reda-alaoui commented on PR #534:
URL: 
https://github.com/apache/maven-build-cache-extension/pull/534#issuecomment-6810196292

   @olamy, a related case that bears on your two comments here (the cost of 
collecting the graph, and filtering out test-scope dependencies).
   
   We hit a second blind spot of the same kind, unrelated to snapshots. A 
change in a `<dependencyManagement>` version that only applies to a transitive 
dependency leaves the project checksum unchanged:
   
   - The normalized effective pom keeps the declared dependencies and drops 
`<dependencyManagement>`.
   - Direct dependencies do pick up their managed version, so changing one of 
those is detected.
   - A managed version that only steers a transitive dependency never reaches 
the checksum.
   
   What happened to us: a dependency-update bot bumped a managed version (a 
`jakarta.xml.bind-api` bump, and in another project an `xml-apis` downgrade 
where the jar content is identical). The checksum stayed the same, so 
`enforcer:enforce` with `RequireUpperBoundDeps` was restored from cache, and 
the change merged green on a broken dependency graph. The breakage only showed 
up later, on unrelated changes that didn't hit the cache.
   
   Hashing the whole `<dependencyManagement>` block is too broad: 
`CacheInvalidationProjectTraitsTest#bomManagedUnusedDepVersionChangeDoesNotInvalidate`
 rightly expects an unused BOM entry not to invalidate anything. What has to be 
hashed is the version Maven actually picks for each dependency in the graph. 
That is the same graph collection #534 adds, which is why I'm raising it here.
   
   On your two points:
   
   - **Cost.** Collecting the graph only reads POMs, most of them already in 
the local repository; it doesn't need to download any jar. I agree it's extra 
work on cache hits, where Maven would otherwise skip dependency resolution. I 
haven't measured it on a large reactor yet.
   - **Test scope.** For this case it can't be filtered out. The enforcer (and 
any other plugin that reads the graph) checks every scope, so a version change 
in a test-scope dependency must invalidate a cached result that depends on it.
   
   Would you accept the resolved versions being recorded as part of the 
effective pom entry, along the lines AlexanderAshitkin suggested? Or would you 
rather have this tracked as its own issue so #534 stays focused on snapshots? 
I'm happy to open the issue with a reproducer, an integration test on the p01 
reference project.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to