n-api: back up env before async work finalize

[gabrielschulhof]

We must back up the value of _env before calling the async work
complete callback, because the complete callback may delete the
instance in which _env is stored by calling napi_delete_async_work,
and because we need to use it after the complete callback has
completed.

Fixes: #20966

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[gabrielschulhof]

I'm not sure if it makes any sense to add a test like this one, but finite.

[nstepien]

Considering this fixes a regression, I'd say a test should be added.

[gabrielschulhof]

@MayhemYDG the problem is it's a memory corruption regression, so a 100% reliable test is hard to create. After all, when c072057 introduced the regression, the async work test which we already have should have segfaulted but didn't.

My only thought with the above comment was to add a second test which essentially runs the first test over and over and over, because that's what I used for reproducing. But you're right - I think I'll add a test like the one in the gist, with, say, 500 iterations. Hopefully in that many iterations it will crash if we ever again have such a bug.

[gabrielschulhof]

Well, the test I added crashes before the fix and passes after the fix, so FWIW, it captures the fix.

[mhdawson]

LGTM

[mhdawson]

LGTM

[gabrielschulhof]

CI: https://ci.nodejs.org/job/node-test-pull-request/15264/

[gabrielschulhof]

Another CI: https://ci.nodejs.org/job/node-test-pull-request/15268/

[gabrielschulhof]

sigh ... forgot to set the name of the async context - weird that it did not throw in debug mode O_o

Here's another CI: https://ci.nodejs.org/job/node-test-pull-request/15273/

[nstepien]

@gabrielschulhof
Nice. Do you think this could get merged in time for the v10.4.0 release?

[refack]

I'm +1 to land this after only 40h so it make it into 10.4.0, since the most significant stakeholders are aware of this.

[gabrielschulhof]

@refack OK, landing it then.

[MylesBorins]

landed in 991f406

[MylesBorins]

Should this be backported to v8.x-staging? If yes please follow the guide and raise a backport PR, if not let me know or add the dont-land-on label.

[gabrielschulhof]

@MylesBorins this actually fixes a regression introduced by c072057, so, unless we backport that commit too, we need not backport this commit. As it stands, it doesn't look like that commit was backported to v8.x.