Cache SyntaxSet between parse_codeblock calls

[mgeisler]

Before we would load and deserialize the syntect syntax trees (Sublime Text syntax definitions) on every call. We now cache this read-only data.

Before, mdbook build took 1.44 s, now it takes 0.56 s instead. I started looking in to optimizing this because I'm considering making mdbook-xgettext an (optional) plugin in book.toml for Comprehensive Rust. This will imply that we run the plugin on every mdbook build — at least for people like me who have the plugin installed.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ec5968c) 90.64% compared to head (62c36d2) 90.64%.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #147   +/-   ##
=======================================
  Coverage   90.64%   90.64%           
=======================================
  Files          11       11           
  Lines        2372     2373    +1     
  Branches     2372     2373    +1     
=======================================
+ Hits         2150     2151    +1     
  Misses        159      159           
  Partials       63       63           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kdarkhan]

Just a thought - maybe we should add several bench targets so that we can find how performance of the code changes. Does not have to be exposed during PR build but something that is easy to trigger locally.

[mgeisler]

Just a thought - maybe we should add several bench targets so that we can find how performance of the code changes. Does not have to be exposed during PR build but something that is easy to trigger locally.

That would indeed be a useful tool to have!

For this, I used https://github.com/sharkdp/hyperfine like this:

% MDBOOK_OUTPUT='{"xgettext": {"pot-file": "messages-10.pot", "granularity": 10}}' hyperfine 'mdbook build -d po'

to test only mdbook-xgettext. But if we could put this into a criterion benchmark as well, then that would be nice.