[makwana-akshat]

Fixes #160976.

Summary

The Gecko sampling profiler exporter used only the basename of Python source filenames, losing directory information when different files shared the same basename.

This PR:

• Preserves the full source filename path in the Gecko function table.
• Adds a regression test verifying that files with identical basenames in different directories retain distinct paths.
• Adds a NEWS entry documenting the fix.

• Issue: profiling.sampling (Tachyon) gecko output: Only source file basename is shown #160976

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[maurycy]

cc @ivonastojanovic ?

[ivonastojanovic]

Thanks for the fix! I’ve left a couple of nits.

One other thing I was wondering about is the resource name. _get_or_create_resource still uses the basename, so both files would show up as Resource: __init__.py in the profiler. Since Gecko resources are meant to group functions, for example by library or origin, would it make more sense to use the module name for the resource instead?

cc @pablogsal

[ivonastojanovic]

I think we could simplify this test quite a bit. Some of the assertions seem redundant, like checking the index types/ranges and that both the indices and strings are different. I think we could keep just the assertions that verify the behavior we actually care about here.

[ivonastojanovic]

nit: Maybe a bit shorter, e.g. “The Gecko reporter (--gecko) in profiling.sampling now records full source file paths instead of only basenames.”

[makwana-akshat]

Thanks for the review! I've simplified the regression test to focus on resolving each function's filename to the correct full source path and ensuring the two files are distinguished. I've also shortened the NEWS entry.

Regarding _get_or_create_resource(), I’ll keep that behavior unchanged in this PR so the fix stays focused on preserving source file paths in the function table. I agree that resource naming is worth considering separately.