I have one question: the parser in the article uses computed goto, which, as far as I know, is not standard C. Are there many cases where even the standard goto yields significantly better performance?
Regarding computed and simple goto I'd like to reference our early article discussing the parser in standard goto https://natsys-lab.blogspot.com/2014/11/the-fast-finite-stat... . My recent talk at scale https://www.socallinuxexpo.org/scale/17x/presentations/fast-... (watch videohttps://www.youtube.com/watch?v=LQc4er8ng64&feature=youtu.be... and slides at http://www.tempesta-tech.com/research/http_str.pdf ) discusses the parser with the compiler extensions.
We also found hat Ragel parser generator (using the goto state machine) generated somewhat faster large state machines than Bison (switch-drived state machine). However, as pointed out in the SCALE talk goto works better only on big enough state machines.
It is very unlikely that the computed gotos make the parser faster than using standard features of a more powerful language.
Can you explain more on how? Use case I have in mind is bytecode interpretation where computed goto should be a big help.
https://github.com/ncm/computed-goto
The "more powerful language" part is to get language-level tools to compose parser combinators, ultimately resolving down to tail calls.