Sharing package.json and stat cache between CJS and ESM loaders

[guybedford]

Currently the ES module resolver and CJS resolver do not share their stat cache or package configuration cache.

Sharing these caches would significantly improve boot performance of applications that contain mixed CommonJS and ESM (currently very few, but soon to be most, we hope!)

I wanted to create this issue to track that optimization work.

[juanarbol]

Could you explainme a bit better how cache thing works? I've read something on cjs loader, but no understood very well.

[shackijj]

Currently the ES module resolver and CJS resolver do not share their stat cache or package configuration cache.

@guybedford Should the cache sharing be implemented in a scope of this issue? If yes, I'd be glad to help.

[guybedford]

@shackijj help would be amazing here. The work has already been simplified by the fact that both resolvers are now in JS wheras previously the ESM resolver was C++.

Should the cache sharing be implemented in a scope of this issue?

Can you clarify the question here a bit more?

[shackijj]

@guybedford
In the original message you wrote:

I wanted to create this issue to track that optimization work.

So, I wanted to clarify if we need to make the optimization (write some code) or this issue was created for tracking purpose only (e.g. after a certain release we should check that performance got better)

I hope this makes sense.

[shackijj]

Spent some time on researching the code base and thinking of the implementation. The diagram below represents the way we can make it. @guybedford could you please share thoughts about it?
If it's OK, please assign the issue to me and I will work on that.

[guybedford]

@shackijj there's no other optimizations or blockers here, this issue is directly for this work of joining the caches.

Glad to hear you managed to take a look at the codebase already. That looks like a great summary. I would suggest tackling the package config cache and stat cache separately either as separate PRs or at least as separate commits. This will also help me for review in keeping the diffs straightforward.

Joining the package cache should be as simple as calling the package cache function https://github.com/nodejs/node/blob/master/lib/internal/modules/cjs/loader.js#L279 from the ESM loader. Note it is a slightly different API to what the ESM loader users, but refactoring the APIs to be similar is an important simplification as part of this work.

As we generalize these routines between both loaders we may also want to move them out to a shared resolution file as well in the process. Focusing on keeping the diff down and ensuring simplifications in the refactoring would be great to see.

Let me know if I can help further at all here, otherwise would be great to see what you come up with.