src: make FSEventWrap/StatWatcher::Start more robust

[TimothyGu]

Fixes: #17430

Checklist

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

Affected core subsystem(s)

fs, src

[TimothyGu]

CI: https://ci.nodejs.org/job/node-test-pull-request/11856/

[bnoordhuis]

For consistency with the surrounding code, maybe write this as return args.GetReturnValue().Set(0); without braces.

[bnoordhuis]

Doesn't the linter complain about using two spaces before //?

[TimothyGu]

No it doesn't, though fixed anyway.

[bnoordhuis]

This is wrong when Start() is called with different arguments than last time. This should work:

uv_fs_poll_stop(wrap->watcher_);
uv_fs_poll_start(wrap->watcher_, Callback, *path, interval);
// Safe, uv_ref/uv_unref are idempotent.if (persistent)
uv_ref(reinterpret_cast<uv_handle_t*>(wrap->watcher_));
elseuv_unref(reinterpret_cast<uv_handle_t*>(wrap->watcher_));

And while we're here, this code really ought to check the return value of uv_fs_poll_start() (and uv_fs_poll_stop() too, although it can't fail in the current implementation.)

[TimothyGu]

Start() is not part of the external API, and no guarantees were made whether it works with different arguments or not. On the other hand, the proposed solution breaks compatibility.

[bnoordhuis]

Then why don't you make it throw an exception? Methods should not silently do the wrong thing, that's arguably worse than aborting.

On the other hand, the proposed solution breaks compatibility.

In what way?

[TimothyGu]

The added test currently throws no exception. If I make it throw an exception it would be breaking compatibility.

[bnoordhuis]

What do you mean? There are no compatibility issues because it was aborting before.

[TimothyGu]

It was aborting in FSEventWrap, not StatWatcher. This was simply tacked on as a similar change that improves the current behavior in a similar way.

[bnoordhuis]

I wouldn't use the word 'improves'. StatWatcher already worked mostly like I proposed in my first comment and that would be an acceptable change for FSEventWrap too, as would throwing an exception. Silently ignoring changes is the worst possible solution however.

[BridgeAR]

CI https://ci.nodejs.org/job/node-test-pull-request/12069/

[addaleax]

Landed in cd71fc1

[bnoordhuis]

#17432 (comment) is still unfixed.

[MylesBorins]

should this be backported to v6.x or 8.x... my gut is no but not 100%

[MylesBorins]

ping re: backport

[MylesBorins]

ping re: backport

edit: landed cleanly on 8.x. Any reason not to land?

[TimothyGu]

@MylesBorins No, not really.