Well... this is weird. Seems like a regression in git.

$ git blame `git rev-parse HEAD` -- xyz
fatal: no such path xyz in /usr/share/locale/en_US/LC_MESSAGES/git.mo

This makes no sense.

$ git blame HEAD -- xyz
fatal: no such path xyz in ȷs?�?

This makes less sense.

Confirmed on git tip of tree (5116f791c1, v2.13).

I'll file a bug upstream. In the meantime, do you think we can leave this test as failing or do you want me to temporarily disable it?

I'll do a full bisect later. For now...

Last known good: e572fef9d459497de2bd719747d5625a27c9b41d (Jan 2016)
First known bad: dc076ae5d9fe8c2fbf238b265c1bc1f6f089fd16 (May 2017)

(That's a big bisect.)

Bisect shows 835c49f7d16c5f0307e71da19618fcfe500c2430 is the culprit ("blame: rework methods that determine 'final' commit").

Update: I have a patch for this which I am going to send to the Git mailing list. I'm just checking whether there is any special process for a potential security vulnerability (since this is a use-after-free in git-blame).

Nice!

Re comment 0: I touch depot_tools maybe once a year, so personally I don't care much. But having passing tests on presubmit seems fairly basic on a project's hierarchy of needs, so disabling it might be nice.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/tools/depot_tools/+/6bac8a849974017b8a78c27b97cd52f668bd033f

commit 6bac8a849974017b8a78c27b97cd52f668bd033f
Author: Matt Giuca <mgiuca@chromium.org>
Date: Thu Jul 13 17:18:50 2017

GitHyperBlameMainTest.testBadFilename: Work around git-blame bug.

A regression in git-blame prints an incorrect error message which causes
this test case to fail. Alter the test to only check the start of the
string, until the bug is fixed upstream.

Bug:  737688 
Change-Id: I4045cb8792d8abe984215c7198e213b23e9f6f5d
Reviewed-on: https://chromium-review.googlesource.com/567778
Commit-Queue: Nico Weber <thakis@chromium.org>
Reviewed-by: Nico Weber <thakis@chromium.org>
Reviewed-by: Jochen Eisinger <jochen@chromium.org>

[modify] https://crrev.com/6bac8a849974017b8a78c27b97cd52f668bd033f/tests/git_hyper_blame_test.py

Fixed test. I will post more info about the upstreaming of the fix to git-blame here.

Also removing security since the fix is posted to a public mailing list.