
9#b
(poke) 2#KB - 1024#B
1024#B

The unit of the result is the greatest common divisor of the units of the operands.

The operators ++ and --, in their prefix and suffix versions, can be applied to offsets as well. The step used in the increment/decrement is one unit.

14.2.5.2 Multiplication by a scalar

Multiplying an offset by a magnitude gives you another offset. Examples:

(poke) 8#b * 2
16#b
(poke) 16#MB * 0
0#MB

The unit of the result is the same as the unit of the offset operand.

Note that multiplying two offsets is not supported. This makes sense, since computer memory is linear, and therefore it wouldn’t make any sense to have units like #B2.

14.2.5.3 Division

Dividing two offsets gives you a magnitude. Examples:

(poke) 16#b / 1#B
2
(poke) 1024#MB / 512#Mb
16

Dividing offsets is the Pokish way of converting memory magnitudes between different units: just use units like you do when doing physics or working with units in other contexts.

For example, using the syntactic trick of omitting the magnitude (in which case it is assumed to be 1) it is very natural to write something like the following to convert from kilobits to bytes:

(poke) 24 #Kb/#B
3072

There is also a ceil-division operator for offsets, with the same semantics as the ceil-division for integers:

(poke) 10#B /^ 3#B
4

14.2.5.4 Division by an integer

Dividing an offset by an integer gives you an offset. Example:

(poke) 8#B / 2
4#B

14.2.5.5 Modulus

The modulus of two offsets gives you another offset with the expected semantics. Examples:

(poke) 9#b % 1#B
1#b
(poke) 1#B % 9#b
8#b

The unit of the result is the greatest common divisor of the units of the operands.

14.2.6 Offset Attributes

The following attributes are defined for offset values.

size

Gives an offset with the storage occupied by the offset. Examples:

(poke) 10#B'size
0x20UL#b
(poke) 10N#B'size
0x4UL#b

magnitude

Gives the magnitude part of the offset. Examples:

(poke) 10#B'magnitude
10
(poke) 2H#b'magnitude
2H

unit

Gives a number with the unit of the offset, expressed in bits. Examples:

(poke) 10#B'unit
8UL
(poke) 2H#b'unit
1UL

mapped

Always 0 for offsets. (see Mapping).

14.3 Strings

Poke supports a notion of strings which is very similar to the C programming language: a string value is a sequence of characters that is terminated by the so-called null character.

The standard library provides functions which process strings. See String Functions.

14.3.1 String Literals

NULL-terminated sequences of ASCII codes can be denoted using the following syntax:

"foo"

Poke string values are very similar to C strings. They comprise a sequence of 8-bit character codes, terminated by the value 0UB.

The following escape sequences are supported inside string literals:

\n

Denotes a new line character.

\t

Denotes an horizontal tab.

\\

Denotes a backlash \ character.

\"

Denotes a double-quote " character.

\num

num is a number in the range 1 to 255 (inclusive) in base 8.

\xnum

num is a number in the range 1 to 255 (inclusive) in base 16.

Strings cannot contain 0UB character, and inserting one using escape sequences (\0 or \x0) leads to compilation error.

14.3.2 String Types

Every string value in Poke is of type string.

14.3.3 String Indexing

Poke supports accessing the characters in a string using the array indexing notation. The indexes are in the [0,n] range, where n is the length of the string minus one. Note the length doesn’t include the null character, i.e. it is not possible to access the terminating null. Examples:

(poke) "foo"[0]
0x66UB
(poke) "foo"[1]
0x6fUB

If the passed index is less than zero or it is too big, an E_out_of_bounds exception is raised:

(poke) "foo"[-1]
unhandled out of bounds exception
(poke) "foo"[3]
unhandled out of bounds exception

14.3.4 String Concatenation

Strings can be concatenated using the + operator. This works like this:

(poke) "foo" + "bar"
"foobar"

Note how the null character terminating the first string is removed. Therefore, the length of the concatenation of two given strings of lengths N and M is always N+M-1.

Concatenation and indexing are useful together for building strings. A string can be created empty, and additional characters added to it by means of concatenation:

(poke) var bytes = "";
(poke) bytes = bytes + 'x' as string;

Then, we can retrieve characters from the string we built using indexing:

(poke) bytes[0]
0x78UB

Additionally, the * operator allows to “multiply” a string by concatenating it with itself a given number of times. This works like this:

(poke) "foo" * 3
"foofoofoo"
(poke) "foo" * 0
""

This is useful for building strings whose length is not known at compile time. For example:

fun make_empty_string = (int length) string:
{
   return " " * length;
}

14.3.5 String Attributes

The following attributes are defined for string values.

length

Gives the number of characters composing the string, not counting the terminating null. Examples:

(poke) "foo"'length
3UL
(poke) ""'length
0UL

size

Gives an offset with the storage occupied by the string. This includes the terminating null. Examples:

(poke) "foo"'size
32UL#b
(poke) ""'size
8UL#b

mapped

Always 0 for strings. (see Mapping).

