fs: document why isPerformingIO is required

[ronag]

Adds some notes to motivate why isPerformingIO is required for any future developers unfamiliar with it.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

[ronag]

@addaleax: PTAL to ensure the description is correct.

[trivikr]

LGTM, minor nits

[ronag]

@addaleax: Would it make sense to add guards for this one level lower, i.e. in fs.read/write and fs.close? I'm a little worried that when users use these methods directly they don't take this edge case into consideration (just like we missed it in fs streams).

[addaleax]

@ronag Yes, but how? We can’t assign those properties to the fd itself, even though that would make the most sense here… it might be worth for FileHandle, though.

[ronag]

We can’t assign those properties to the fd itself

We could have a Map<FD, state>? With some performance overhead...

Would it be possible to use FileHandle for the callback version as well?

[addaleax]

@ronag Yes… I’m just worried we’ll run into problems – what if the user passes the fd to an addon? We can’t know what happens with that fd in that case. I’m all for it, though – I’ve thought about doing this for Worker threads too, at least optionally, because they should really close all fds that they have created during their lifetime.

[ronag]

We can’t know what happens with that fd in that case.

Is it maybe possible to fix this whole thing in libuv? It kind of feels like a "bug" to me...

[addaleax]

@ronag I don’t think we could do it any better than in Node.js, i.e. it would have to be on a best-effort basis.

[ronag]

Would moving to e.g. io_uring on at least linux resolve this?

[ronag]

what if the user passes the fd to an addon?

Could napi have guards?

[addaleax]

Would moving to e.g. io_uring on at least linux resolve this?

Not that I know of – they still pass fds as raw numbers?

what if the user passes the fd to an addon?

Could napi have guards?

No, we can’t tell fds apart from other integers, unfortunately.

(These are all reasons why FileHandle is a lot nicer than the standard fs API.)

[ronag]

Not that I know of – they still pass fds as raw numbers?

Yes, I think ioring would avoid the need for threads and sync api's... i.e. removing the underlying issue.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/31996/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/31999/

[ronag]

Landed in 51a2df4

[MylesBorins]

This is not landing cleanly on v14.x. @ronag should we backport?

[ronag]

@MylesBorins: No I think we can skip this. It's mostly for future reference in development.