Our CI runs with

pants test --changed-since=origin/main

now I observed that one PR changed a file name that didn’t trigger a test that depended on that file. Has this been observed before? (I searched but couldn’t find a matching issue)

Isn't

origin/main

the current commit when your CI runs? I use HEAD^ for CI to compare against previous commit (we always squash-merge).

I meant the CI runs inside the branch associated with the PR. Not CI runs on the main branch itself.

Ah, makes sense. Are you doing shallow checkouts or something like that so it doesn't fetch the whole git history? This is quite common on GitHub actions for example. I think Buildkite might also do that in some configurations.

Hi @gentle-gigabyte-52115, can you provide a small github repo that demonstrates the problem? Many people, myself included, use

--changed-since

frequently, and I haven't noticed this issue. So maybe this is a git diffing issue of some kind as @gorgeous-winter-99296 suggests?

we use

fetch-depth: 0

in actions. I’ll create a minimal example tomorrow

Maybe the

--changed-dependents

(https://www.pantsbuild.org/docs/reference-changed#dependents) would help here?

Ah yes, of course 🤦‍♂️

--changed-since

only retrieves files that have changed.

--changed-dependents

also retrieves all files that transitively depend on files that have changed. So based on your use case it sounds like you need that.

Here is a minimal example https://github.com/JoostvDoorn/pants-changed-since-reproduction See the README for the steps to reproduce. I know this can be caught by turning

unowned_dependency_behavior

to error, but we currently use the default value.