14.4 Arrays

Arrays are homogeneous collections of values.

14.4.1 Array Literals

Array literals are constructed using the following syntax:

[exp,exp…]

where exp is an arbitrary expression.

For example, [1,2,3] constructs an array of three signed 32-bit integers. Likewise, ['a','b','c'] constructs an array of three unsigned 8-bit integers (characters). For convenience, a trailing comma is accepted but not required.

The type of the array literal is inferred from the type of its elements. Accordingly, all the elements in an array literal must be of the same type. Examples of invalid array literals, that will raise a compilation-time error if evaluated, are:

[1,2u,3]
[1,0xffff_ffff,3]
['a',"b",'c']

Array literals must contain at least one element. Accordingly, [] is not a valid array literal.

This is how a 3x3 matrix could be constructed using an array of arrays:

[[1,2,3],[4,5,6],[7,8,9],]

It is possible to refer to specific elements when constructing array literals. For example, [1,2,3,.[3] = 4] denotes the same array as [1,2,3,4].

This allows creating arrays without having to specify all its elements; every unspecified element takes the value of the first specified element to its right. For example, [.[2] = 2] denotes the same array as [2,2,2].

Note that an array element can be referenced more than once. When that happens, the final value of the element is the last specified. For example, [1,2,3,.[1]=10] denotes the array [1,10,3].

14.4.2 Array Types

There are three different kind of array types in Poke.

Unbounded arrays have no explicit boundaries. Examples are int[] or Elf64_Shdr[]. Arrays can be bounded by number of elements specifying a Poke expression that evaluates to an integer value. For example, int[2]. Finally, arrays can be bounded by size specifying a Poke expression that evaluates to an offset value. For example, int[8#B].

14.4.2.1 Writing unbounded array literals

The type of an array literal is always bounded by number of elements. For example, the type of [1,2,3] is int[3]. If what we want is an unbounded array literal we can obtain it with a case like [1,2,3] as int[].

14.4.2.2 Array boundaries and closures

Poke arrays are rather peculiar. One of their seemingly bizarre characteristics is the fact that the expressions calculating their boundaries (when they are bounded) evaluate in their own lexical environment, which is captured. In other words: the expressions denoting the boundaries of Poke arrays conform closures. Also, the way they evaluate may be surprising. This is no capricious.

When an array type is bounded, be it by number of elements or by size, the expression indicating the boundary doesn’t need to be constant and it can involve variables. For example, consider the following type definition:

var N = 2;
type List = int[N*2];

Let’s map a List at some offset:

(poke) List @ 0#B
[0x746f6f72,0x303a783a,0x723a303a,0x3a746f6f]

As expected, we get an array of four integers. Very good, obviously the boundary expression N*2 got evaluated when defining the type List, and the result of the evaluation was 4, right?. Typical semantics like in my garden variety programming language…? Right?!?

Well, not really. Let’s modify the value of N and map a List again…

(poke) N = 1
(poke) List @ 0#B
[0x746f6f72,0x303a783a]

Yes, The boundary of the array type changed… on, this is Poke, was you really expecting something typical? ;)

What happens is that at type definition time the lexical environment is captured and a closure is created. The body of the closure is the expression. Every time the type is referred, the closure is re-evaluated and a new value is computed.

Consequently, if the value of a variable referred in the expression changes, like in our example, the type itself gets updated auto-magically. Very nice but, why is Poke designed like this? Just to impress the cat? Nope.

In binary formats, and also in protocols, the size of some given data is often defined in terms of some other data that should be decoded first. Consider for example the following definition of a Packet:

type Packet =
  struct
  {
    byte size;
    byte[size] payload;
  };

Each packet contains a 8-bit integer specifying the size of the payload transported in the packet. The payload, a sequence of size bytes, follows.

In struct types like the above, the boundaries of arrays depend on fields that have been decoded before and that exist, like variables, in the lexical scope captured by the struct type definition (yes, these are also closures, but that’s for another article.) This absolutely depends on having the array types evaluate their bounding expressions when the type is used, and not at type definition time.

To show this property in action, let’s play a bit:

(poke) var data = byte[4] @ 0#B
(poke) data[0] = 2
(poke) data[1] = 3
(poke) data[2] = 4
(poke) data[3] = 5
(poke) dump
76543210  0011 2233 4455 6677 8899 aabb ccdd eeff
00000000: 0203 0405 0000 0000 0000 0000 0000 0000
00000010: 0000 0000 0000 0000 0000 0000 0000 0000
(poke) var p1 = Packet @ 0#B
(poke) var p2 = Packet @ 1#B
(poke) p1
Packet {size=0x2UB,payload=[0x3UB,0x4UB]}
(poke) p2
Packet {size=0x3UB,payload=[0x4UB,0x5UB,0x0UB]}

Now, let’s change the data and see how the sizes of the payloads are adjusted accordingly:

(poke) data[0] = 1
(poke) data[1] = 0
(poke) p1
Packet {size=0x1UB,payload=[0x0UB]}
(poke) p2
Packet {size=0x0UB,payload=[]}

So, as we have seen, Poke’s way of handling boundaries in array types allows data structures to adjust to the particular data they contain, so usual in binary formats. This is an important feature, that gives Poke part of its feel and magic.

