[Wesley Wigham (weswigham)]

RWC has been failing for awhile (2 weeks); it wasn't really noticeable, since we still got normal diff PRs and merged them as needed, which seemed fine, but today I noticed that it was still failing even immediately after I merged an update PR - finding that suspect, I checked the logs and found out that pretty much every test was giving a spurious error saying it was missing every file in the test. Definitely not right. The story ended up being that #46086 made our readDirectory return absolute paths, while the rwc harness was very much relying on relative paths to perform a simple path comparison.

[Andrew Branch (andrewbranch)]

Hm, do we have an opinion on what the correct thing for readDirectory to return is? There seemed to be some internal inconsistency that led to a bug, but I didn't fully realize that fixing the bug also resulted in such an observable API change.

[Wesley Wigham (weswigham)]

Hm, do we have an opinion on what the correct thing for readDirectory to return is?

I'd defer to Ron Buckton (@rbuckton) personally, since he probably knows what the intent was, since it was written to support file globbing, iirc.

[Andrew Casey (amcasey)]

For the record, I'm totally cool with rolling back the original change (#44710) - ideally just this part, but the whole thing if that's easier. It was a minor quality-of-life improvement (can enable exceptions during debugging with minimal noise).

[Andrew Casey (amcasey)]

Also, I'd expect to get names and not paths from readDirectory - I'm pretty sure that's what fs.readdir returns.

[Ron Buckton (rbuckton)]

Also, I'd expect to get names and not paths from readDirectory - I'm pretty sure that's what fs.readdir returns.

I mentioned this in #46788, but readDirectory wasn't intended to emulate fs.readdir. When globs were added, one of the requirements was that readDirectory return all of the matched results in one API call. This was because VSTS wanted to be able to swap out a call to readDirectory with one performed in-process to avoid the overhead of multiple round-trips from .NET to JavaScript using recursive calls that only returned file names.