commit 63e99e8732ab04c88c6e8e16978d207a33d90720
author Youssef Elghareeb <ghareeb@google.com> Tue Sep 01 18:29:07 2020 +0200
committer Youssef Elghareeb <ghareeb@google.com> Wed Oct 28 00:14:44 2020 +0100
tree c10f0329cf810eef706626fd335c859687c9a088
parent 03e21686e74bfa2374f336ea3eae54499a250b74

Add the GitModifiedFilesCache and ModifiedFilesCache

The two new caches are part of the refactoring done to the Diff cache,
previously implemented in PatchListLoader.java.

The 2 new caches compute the set of modified files (added, deleted,
modified, etc...) between 2 specific commits.

1) The GitModifiedFilesCache computes the modified files between two
commits according to a Git (tree) diff. File contents are not available
with the cache values. The cache key uses the SHA-1s of the trees
instead of the SHA-1s of the commits. This allows the re-use of
computations when commits keep their trees but change in other ways
(e.g. for different commit messages).

2) The ModifiedFilesCache is a wrapper on top of the first cache and
adds extra Gerrit logic. This cache filters out the files that were not
touched between the 2 commits, previously implemented in
PatchListLoader. Similar to the first cache, the file contents and edits
are not available with the cache values.

For now, the magic file paths "/COMMIT_MSG", and "/MERGE_LIST" are not
returned from this cache. In a follow up change, the single file diff
caches will be added and the caller can query the magic file paths from
them. I excluded them from here because:
  * Checking for the cases when we should compute the merge list (e.g.
  determining if the diff is against base / patchsets) would require
  adding some more logic.
  * For commit messages between patchsets, to determine if there's a
  change or not, we would need to parse the actual commit message text.

In the next follow up change, I will implement the protobuf
serialization for the cache entities as marked with TODOs in this
change.

Change-Id: I0c6fc4e015009624c4f93471a2e6335d180f6c82