@wilbaker noted in #122 that I had changed the parallel test to git rev-list --objects HEAD^{tree} and should not have. This makes the test do nothing of value: we already loaded all blobs at HEAD due to the non-sparse clone. (If we made the test do a sparse clone, then that would be interesting.)
However, without that change we get test failures in ParallelDownloadsInSharedCache().
After investigating, I think we have a different race condition that we didn't see before.
@wilbaker noted in #122 that I had changed the parallel test to
git rev-list --objects HEAD^{tree}and should not have. This makes the test do nothing of value: we already loaded all blobs at HEAD due to the non-sparse clone. (If we made the test do a sparse clone, then that would be interesting.)However, without that change we get test failures in
ParallelDownloadsInSharedCache().After investigating, I think we have a different race condition that we didn't see before.