Implementation of the LaTeX3 Bundle in Tralics

Introduction

This document explains the implementation of the LaTeX3 expl3.sty package. This consists in an extension of the C++ code, a style file (using normal latex syntax) and a big tex file (using latex3 syntax) which is expl3-code.tex minus the C++ code.

A partial implementation is available in Tralics version 2.15.4.

The most evident feature of LaTeX3 is that command names contain underscore and colon characters (no atsign) and are much more explicit. For instance \@ne and \thr@@ are LaTeX constants, with value one and three; in LaTeX3, they are named \c_one and \c_three. instead of \@tempboxa and \@tempcnta there will be \l_tmpa_box and \g_tmpa_int; the first one should be used for local assignment, the second one for global assignment. A variable of the form \l_module_foo_tl is local, defined in the module module, named foo and contains a token list, while \l__module_foo_tl is a private variable of that module. In the case of the kernel, the module part is omitted from the name.

The two commands \module_foo:nnn and \__module_foo:nnn are macro names, respectively public and private, defined in the module; they take three arguments (there are three letters after the colon). The meaning of these letters is the following

None. A command whose name ends with a colon takes no argument. Example: \prg_do_nothing: (corresponds to \@empty, is a command whose expansion is void, while \scan_stop: is the sames as \relax.
n: Unexpanded token or braced token list. Example \use:nn. This is a command that takes two arguments, whose expansion is formed of these arguments. For instance \use:nn {a b}{{c d}} expands to a b{c d} (one pair of braces disappears), while \use:nn a b expands to ab (an argument can be preceded by an optional space).
N: Single token (unlike n, the argument must not be surrounded by braces). Example \cs_if_eq:NNTF takes for arguments, compares the first two, and expands to the third if the test is true, to the fourth otherwise. The test is done by \if_meaning:w (aka \ifx) that compares two tokens. It is very complicated to grab a single token in TeX, so, generally, all that matters is that the arguments consists in a single token; so \cs_if_eq:NNTF {a}{a} {ok}{\bad} expands to ok. If the first argument is {ab}, LaTeX compares a and b, and the test is false; if the first argument is empty, you could be in trouble. An error is signaled in such cases by Tralics. On the other hand \cs_gset_eq:NN behaves like \global\let. So \cs_set_eq:NN {=} tries to redefine the opening brace, which is an error (unless the opening brace is an active character, an unusual situation).
T, F: These are special cases of n arguments. Example: \cs_if_eq:NNTF, \cs_if_eq:NNT, \cs_if_eq:NNF and \cs_if_eq_p:NN, are four commands that process the first two arguments in the same way. The command that takes only two parameters expand to true or false (this will be explained later on), the first one (with four arguments) is explained aove; the other one take three argument; the expansion is this argument (if the truth value matches the extension) and is empty otherwise. So both \cs_if_eq:NNT aa {ok} and \cs_if_eq:NNF ab {ok} expand to ok. The letters T and F are only used in this case.
p: Primitive TeX parameter specification. This is used in command that define other commands whose arguments are not specified by a suffix. Example: \cs_set_nopar:Npn is \def and \cs_set:Npn behaves like \long\def.
w: This means that the argument syntaxis weird in that it does not follow any standard rule. Example \if_meaning:w cited above; this command reads two argument and compares them. It this executes the true or false code and skips the other one; this is the weird part of the syntax. LaTeX3 never defines commands with such a syntax: see T and F above.
D: This means Do not use. Example \tex_ifx:D and \tex_let:D; these are alternate names for \ifx and \let; the idea is to always use macros with a colon in their name in the package files. The first macro is also known as \if_meaning:w, and the second is like \cs_set_eq:NN (the difference is that you can say \let\foo= 3 where the equals sign and the space that follows it are optional, and \cs_set_eq:NN~ 3 makes the active character tilde behave like a space and typesets 3.
Variants. For instance \cs_set_eq:cN {foo}\bar is equivalent to \exp_after:wN \cs_set_eq:cN \cs:w foo\cs_end:\bar or, with TeX syntax: \expandafter \let \csname foo\endcsname= \bar. In general a variant calls the base function (here \cs_set_eq:NN, unless optimiseed for speed.
c variant. As explained above, this means that the argument is processed by \csname (so is a command name). Example \use:c is the same as \@nameuse it reads a sequence of tokens, fully expands them, converts the resulting characters into a control sequence name (which is the result of the expansion). If the expansion contains \par, an error is signaled; note that \@nameuse signals an error is \par is part of the argument. The command \cs_if_exist:cTF is like \cs_if_exist:NTF but uses \if_cs_exist:w rather than \if_cs_exist:N (i.e., \ifcsname or \ifdefined); the idea is to avoid filling the hash table with an undefined token.
x variant. Fully-expanded token or braced token list. Example \iow_log:x, which is the same as \wlog; it prints its argument in the transcript file; the expansion is done by TeX. Other example \cs_set_nopar:Npx (aka \edef) is the base function for doing full expansion. Now \use:x{foo} is the same as \edef\tmp{foo}\tmp. For this reason, the command is not expandable, and x-type arguments cannot be nested. Note that \str_if_eq_x:nnTF is similar to \str_if_eq:nnTF. The idea here is that the pdftex function that compares two string fully expands them; and you have to use \unexpanded in order to avoid this expansion.