`git cl split` operates by creating a new git branch for each CL
it uploads. It names this branch based on the directory of the files
in the CL. It uses this convention when resuming an interrupted upload:
it skips branches that already exist.
However, this approach is brittle for several reasons:
- The splitting is nondeterministic, so there's no guarantee that CLs
with the same directories will be generated when re-run. Even if they
are, there's no guarantee they'll have exactly the same files.
- For CLs with multiple directories, it uses an arbitrary one, so it
may not choose the same name a second time
- It bakes in an assumption that every CL is defined by the directories
it draws files from.
This CL alleviates this issues by instead naming branches using the hash
of their contained files. This ensures that, with extremely high
probability, branches will have the same name if and only if they have
the same (sorted) list of files. Thus resumed uploads can safely skip
existing branches.
Bug: 389069356
Change-Id: I19bf7cd6ff8b0127f5906384f7f43b3e40ac8fee
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/6236622
Reviewed-by: Josip Sokcevic <sokcevic@chromium.org>
Commit-Queue: Devon Loehr <dloehr@google.com>
5a3c9438e6