ShabbirHasan1 / tex-autogolfer

TeX AutoGolfer

(Heavily Work-in-Progress)

Highly-golfed TeX programs are hard to maintain. This tool strives to:

• shrink identifiers to shorter identifiers, in some situations:
  • ✓ renaming stuff defined by \def or \let
  • ✓ Using \let to re-define builtin names
  • Using \catcode to make some bytes active
• ✓ handle transformation of e.g. \newcount\x, \x to \count1
• ✓ remove whitespace and comments

Example

Input (71 bytes):

\usegolf{rename}
\def\double#1{#1#1}
\def\triple#1{#1#1#1}
\double a
\triple c
\double b

Output (36 bytes):

\def~#1{#1#1}\def\!#1{#1#1#1}~a\!c~b

Configuration

Most golfing configuration is on the input files of the form of LaTeX-looking \usegolf commands. While \usegolf can be specified in any order, golfing transforms are always executed in the same order.

1. \usegolf{count}: convert \newcount usages to \count.

   • Write ⫽ after any counter control sequence that needs a space after it if converted to a \count
     • ⫽ is automatically inserted before any digit preceded by a control sequence
     • e.g. a macro to mod counter \d by a number literal can be written \def\m#1;{\u\d\divide\u⫽#1\multiply\u⫽#1\advance\d-\u} and used \m17;
   • Write ⦃ ⦄ around and counter control sequence that needs curly braces around it if convered to a \count
     • e.g. write \argv⦃\x⦄ so it can be transformed to \argv{\count1} instead of \argv\count1.
     • If there is no \usegolf{count}, then the curly braces are removed.

2. \usegolf{rebind<control-seq>}: really just a convenience

   • For example, \usegolf{rebind\def} replaces every \def with a \defRebind and puts a \let\defRebind\def to the program. This lets \usegolf{rename} below rename \def
   • \usegolf{count} automatically adds a \usegolf{rebind\count} at the top of the file
   • The \let is placed at the position of the \usegolf, so writing \usegolf{rebind\newcount~}\newcount\x\usegolf{rebind\advance~}\advance\x1 will work.

3. \usegolf{rename}: rename identifiers bound with \let, \def, and \newcount

   • The identifiers are renamed by default to identifiers like \$. These have the slight advantages (over \a) of less often needing spaces after them
   • The most-common identifier is renamed to ~, which is the only default active character that works everywhere. Make sure not to use ~ for a space: use ␣ instead.
   • (Not yet implemented) some way to allow form-feed and other single-byte active characters

4. \usegolf{rename<control-seq-1><control-seq-2>}: specify a particular rename to use

   • E.g. \usegolf{rename\def\D} renames \def to \D instead of whatever arbitrarily chosen identifier like \$.
   • Has no effect if \usegolf{rename} is not provided
     • (for forwards compatibility with a future decision of either enabling renaming everything, or just renaming that one thing)
   • Note that formfeed  (0x0C) is an active character, but only works in some contexts, so it's not automatically chosen. You can use it like \usegolf{rename\def}
   • Write \usegolf{rename-add\name} if you want to prioritize consideration of that control sequence. For example, if you change $ to be an active character (\catcode`$=13), then write \usegolf{rename-add$}
   • If you change % to be an active character, then write \usegolf{rename-add\char37} (since % is byte 37). Same applies to any other single byte, but if it's not a special character, you can also write the name directly.

5. If an operation such as \rebind would rename too much stuff, name the control sequences you don't want renamed with ↦. E.g. \def↦\def\f{hi} (instead of \def\f{hi}) if you rebind \def later on.

6. Not a golf option, but just an ease of input: write ⦃97⦄ to get an a, for example. Specify the codepoint in decimal (or in hex, with a 0x prefix, e.g. ⦃0x61⦄). This is in the pre-processer, before any golfs happen. In particular, ⦃12⦄ gives form feed.

There's one more option. If you write --preserve-newlines on the command line, then the parser will not delete newlines.

Disclaimer

This tool makes no guarantee about preserving program correctness. However, there are some ways you can help the tool ensure it doesn't change program behavior:

• Mark required spaces (for typography) as ␣. These convert to (or \ if the would be gobbled)
• Mark required spaces (for preventing e.g. a number from running on too long) as ⫽⫽. These convert to (or {} if the would be gobbled)
• Mark required spaces (for \catcode32=12) as …. These convert to unconditionally
• Messing with \catcodes is risky.
• Don't use multi-byte UTF-8 characters in places that would confuse the tool
  • TeX doesn't understand multi-byte characters (your code gets broken into bytes), but I decided to preserve them.
• Ensure each identifier is only for one purpose (i.e. isn't re-defined)
• More suggestions, such as \x⫽ and ⦃\x⦄ are listed above in the "Configuration" section.

XCompose

A sample .XCompose for the funny unicode.

<Multi_key> <space> <space>		          : "␣"
<Multi_key> <slash> <slash>		          : "⫽"
<Multi_key> <braceleft> <braceleft>	    : "⦃"
<Multi_key> <braceright> <braceright>	  : "⦄"
<Multi_key> <t> <o>                     : "↦"
<Multi_key> <slash> <period>            : "⏎"

Dev

Run npm install to install.

Command recipes:

npm run build --watch
npm run lint --watch
# CLI, only after a build
node ./dist/cli.js scratch/a.tex > scratch/b.tex
# View form-feeds in output, and add trailing newline
node ./dist/cli.js examples/ex6.tex | sed -e 's/\f/�/g'; echo