Re: What's cooking in git.git (May 2019, #04; Tue, 28)

On 5/30/2019 11:01 AM, Johannes Schindelin wrote:

Hi Peff,

On Thu, 30 May 2019, Jeff King wrote:

quoted

On Wed, May 29, 2019 at 09:53:44AM -0700, Junio C Hamano wrote:

quoted

quoted

quoted

* ds/object-info-for-prefetch-fix (2019-05-28) 1 commit
 - sha1-file: split OBJECT_INFO_FOR_PREFETCH

 Code cleanup.

 Will merge to 'next'.

I think this one is actually a bug-fix (we are refusing to prefetch for
"QUICK" calls even though was not the intent), and it is new in this
release.

I'm not sure of the user-visible impacts, though. There are a lot of
QUICK calls, and I'm not sure for which ones it is important to fetch.

Hmph.  I took it as primarily futureproofing, as I didn't find a way
to trigger bad behaviour from within the current codebase.

Hmm. Looking over the uses of OBJECT_INFO_QUICK, they all seem to be in
either index-pack or as part of a fetch operation. And in both of those
cases, we'd disable the whole feature anyway with fetch_if_missing.

So I _think_ you are right, and there isn't a way to trigger it.

FWIW that was also my impression, but then, I did not look very closely.

In an attempt to find regressions (and to fix them) during the -rc phase,
we rebased VFSforGit's patches on top of current `master`, and saw this
issue. But yeah, the issue fixed by Stolee's patch seems to be caused by
the interaction between current `master` and the VFSforGit patches.

Just to complete the circle, I found this *possible bug* while investigating
test failures with VFS for Git, but it turns out it was related to a
behavior change in the multi-pack-index and the test we were running only
worked by accident before as it was testing a scenario that never occurs
in practice.

The fix required a change to our VFS for Git test. See [1] for full details
if you are interested.

So I agree, this is not a priority to ship in 2.22.0.

Thanks,
-Stolee

[1] https://github.com/microsoft/VFSForGit/pull/1213