Uncaught lstat EPERM errors—potential error handling bug in node_watcher about sane

[marcello3d]

Something like this: https://github.com/amasad/sane/compare/master...marcello3d:walker-errors?expand=1

[wtgtybhertgeghgtwtg]

Sorry for the late response. Do your users have examples of where this error is occurring?

[joshkel]

I've run into this too. When using Facebook's Jest in watch mode, I'll sporadically get errors similar to the following:

events.js:163
      throw er; // Unhandled 'error' event
      ^

Error: EPERM: operation not permitted, lstat 'F:\trunk\app\test\containers\RunNumberHandlerTest.js~RF236e5f90.TMP'

My diagnosis was the same as @marcello3d's (node_watcher's recReaddir -> 'error' event -> no error listener).

From what I can tell, the culprit in my case is an editor creating temporary files with ShareMode 0 (none). (See MSDN's CreateFile docs.) Node then attempts its own CreateFile call as part of its lstat implementation, and that fails.

I can reproduce the error by creating a file, changing the file permissions to deny access to everyone, then running Jest.

[amasad]

@joshkel || @marcello3d patches welcome :)

[joshkel]

@amasad - I'd be happy to submit a patch, but I'm not using sane directly and am really not familiar with its design.

Should it ignore errors completely, under the assumption that callers don't assume they'll have to handle errors and that there aren't any actionable events on inaccessible or erroneous files?

Should it try to special case a couple of errors for walker, like it does for lstat?

Something else?

[marcello3d]

How does my patch from up above look?

[amasad]

@marcello3d the patch looks good. But I don't understand why should we also ignore "ENOENT"? EPERM seems to be the problem here

[marcello3d]

I refactored the existing code into an new helper function, see https://github.com/amasad/sane/compare/master...marcello3d:walker-errors?expand=1#diff-18c7836cd56d6aee3a621e821c9c121fL223

As to why the existing code did that, I couldn't tell you. :-)