Function texlang_stdlib::expansion::get_expandafter_optimized

pub fn get_expandafter_optimized<S: TexlangState>() -> BuiltIn<S>

Get the optimized \expandafter command.

This is a more complex implementation of \expandafter that is optimized for handling repeated \expandafter tokens. It contains two optimizations, as described below. These optimizations where developed “theoretically” and with no benchmarking, so it remains to be seen if they actually make expansion faster. For this reason both the optimized and simple \expandafter implementations are maintained.

We now describe the two optimizations.

First, \expandafter control sequences are often linked together in the following format:

\expandafter<token 1>\expandafter<token 2>...\expandafter<token n><token n+1>

Here, to expand the first \expandafter we just need to expand <token n+1>. In the original implementation of TeX this works via recursion: the ith \expandafter requests expansion of the second-to-next token, which is the (i+1)th \expandafter. After n recursions, the last token is finally expanded. In the optimized implementation here, the token stream is scanned ahead for as long as the pattern \expandafter<token i> repeats. The token <token n+1> is expanded and the intermediate \expandafter tokens are dropped from the input. This is still an O(n) operation, but results in only 1 Rust function stack being used, rather than n.

Second, \expandafter commands are often grouped together like this:

\expandafter\expandafter\expandafter\A\expandafter\B\C

This TeX code causes \C to be expanded first, then \B\ and finally \A. When the leading \expandafter is expanded, the first optimization kicks in and \C will be expanded, leaving:

\expandafter\A\B\Cexpanded

The second optimization is that the leading \expandafter that is left over will also be expanded without yielding control to the main expansion loop. If, after this pass, the leading token is again an \expandafter token, it will be expanded too. This process continues repeatedly until no \expandafter tokens are left at the start of the token stream.