Optimise the Inline Requires plugin

[marcins]

↪️ Pull Request

In our large application build, the Inline Requires plugin adds about 25% to the build time (6 min) - so we have looked at some quick optimisation wins.

While the ideal solution for inline requires would be to implement it in the ScopeHoistingPackager, a good time to do that would be when we move the packager to native code rather than working on a complex change to it now.

With a combination of tracing, profiling, and eyeballing (thanks @yamadapc!) we identified a few small performance improvements to the current implementation

• remove potentially expensive object allocation for comparing argument values and replace with a verbose if (this saves some GC overhead and provides a 10-15% perf improvement in our tests)
• remove unnecessary usage of an array from an earlier iteration where a boolean would suffice (this saves GC overhead and reduces time spent in the visitor)
• remove verbose logging - even when not logging, there was still significant overhead from this due to the volume (another 10-15%)

The main contributor now to this plugin's run time is the SWC print of the AST back to JS. We have considered a couple of other potential optimisations (that really still just avoid moving the code into the pacakger where it should be...):

• replace the SWC code with regex matching and replacement - not sure if this would be faster without trying it, but it would complicate source maps for sure.
• move the JS code to Rust - this will avoid serialising/deserialising the AST that SWC has to do with the parse/print pattern.
• implement minification support in the plugin - this means that we can skip running the SWC minifier, and just have this plugin produce the minified output - saving another print/parse that plugin has to do.

🚨 Test instructions

Verified application still works (unit tests for the visitor pass)

✔️ PR Todo

• Added/updated unit tests for this change
• Filled out test instructions (In case there aren't any unit tests)
• Included links to related issues/PRs

[marcins]

Thought of another idea while folding laundry, but not sure if it'd make a difference - there's no need to maintain a Map of public asset ids -> side effects + filePath. The filePath is only used as a convenience for logging, so this could easily be a Set of only public asset ids that have side effects. This turns a map.has + map.get to work out if the asset has side-effects, into just a set.has.

parcel/packages/optimizers/inline-requires/src/InlineRequires.js

Line 22 in db2984d

publicIdToAssetSideEffects = new Map<string, SideEffectsMap>();

[marcins]

I want to cut another dev release with the asset map -> set change and test it before merging this.. currently blocked by the OpenSSL cross-compile issue though, and I don't want to "revert" the Sentry PR in this branch again just to test.

Just waiting for a dev release to build now, so I can test this change in CI.