Discussion: Regression and Process Improvements (Ref: PR #389)

[sinhasubham]

Hi team,

I'm opening this issue to discuss a recent regression that impacted clients (like google-cloud-spanner) and to explore how we can improve our processes to prevent similar breaking changes in the future.

Context
The Change: PR #389 ("Cleanup the resource detector code a bit") was merged, which refactored the GoogleCloudResourceDetector and deprecated the old one.

The Bug: This refactor appears to have introduced a TypeError. The new resource detector's code was attempting to access attributes on an OpenTelemetry Resource object using dictionary-style access (e.g., resource['key']) instead of the correct attribute-style access (e.g., resource.attributes['key']).

The Impact: This regression introduced a runtime TypeError that broke downstream clients. For example, here are the google-cloud-spanner library reported failures: https://github.com/googleapis/python-spanner/issues.

The Fix: A subsequent PR was merged (fix PR) that correctly changed the access pattern back to resource.attributes['key'], resolving the TypeError.

[aabmass]

@DylanRussell

[DylanRussell]

Neither the old or new resource detector supports that dictionary bracket kind of access.. maybe the new detector is correctly typed and the old one wasn't and it correctly uncovered a bug in the code? I'm confused

[DylanRussell]

Both old and new code were correctly annotated as returning a Resource though

[aabmass]

The Impact: This regression introduced a runtime TypeError that broke downstream clients. For example, here are the google-cloud-spanner library reported failures: https://github.com/googleapis/python-spanner/issues.

@sinhasubham can you link an actual issue? Looking at your PR I'm also confused why the change was necessary