[bug] `repo:` search filter is returning false positives

[stwist-dd]

Describe the bug

Seeing false positives when using the repo: search filter

To reproduce

I'm not sure how to provide a simple repro for this, as it depends on our repositories. We have hundreds of repositories, organized into different GitLab groups, e.g. digitaldomain/domains/lighting/foo.git and digitaldomain/domains/animation/bar.git

When I try to filter based on certain paths, I'm getting false positives. It's not consistent, some searches work, but others do not.

E.g. repo:domains/lighting file:manifest.yaml should (IMO) only show manifest.yaml from repositories inside digitaldomain/domains/lighting. It includes those repositories, but I'm also getting two false hits that don't in any way match: digitaldomain/domains/core/coresoftware/dpa_web_console and a repo in digitaldomain/domains/core/pipeline (I can't give the repo name, as it's confidential, but it doesn't contain "domains" or "lighting" in the repo name).

I'm only getting those two false positives, not the hundreds of other repositiories under /digitaldomain.

If I use repo:lighting file:manifest.yaml, I get all the repositories under digitaldomain/domains/lighting (good, expected), any repository with "lighting" in its repo name (good, expected), but I also get the same two false positives as before.

If I use repo:digitaldomain/domains/lighting, I only get repositories under digitaldomain/domains/lighting, and do not get the two false positives.

I also get false positives with repo:domains/anim (including repositories that aren't even under digitaldomain/domains, but are under other top-level groups (i.e. digitaldomain/foo) and repo:domains/layout.

Not every search term is giving me false positives, many work as expected.

Sourcebot deployment information

v4.10.29 (081dc67)

Additional information

No response

[stwist-dd]

In addition, if I search for e.g. repo:domains/lighting -repo:dpa_web_console file:manifest.yaml (in an attempt to filter out the false positive), the false positive remains.

Searching for e.g. repo:domains/lighting -repo:bar file:manifest.yaml does filter out all repos that contain "bar" in the name (but not the false positives).

[brendan-kellam]

Hey @stwist-dd - sorry for the late reply. I tried to replicate your error by creating matching subgroups (digitialdomain/domains/lighting and digitaldomain/domains/core) and ran the same query (repo:domains/lighting file:manifest.yaml), but was unable to repro unfortunately. Are you still running into this?

Going to close this for now but please feel free to reopen.

[stwist-dd]

Hey @brendan-kellam, thanks for looking into this.

Unfortunately I am still able to reproduce this on my end. We have many of repositories under both those groups, and only a few that show up incorrectly, so I suspect there might be something else at play here beyond just the repository groups and their names. But without knowing what... this is going to be very hard to reproduce.

I think there's a person on our end that has been in contact with someone at sourcebot (in general). I'll reach out to them, see if there's some way we can get you access to our instance to help debug.

On the plus side, I am able to accurately filter by using the full group path (e.g. repo:digitaldomain/domains/lighting), so this isn't a blocker. Just a mild inconvenience.

I don't have permission to re-open this. I'm not sure if you want to re-open it or not, given we can't reproduce the issue at this time.

[brendan-kellam]

I think there's a person on our end that has been in contact with someone at sourcebot (in general). I'll reach out to them, see if there's some way we can get you access to our instance to help debug.

Yea that would be great - you could also send me a email at brendan [at] sourcebot [dot] dev and we can setup sometime to chat on a call. Would be helpful to see the issue reproduced live

[brendan-kellam]

@stwist-dd following up on this

[brendan-kellam]

@stwist-dd closing for now as per our bug policy (#715). I think it would be most helpful if we could jump on a call to debug, so please shoot me a email at brendan@sourcebot.dev