fs: avoid emitting error EBADF for double close

[mcollina]

Changed the logic in fs.ReadStream and fs.WriteStream so that
close always calls the prototype method rather than the internal
event listener.

Fixes: #2950

cc @evanlucas @Trott @bnoordhuis

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

Affected core subsystem(s)

fs, stream

[jasnell]

Given that this affects error handling it technically should be considered for semver-major. However, I think it qualifies as a bug fix so leaving it as semver-patch.

[cjihrig]

Can you remove the / before rw.

[mcollina]

fixed

[mcollina]

@jasnell I added the don't land on LTS labels, just in case.

[mscdex]

Minor nit: there's a typo in the commit message

[mscdex]

Another reason not to land on v4.x and v6.x is the use of .bind(), which won't be very fast in those versions.

[evanlucas]

If we aren't planning on backporting this, I think we should go ahead and mark as semver-major (especially given that it seems like a breaking change)

[Fishrock123]

Another reason not to land on v4.x and v6.x is the use of .bind(), which won't be very fast in those versions.

We can just use an arrow function instead?

[mcollina]

Another reason not to land on v4.x and v6.x is the use of .bind(), which won't be very fast in those versions.

We can just use an arrow function instead?

If that's the only problem, yes of course.

[mcollina]

Landed as b1fc774

[Fishrock123]

@jasnell @mcollina Was this intentionally left as semver-major?

[mcollina]

@Fishrock123 yes. I'm happy for it to be backported, but we were not in agreement. If we agree that this is not semver-major, we can backport. I can fire a PR.

cc @jasnell @evanlucas @mscdex @cjihrig

[cjihrig]

I don't know if this is pressing enough that it needs to be backported, but I don't feel strongly either way.

[mscdex]

@mcollina Now that I look at this again, I think we could have avoided the bind() with just:

this.once('open', this.close);

which is more easily backportable. Maybe submit a separate PR for the bind() removal and then backport this and that PR at the same time?

[mcollina]

@mscdex we cannot, because open will be emitted with the fd as the first argument. If we want to change this, we need to change the check on cb, so that it ignores anything that is not a function. As calling close()  when a stream is not open yet is likely not a hot path, I don't think it's worth the added complexity.

[mscdex]

@mcollina Ok well we could still pull out a reusable function and do it that way:

this.once('open', closeOnOpen);

// ...

function closeOnOpen() {
  this.close(null);
}

[mcollina]

here it is: #13474