picotool: Tools and Python libraries for manipulating PICO-8 game files

PICO-8 is a fantasy game console by Lexaloffle Games. The PICO-8 runtime environment runs cartridges (or carts): game files containing code, graphics, sound, and music data. The console includes a built-in editor for writing games. Game cartridge files can be played in a browser, and can be posted to the Lexaloffle bulletin board or exported to any website.

picotool is a suite of tools and libraries for building and manipulating PICO-8 game cartridge files. The suite is implemented in, and requires, Python 3. The tools can examine and transform cartridges in various ways, and you can implement your own tools to access and modify cartridge data with the Python libraries.

Useful tools include:

• p8tool build: assembles cartridges from multiple sources, as part of a game development workflow
• p8tool stats: reports statistics on one or many cartridge files
• p8tool listlua: prints the Lua code of a cartridge
• p8tool luafind: searches the Lua code of a collection of cartridges
• p8tool luafmt: formats the Lua code of a cartridge to make it easier to read

There are additional tools that are mostly useful for demonstrating and troubleshooting the library: writep8, listtokens, printast, luamin. A separate demo, p8upsidedown, uses picotool to transform the code and data of a game to turn it upsidedown.

picotool supports reading and writing both of PICO-8's cartridge file formats: the text-based format .p8, and the PNG-based binary format .p8.png.

How do most people use picotool?

Most people use picotool for the luamin command, which condenses the Lua code of a cart to use as few characters as possible. This is useful if you have a large game whose code is above PICO-8's character limit but below the token limit. In this rare case, luamin helps you keep your developer version easy to read and modify while you polish the game for publication. In most cases, your cart will exceed the token limit first, and minification doesn't help with unusual cases such as long string data.

I originally created picotool for general purpose build workflows, especially the build tool. Since then, PICO-8 has added features such as #include support, making some of the build features unnecessary. (#include doesn't quite work like require() but it's good enough for most things!) I'm still interested in the potential for token optimization features such as dead code elimination, but these are not yet implemented.

I hope picotool can be the basis for your custom workflows and cart manipulation experiments, so you don't have to write your own cart read/write routines. If there's a feature you'd like to see in the libraries or the tools, please let me know!

Installing picotool

To install the picotool tools and libraries:

1. Install Python 3 version 3.4 or later, if necessary. (picotool has not been tested with Python 2.)

2. Download and unpack the zip archive, or use Git to clone the Github repository.

   • Unpacking the zip archive creates a root directory named picotool-master.
   • When cloning the repo, this is just picotool, or whatever you named it when you cloned it.

3. Change to the unpacked archive or clone directory from the above step.

4. Install the software. The dot (.) tells pip install to use the current working directory, which should be picotool-master (from the .zip file) or picotool (from git clone).

   pip install .

You should now have a p8tool command in your path.

Using picotool

To use a tool, you run the p8tool command with the appropriate arguments. Without arguments, it prints a help message. The first argument is the name of the tool to run (such as stats), followed by the arguments expected by that tool.

For example, to print statistics about a cart named helloworld.p8.png:

p8tool stats helloworld.p8.png

p8tool build

The build tool creates or updates a cartridge file using other files as sources. It is intended as a part of a game development workflow, producing the final output cartridge.

The tool takes the filename of the output cartridge, with additional arguments describing the build. If the output cartridge does not exist, the build starts with an empty cartridge. Otherwise, it uses the existing cartridge as the default, and overwrites sections of it based on the arguments.

For example, you can create a cartridge in PICO-8, use PICO-8's built-in graphics and sound editors, then use p8tool build to replace the Lua code with the contents of a .lua file:

p8tool build mygame.p8.png --lua mygame.lua

As another example, to create a new cartridge using the spritesheet (gfx) from one cartridge file, music (sfx, music) from another, and Lua code from a .lua file:

p8tool build mygame.p8.png --gfx mygamegfx.p8 --sfx radsnds.p8.png --music radsnds.p8.png --lua mygame.lua

You can also erase a section of an existing cart with an argument such as --empty-map.

The available arguments are as follows:

• --lua LUA: use Lua code from the given cart or .lua file
• --gfx GFX: use spritesheet from the given cart
• --gff GFF: use sprite flags from the given cart
• --map MAP: use map from the given cart
• --sfx SFX: use sound effects from the given cart
• --music MUSIC: use music patterns from the given cart
• --empty-lua: use an empty Lua code section
• --empty-gfx: use an empty spritesheet
• --empty-gff: use empty sprite flags
• --empty-map: use an empty map
• --empty-sfx: use empty sound effects
• --empty-music: use empty music patterns

If the output cart filename ends with .p8.png, the result is a cartridge with a label image. If the file already exists, the cartridge label is reused. If the file does not exist, an empty cartridge label is used. To use a non-empty label, you must open the cart in PICO-8, take a screenshot (press F6 while running), set the title and byline in the first two lines of code (as Lua comments), then save the .p8.png file from PICO-8. Future runs of p8tool build will reuse the label.