14.4.3 Array Constructors

Array literals can only specify arrays which a constant number of elements. They also require initial values for their elements, and it is also not possible to construct an array literal denoting an empty array of some given type.

Array constructors provide a suitable way to build such array values. The syntax is:

array_type ([value])

Where array_type is an array type, or the name of an array type, and value is an optional argument that specify the value to which the elements of the new array are initialized. If no value is provided in the constructor then default values are calculated (XXX: xref to default values.)

Examples:

(poke) int[]()
[]
(poke) int[2]()
[0,0]
(poke) int[2](10)
[10,10]
(poke) offset<int,B>[2](16#b)
[2#B,2#B]
(poke) string[3]()
["","",""]
(poke) int[][3](int[2](666))
[[666,666],[666,666],[666,666]]
(poke) Packet[3]()
[Packet {...}, Packet {...}, Packet {...}]

14.4.4 Array Comparison

The equality operator (==) and the inequality operator (!=) can be applied to arrays. Examples:

(poke) [1,2,3] == [1,2,3]
1
(poke) [[1,2],[3,4]] != [[5,6],[7,8]]
1

Note that the array elements are compared recursively.

14.4.5 Array Indexing

Arrays are indexed using the usual notation, providing an index enclosed between square brackets with [ and ]:

(poke) [1,2,3][0]
1
(poke) [1,2,3][1]
2

The index should be an expression that evaluates to an integer value, and it is promoted to an unsigned 64-bit integer when needed.

The valid range for the index is [0,n] where n is the number of elements stored in the array minus one. If the passed integer is out of that range, an E_out_of_bounds exception is raised:

(poke) [1,2,3][-1]
unhandled out of bounds exception
(poke) [1,2,3][3]
unhandled out of bounds exception

14.4.6 Array Trimming

Indexing is used to fetch elements from arrays. Another operation, called trimming, allows you to extract a subset of the array, as another array.

Trims use the following notation, where a range is specified between square brackets. The left sides of the range are closed, whereas the right sides of the range are open:

(poke) [1,2,3][0:2]
[1,2]
(poke) [1,2,3][1:2]
[2]
(poke) [1,2,3][0:3]
[1,2,3]

If the minimum side of the range is omitted, it is assumed to be zero. If the maximum side of the range is omitted, it is assumed to be the length of the trimmed array:

(poke) [1,2,3][:2]
[1,2]
(poke) [1,2,3][1:]
[2,3]
(poke) [1,2,3][:]
[1,2,3]

The elements of the base array and the trimmed sub-array are copied by shared value, exactly like when passing arguments to functions. This means that for simple types, copies of the elements are done:

(poke) var a = [1,2,3]
(poke) var s = a[1:2]
(poke) s[0] = 66
(poke) a
[1,2,3]

However, for complex types like arrays and structs, the values are shared:

(poke) var a = Packet[] @ 0#B
(poke) var s = a[1:2]
(poke) s[0].field = 66
(poke) a[1].field
66

There is an alternative syntax that can be used in order to denote trims, in which the length of the trim is specified instead of the index of the second element. It has this form:

(poke) [1,2,3][0+:3]
[1,2,3]
(poke) [1,2,3][1+:2]
[2,3]
(poke) [1,2,3][1+:0]
[]

14.4.7 Array Elements

The in operator can be used to determine whether a given element is stored in an array. Examples:

(poke) 2 in [1,2,3]
1
(poke) 10#B in [2#b,10*8#b,3#b]
1
(poke) 30#B in [2#b,10*8#b,3#b]
0

14.4.8 Array Concatenation

The operator + can be used to build new arrays resulting from the concatenation of the elements of two arrays given as operands:

(poke) [1,2] + [3,4,5]
[1,2,3,4,5]

The two arrays given as operands shall have the same elements of the same type. The resulting array is always unbounded.

14.4.9 Array Attributes

The following attributes are defined for array values.

size

Gives an offset with the storage occupied by the complete array. Example:

(poke) [1,2,3]'size
96UL#b

length

Gives the number of elements stored in the array. Example:

(poke) [1,2,3]'length
3

mapped

Boolean indicating whether the array is mapped. Examples:

(poke) var a = [1,2,3]
(poke) var b = int[3] @ 0#B
(poke) a'mapped
0
(poke) b'mapped
1

strict

Boolean indicating whether the array is mapped in strict mode. Examples:

(poke) var a = int[3] @ 0#B
(poke) a'strict
1
(poke) var a = int[3] @! 0#B
(poke) a'strict
0

14.5 Structs

Structs are the main abstraction that Poke provides to structure data. They contain heterogeneous collections of values.