Ergonomic, hermetic, user-friendly software build system for Python, Go, Java, Kotlin, Protobuf, Scala, Shell. Pants lets you fearlessly scale up your codebase!

Pants

I tried both variants with a local publish of jar tool itself and fast_relpath and os.path.relapth both generate normal jars

Have you figured out a reproducible way to create the bad jars?

we populate classes_by_target by walking the jar.

I switched to os.path.relpath with the config override and it goes away

yes | ./pants --config-override=pants.ini.isolated clean-all publish.jar --no-dryrun --local=/tmp/fast_relpath_isolated src/java/org/pantsbuild/tools/jar/::

Then
```
 $ git diff -U1
diff --git a/src/python/pants/goal/products.py b/src/python/pants/goal/products.py
index d08e3b3..1967849 100644
--- a/src/python/pants/goal/products.py
+++ b/src/python/pants/goal/products.py
@@ -86,3 +86,3 @@ class RootedProducts(object):
     for abs_path in abs_paths:
-      self._rel_paths.add(fast_relpath(abs_path, self._root))
+      self._rel_paths.add(os.path.relpath(abs_path, self._root))
 
```

$ yes | ./pants --config-override=pants.ini.isolated clean-all publish.jar --no-dryrun --local=/tmp/os_path_relpath_isolated src/java/org/pantsbuild/tools/jar/::

just fixing tha unit test difference above should be enough to fix

but understanding the trailing / that triggers would be good to

<@U06A03HV1>: Is twitter using name hashing?

<@U053HNE90>:  no, but we'd like to. <https://github.com/pantsbuild/pants/issues/1861>

Do you have a specific trace where it blows up?