Packages and the require() function

p8tool build supports a special feature for organizing your Lua code, called packages. When loading Lua code from a file with the --lua mygame.lua argument, your program can call a function named require() to load Lua code from another file. This is similar to the require() function available in some other Lua environments, with some subtle differences due to how picotool does this at build time instead of at run time.

Consider the following simple example. Say you have a function you like to use in several games in a file called mylib.lua:

function handyfunc(x, y)
  return x + y
end

handynumber = 3.14

Your main game code is in a file named mygame.lua. To use the handyfunc() function within mygame.lua, call require() to load it:

require("mylib")

result = handyfunc(2, 3)
print(result)

r = 5
area = handynumber * r * r

All globals defined in the required file are set as globals in your program when require() is called. While this is easy enough to understand, this has the disadvantage of polluting the main program's global namespace.

A more typical way to write a Lua package is to put everything intended to be used by other programs in a table:

HandyPackage = {
  handyfunc = function(x, y)
    return x + y
  end,
  handynumber = 3.14,
}

Then in mygame.lua:

require("mylib")

result = HandyPackage.handyfunc(2, 3)

This is cleaner, but still has the disadvantage that the package must be known by the global name HandyPackage wihtin mygame.lua. To fix this, Lua packages can return a value with the return statement. This becomes the return value for the require() call. Furthermore, Lua packages can declare local variables that are not accessible to the main program. You can use these features to hide explicit names and return the table for the package:

local HandyPackage = {
  handyfunc = function(x, y)
    return x + y
  end,
  handynumber = 3.14,
}

return HandyPackage

The main program uses the return value of require() to access the package:

HandyPackage = require("mylib")

result = HandyPackage.handyfunc(2, 3)

The require() function only evaluates the package's code once. Subsequent calls to require() with the same string name do not reevaluate the code. They just return the package's return value. Packages can safely require other packages, and only the first encountered require() call evaluates the package's code.

Where packages are located

The first argument to require() is a string name. picotool finds the file that goes with the string name using a library lookup path. This is a semicolon-delimited (;) list of filesystem path patterns, where each pattern uses a question mark (?) where the string name would go.

The default lookup path is ?;?.lua. With this path, require("mylib") would check for a file named mylib, then for a file named mylib.lua, each in the same directory as the file containing the require() call. The lookup path can also use absolute filesystem paths (such as /usr/share/pico8/lib/?.lua). You can customize the lookup path either by passing the --lua-path=... argument on the command line, or by setting the PICO8_LUA_PATH environment variable.

For example, with this environment variable set:

PICO8_LUA_PATH=?;?.lua;/home/dan/p8libs/?/?.p8

The require("3dlib") statement would look for these files, in this order, with paths relative to the file containing the require() statement:

3dlib
3dlib.lua
/home/dan/p8libs/3dlib/3dlib.p8

To prevent malicious code from accessing arbitrary files on your hard drive (unlikely but it's nice to prevent it), the require() string cannot refer to files in parent directories with ../. It can refer to child directories, such as require("math/linear").

As with Lua, packages are remembered by the string name used with require(). This means it is possible to have two copies of the same package, each known by a different name, if it can be reached two ways with the lookup path. For example, if the file is named foo.lua and the lookup path is ?;?.lua, require("foo") and require("foo.lua") treat the same file as two different packages.

Packages and game loop functions

When you write a library of routines for PICO-8, you probably want to write test code for those routines. picotool assumes that this test code would be executed in a PICO-8 game loop, such that the library can be in its own test cart. For this purpose, you can write your library file with _init(), _update() or _update60(), and _draw() functions that test the library. By default, require() will strip the game loop functions from the library when including it in your game code so they don't cause conflicts or consume tokens.

For example:

local HandyPackage = {
  handyfunc = function(x, y)
    return x + y
  end,
  handynumber = 3.14,
}

function _update()
  test1 = HandyPackage.handyfunc(2, 3)
end
function _draw()
  cls()
  print('test1 = '..test1)
end

return HandyPackage

If you want to keep the game loop functions present in a package, you can request them with a second argument to require(), like so:

require("mylib", {use_game_loop=true})

How require() actually works

Of course, PICO-8 does not actually load packages from disk when it runs the cartridge. Instead, picotool inserts each package into the cartridge code in a special way that replicates the behavior of the Lua require() feature.

