Updates spk's FileMatcher to lazily create its internal gitignore object

[dcookspi]

This updates spk's FileMatcher objects to lazily create its internal gitignore object on first call to the matches() method.

The FileMatcher and its gitignore field was accounting for about a quarter of the load/read time for a package build spec. The matching is only carried out when building a package, so setting it up lazily saves time in other parts of spk (e.g. a solve that reads 15k package build specs).

Todo:

• Add check for empty rules before creating a gitignore
• Include result to avoid repeated recreation failures
• Profile vs ArcSwap

[codecov]

Codecov Report

❌ Patch coverage is 63.63636% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ema/crates/foundation/src/spec_ops/file_matcher.rs	62.50%	12 Missing ⚠️

📢 Thoughts on this report? Let us know!

[jrray]

Suggested change

	gitignore: Arc<Mutex<Option<ignore::gitignore::Gitignore>>>,
	gitignore: Arc<Mutex<Option<Result<ignore::gitignore::Gitignore>>>>,

I suggest making it distinguishable between an uninitialized value and a value that failed to initialize, to avoid repeatedly trying to initialize something that will fail.

[dcookspi]

Updated to include this and store the fail error.

[jrray]

Profile first, but ArcSwap may be a cheaper way to manage this field. Instead of creating numerous Mutex instances. I'm more concerned about this significantly slowing down builds for packages with a lot of files and calling matches on each file.

[rydrman]

maybe a OnceCell to leverage get_or_try_init? Maybe that's been tried, but leaving the note anyway

[jrray]

Before acquiring the lock you could check if self.rules is empty, if so then this could early exit. I'm assuming an empty Gitignore returns false for every input. This may alleviate some of the concerns about the Mutex overhead.

[dcookspi]

Added a check for a FileMatcher with no rules.