A difference in KSP task

classpathStructure

input was found to be the main local-local cache miss in a project. Has anyone encountered this? Any ideas on how to find what the actual difference is between builds?

It's the only difference shown in the build scan comparison?

Hi! It's the "root" difference. This is from exp. 2 of build-validation-scripts (local-local, caching same location), so same commit on both builds, etc. Other than this, differences only on some custom values we log in build scans (GC time, things like that)

what's the version for Ksp and Kotlin, I will try to reproduce it locally

Thank you! We're on Kotlin 2.0.0 and KSP 2.0.0-1.0.22. I couldn't find anything of relevance in later release notes of either plugin.

I was not able to reproduce it with nowinandroid aligning versions. But that's something that I saw recently in other build scans. I continue the investigation. if I'm not wrong this hash represents the fingerprinting input. We will confirm with the team

Good to know I’m not alone 😅 This is a large project, so there was a fat chance of being something specific to us

one more question about the experiment 2 showing the differences: can you check what's the outcome for the transforms: • ClasspathEntrySnapshotTransform • BuildToolsApiClasspathEntrySnapshotTransform in both builds? (Performance > Transform Execution)

Despite not knowing what the hashes mean, I was able to narrow it down one of the transforms that was different. Out of the image before, there was 1 for an "internal.jar" input and 1 for a "public.jar" that were executed with a different output in the 2nd build. There were other transforms of a "public.jar" re-executed, but only one of an "internal.jar" 🙌🏻. Based on this transform, it might be related to empty projects. The one of "internal.jar" has no src. I also realized that "public/internal.jar" is the de facto "name" of the subproject, like "featureexample:public" 🤦🏻‍♂️ . So I should be able to find the other changed transform, "public.jar", by renaming the projects. PS: In Develocity UI, I couldn't search transforms by input name, but using the API I was able to find them easily

looks like that this is the issue, can you confirm after renaming you don't observe task differences? After looking for similar issues, I found a case with another customer where the difference was in the

kapt

task, but the input was the same:

classpathSnapshotProperties.classpathSnapshot

. In that instance, the problem was that the modularized project contained different modules named "foo," such as "corefoo," "featurefoo", ":foo" and so on. The build scan comparison identified the "foo" snapshot bin as the source of the differences. After they fixed the naming the differences disappeared

Hi, @Inaki Villar! I can confirm that either • renaming each project to a unique name, or • deleting the projects can fix the cache misses. I tested both separately. I also couldn't reproduce it in isolation so far, but I'll try a bit more.

Thanks for the detailed investigation Gabriel, I was not able to reproduce it locally, it's the second time we have seen this issue with project names and

classpathSnapshotProperties.classpathSnapshot

will inform the team internally