When you run p8tool build with the --lua=... argument, picotool scans the code for calls to the require() function. If it sees any, it loads and parses the file associated with the string name, and does so again if the required file also has require() calls.

Each required library is stored once as a function object in a table inserted at the top of the final cartridge's code. A definition of the require() function is also inserted that finds and evaluates the package code in the table as needed.

To match Lua's behavior, require() maintains a table named package.loaded that maps string names to return values. As with Lua, you can reset this value to nil to force a require() to reevaluate a package.

This feature incurs a small amount of overhead in terms of tokens. Each library uses tokens for its own code, plus a few additional tokens for storing it in the table. The definition for require() is another 40 tokens or so. Naturally, the inserted code also consumes characters.

Formatting or minifying Lua in a built cart

You can tell p8tool build to format or minify the code in the built output using the --lua-format or --lua-minify command line arguments, respectively.

p8tool build mycart.p8.png --lua=mygame.lua --lua-format

This is equivalent to building the cart then running p8tool luafmt or p8tool luamin on the result.

The build command supports the options --keep-names-from-file=<filename> and --keep-all-names. See luamin, below, for more information.

p8tool stats

The stats tool prints statistics about one or more carts. Given one or more cart filenames, it analyzes each cart, then prints information about it.

% p8tool stats helloworld.p8.png
hello world (helloworld.p8.png)
by zep
version: 0  lines: 48  chars: 419  tokens: 134

This command accepts an optional --csv argument. If provided, the command prints the statistics in a CSV format suitable for importing into a spreadsheet. This is useful when tallying statistics about multiple carts for comparative analysis.

p8tool --csv stats mycarts/*.p8* >cartstats.csv

p8tool listlua

The listlua tool extracts the Lua code from a cart, then prints it exactly as it appears in the cart.

% p8tool listlua helloworld.p8.png
-- hello world
-- by zep

t = 0

music(0)

function _update()
 t += 1
end

function _draw()
 cls()

...

p8tool luafmt

The luafmt tool rewrites the Lua region of a cart to make it easier to read, using regular indentation and spacing. This does not change the token count, but it may increase the character count, depending on the initial state of the code.

The command takes one or more cart filenames as arguments. For each cart with a name like xxx.p8.png, it writes a new cart with a name like xxx_fmt.p8.

% p8tool luafmt helloworld.p8.png
% cat helloworld_fmt.p8
pico-8 cartridge // http://www.pico-8.com
version 5
__lua__
-- hello world
-- by zep

t = 0

music(0)

function _update()
  t += 1
end

function _draw()
  cls()

  for i=1,11 do
    for j0=0,7 do
      j = 7-j0
      col = 7+j
...

By default, the indentation width is 2 spaces. You can change the desired indentation width by specifying the --indentwidth=... argument:

% p8tool luafmt --indentwidth=4 helloworld.p8.png
% cat helloworld_fmt.p8
...
function _update()
    t += 1
end

function _draw()
    cls()

    for i=1,11 do
        for j0=0,7 do
            j = 7-j0
            col = 7+j
...

The current version of luafmt is simple and mostly just adjusts indentation. It does not adjust spaces between tokens on a line, align elements to brackets, or wrap long lines.

p8tool luafind

The luafind tool searches for a string or pattern in the code of one or more carts. The pattern can be a simple string or a regular expression that matches a single line of code.

Unlike common tools like grep, luafind can search code in .p8.png carts as well as .p8 carts. This tool is otherwise not particularly smart: it's slow (it runs every file through the parser), and doesn't support fancier grep-like features.

% p8tool luafind 'boards\[.*\]' *.p8*
test_gol.p8.png:11:  boards[1][y] = {}
test_gol.p8.png:12:  boards[2][y] = {}
test_gol.p8.png:14:    boards[1][y][x] = 0
test_gol.p8.png:15:    boards[2][y][x] = 0
test_gol.p8.png:20:boards[1][60][64] = 1
test_gol.p8.png:21:boards[1][60][65] = 1
test_gol.p8.png:22:boards[1][61][63] = 1
test_gol.p8.png:23:boards[1][61][64] = 1
test_gol.p8.png:24:boards[1][62][64] = 1
test_gol.p8.png:30:  return boards[bi][y][x]
test_gol.p8.png:36:      pset(x-1,y-1,boards[board_i][y][x] * alive_color)
test_gol.p8.png:54:          ((boards[board_i][y][x] == 1) and neighbors == 2)) then
test_gol.p8.png:55:        boards[other_i][y][x] = 1
test_gol.p8.png:57:        boards[other_i][y][x] = 0

You can tell luafind to just list the names of files containing the pattern without printing the lines using the --listfiles argument. Here's an example that looks for carts that contain examples of Lua OO programming:

% p8tool luafind --listfiles 'self,' *.p8*
11243.p8.png
12029.p8.png
12997.p8.png
13350.p8.png
13375.p8.png
13739.p8.png
15216.p8.png
...

p8tool writep8

The writep8 tool writes a game's data to a .p8 file. This is mostly useful for converting a .p8.png file to a .p8 file. If the input is a .p8 already, then this just makes a copy of the file. (This can be used to validate that the picotool library can output its input.)

