GitLab

We've recently received reports of DCHECK failures reflecting an
inconsistent state in the processor's tracker. The problematic scenario
is as follows:
1. There's a pending local creation, which starts with a temporary
   server ID. At the processor/tracker level, this means an empty ID.

2. A remote deletion is received for the same entity (same client tag
   hash), which is treated as conflict.

3. The default conflict resolution will choose to "undelete" the entity,
   that is, the creation should prevail over the deletion. However, the
   temporary server ID must be overriden with the real one as provided
   in the remote update (deletion).

Prior to this patch, ProcessorEntityTracker::RecordIgnoredUpdate() did
partially handle this scenario because the server ID in |metadata_| was
updated. However, the analogous ID in cached |commit_data_| also needs
to be updated.

The underlying issue has been around for very long but was reproduced
consistently with recent ongoing experiments that exercise local
deletions far more often.

Bug: 896138
Change-Id: I53a4e0447d2b809c5691a739a220046c8642156a
Reviewed-on: https://chromium-review.googlesource.com/c/1296169
Commit-Queue: Mikel Astiz <mastiz@chromium.org>
Reviewed-by: Jan Krcal <jkrcal@chromium.org>
Cr-Commit-Position: refs/heads/master@{#601891}