The command takes one or more cart filenames as arguments. For each cart with a name like xxx.p8.png, it writes a new cart with a name like xxx_fmt.p8.

% p8tool writep8 helloworld.p8.png
% cat helloworld_fmt.p8
pico-8 cartridge // http://www.pico-8.com
version 5
__lua__
-- hello world
-- by zep

t = 0

music(0)

function _update()
 t += 1
end

function _draw()
 cls()

...

p8tool luamin

The luamin tool rewrites the Lua region of a cart to use as few characters as possible. It does this by discarding comments and extraneous space characters, and renaming variables and functions. This does not change the token count.

The command takes one or more cart filenames as arguments. For each cart with a name like xxx.p8.png, it writes a new cart with a name like xxx_fmt.p8.

By default, luamin renames variables, attributes, and labels to use as few characters as possible. You can disable this behavior with a command line flag: --keep-all-names

You can disable renaming just for specific names by providing a text file listing names, with this command line option: --keep-names-from-file=<filename> This file supports the following features:

• Each name appears on its own line.
• Blank lines are ignored.
• Lines that begin with a # are ignored, so you can have comments in this file.

Disabling renaming is useful in cases where code refers to object attributes using both dot syntax (foo.bar) and indexing syntax (foo["bar"]). In this case, luamin renames the dot syntax but can't rename the indexing syntax (because "bar" can be any expression). You can add bar to the list of names to keep to prevent this, while still allowing renaming for other symbols (including foo in this example).

luamin always preserves the first two comments that appear before any other code. PICO-8 uses these comments as the title and author metadata for the label when exporting a .p8.png.

Statistically, you will run out of tokens before you run out of characters, and minifying is unlikely to affect the compressed character count. Carts are more useful to the PICO-8 community if the code in a published cart is readable and well-commented. That said, some authors have found luamin useful in the late stages of writing a large cart that includes blob data strings, where you might hit the character limit first and you need to squeeze in a few more characters to finish the game.

% p8tool luamin helloworld.p8.png
% cat helloworld_fmt.p8
pico-8 cartridge // http://www.pico-8.com
version 5
__lua__
a = 0
music(0)
function _update()
a += 1
end
function _draw()
cls()
...

p8tool listtokens

The listtokens tool is similar to listlua, but it identifies which characters picotool recognizes as a single token.

% p8tool listtokens ./picotool-master/tests/testdata/helloworld.p8.png
<-- hello world>
<-- by zep>

<0:t>< ><1:=>< ><2:0>

<3:music><4:(><5:0><6:)>

<7:function>< ><8:_update><9:(><10:)>
< ><11:t>< ><12:+=>< ><13:1>
<14:end>

<15:function>< ><16:_draw><17:(><18:)>
< ><19:cls><20:(><21:)>

...

When picotool parses Lua code, it separates out comments, newlines, and spaces, as well as proper Lua tokens. The Lua tokens appear with an ascending number, illustrating how picotool counts the tokens. Non-token elements appear with similar angle brackets but no number. Newlines are rendered as is, without brackets, to make them easy to read.

Note: picotool does not currently count tokens the same way PICO-8 does. One purpose of listtokens is to help troubleshoot and fix this discrepancy. See "Known issues."

p8tool printast

The printast tool prints a visualization of the abstract syntax tree (AST) determined by the parser. This is a representation of the structure of the Lua code. This is useful for understanding the AST structure when writing a new tool based on the picotool library.

% p8tool printast ./picotool-master/tests/testdata/helloworld.p8.png

Chunk
  * stats: [list:]
    - StatAssignment
      * varlist: VarList
        * vars: [list:]
          - VarName
            * name: TokName<'t', line 3 char 0>
      * explist: ExpList
        * exps: [list:]
          - ExpValue
            * value: 0
    - StatFunctionCall
      * functioncall: FunctionCall
        * exp_prefix: VarName
          * name: TokName<'music', line 5 char 0>
        * args: FunctionArgs
          * explist: ExpList
            * exps: [list:]
              - ExpValue
                * value: 0
    - StatFunction
      * funcname: FunctionName
        * namepath: [list:]
          - TokName<'_update', line 7 char 9>
        * methodname: None
      * funcbody: FunctionBody
        * parlist: None
        * dots: None
        * block: Chunk

...

Building new tools

picotool provides a general purpose library for accessing and manipulating PICO-8 cart data. You can add the picotool directory to your PYTHONPATH environment variable (or append sys.path in code), or just copy the pico8 module to the directory that contains your code.

The easiest way to load a cart from a file is with the from_file() function, in the pico8.game.file module:

#!/usr/bin/env python3

from pico8.game import file

g = file.from_file('mycart.p8.png')
print('Tokens: {}'.format(g.lua.get_token_count()))

Aspects of the game are accessible as attributes of the Game object:

• lua
• gfx
• gff
• map
• sfx
• music

Lua code is treated as bytestrings throughout the API. This is because PICO-8 uses a custom text encoding equivalent to lower ASCII plus arbitrary high characters for the glyph set. Take care to use b'bytestring literals' when creating or comparing values.

API under construction

While the library in its current state is featureful enough for building simple tools, it is not yet ready to promise backwards compatibility in future releases. Feel free to mess with it, but please be patient if I change things.

Important: The parser in particular is kinda jank and I may completely redo the API for the AST before v1.0.

Developing picotool

To start developing picotool, clone the repo, set up a Python virtual environment, and install p8tool in editable mode:

git clone git@github.com:dansanderson/picotool.git
cd picotool
python3 -m venv pyenv
source pyenv/bin/activate

# Install picotool in the virtual environment, with symlinks to source:
pip install -e .

# Install development tools:
pip install --upgrade pip
pip install -r pydevtools.txt

With everything installed and the virtual environment active, you can run some useful commands:

# To run the development version of p8tool:
p8tool

# To open an ipython REPL with picotool on the load path:
ipython
# To enable module auto-reloading:
#   %autoreload 2
# To import a pico8 module:
#   from pico8.lua import lua
# Autocompletion works with the tab key:
#   lua.unicode<TAB>

# To run all tests:
pytest

# To run specific tests:
pytest tests/pico8/lua/lua_tests.py

# To run the tests and calculate a coverage report:
coverage run pytest

# To display the coverage report on the console:
coverage report -m

# To produce an interactive HTML document based on the coverage report, as ./htmlcov/index.html:
coverage html

See pytest documentation, coverage documentation.

Visual Studio Code set-up

I use Visual Studio Code for editing with its powerful Python extension. When you open the picotool root folder in VSCode, it finds the active Python virtual environment and use ./pyenv/bin/python as its Python interpreter. If it doesn't, click "Python" in the left of the bottom bar, then select this path for the interpreter.

When you open a new Terminal within VSCode (press Ctrl + backtick), it automatically activates the Python environment. If it doesn't, or if you prefer a standalone terminal app, remember to activate the environment at the beginning of each session:

source picotool/pyenv/bin/activate

Building Sphinx documentation

The picotool documentation lives in docs/ as a Sphinx project. To build the documentation as web pages:

cd docs
make html

The generated HTML pages are in docs/_build/html/. Open the index.html file in your browser to use them.

You are welcome and encouraged to contribute pull requests to the Sphinx documentation. Once accepted, I'll take care of publishing the changes to the documentation site.

(For my own reference: The documentation site uses the gh-pages branch of the Github repo as its source. There is a crude make publish target that builds the HTML and pushes the built files to this branch.)

Known issues

For a complete list of known issues, see the issues list in the Github project:

[https://github.com/dansanderson/picotool/issues]

picotool is a hobby project and I have to ration the time I spend on issues. I welcome pull requests, and preemptively apologize if I don't review them in a timely manner.

I try to triage issues such that picotool's main features work with all cartridges made with the '''latest knwon version''' of PICO-8. During the PICO-8 beta, picotool has lagged behind on new Lua syntax shortcuts and built-in keywords, though this should settle down as PICO-8 reaches version 1.0. So far, PICO-8 has been largely backwards compatible, but I cannot promise continued support for carts made with old beta versions (0.x) of PICO-8.

picotool and PICO-8 count tokens in slightly different ways, resulting in different counts. More refinement is needed so that picotool matches PICO-8. As far as I can tell, with picotool making some concessions to match PICO-8 in known cases, PICO-8's counts are consistently higher. In most cases, the difference is only by a few tokens, even for large carts. If you find specific cases of token count mismatches, please file them as issues with reproducable examples.