[PATCH] rebase: ignore failures from "gc --auto"

[PATCH] rebase: ignore failures from "gc --auto"

[Jeff King]

After rebasing, we call "gc --auto" to clean up if we
created a lot of loose objects. However, we do so inside an
&&-chain. If "gc --auto" fails (e.g., because a previous
background gc blocked us by leaving "gc.log" in place),
then:

  1. We will fail to clean up the state directory, leaving
     the user stuck in the rebase forever (even "git am
     --abort" doesn't work, because it calls "gc --auto"!).

  2. In some cases, we may return a bogus exit code from
     rebase, indicating failure when everything except the
     auto-gc succeeded.

We can fix this by ignoring the exit code of "gc --auto".

Signed-off-by: Jeff King <peff@peff.net>
---
This bit me just now. The real blocker was "you have too many loose
objects, run 'git prune'" left in the gc.log file.

It looks like this is the only "gc --auto" we call from a shell script,
and all of the C callers already ignore the exit code. So no other spots
need a similar fix.

 git-rebase.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/git-rebase.sh b/git-rebase.sh
index af7ba5f..cf60c43 100755
--- a/git-rebase.sh
+++ b/git-rebase.sh
@@ -176,7 +176,7 @@ You can run "git stash pop" or "git stash drop" at any time.
 
 finish_rebase () {
 	apply_autostash &&
-	git gc --auto &&
+	{ git gc --auto || true; } &&
 	rm -rf "$state_dir"
 }
 
-- 
2.7.0.242.gdeece69

Re: [PATCH] rebase: ignore failures from "gc --auto"

[Jeff King]

On Wed, Jan 13, 2016 at 01:47:18PM -0500, Jeff King wrote:

> This bit me just now. The real blocker was "you have too many loose
> objects, run 'git prune'" left in the gc.log file.

Which was of course annoying in itself. I was running a scripted set of
rebases, so even with this fix, my terminal got spammed with the
"whoops, we can't auto-gc" message over and over.

The worst part is that when I looked at .git/objects, there were a lot
of objects, but most of them could be cleaned up! I wonder if
too_many_loose_objects() should take into account the prune time of the
objects it finds. I also wonder how I got into that situation, since
that check should come right _after_ we've finished running "gc". I
wonder if I had two backgrounded auto-gc's racing or something.

-Peff

Re: [PATCH] rebase: ignore failures from "gc --auto"

[Rodrigo Campos]

On Wed, Jan 13, 2016 at 01:47:18PM -0500, Jeff King wrote:
> After rebasing, we call "gc --auto" to clean up if we
> created a lot of loose objects. However, we do so inside an
> &&-chain. If "gc --auto" fails (e.g., because a previous
> background gc blocked us by leaving "gc.log" in place),
> then:
> 
>   1. We will fail to clean up the state directory, leaving
>      the user stuck in the rebase forever (even "git am
>      --abort" doesn't work, because it calls "gc --auto"!).

Just curious, but do you mean "git rebase --abort" ?




Thanks,
Rodrigo

Re: [PATCH] rebase: ignore failures from "gc --auto"

[Jeff King]

On Wed, Jan 13, 2016 at 04:40:11PM -0300, Rodrigo Campos wrote:

> On Wed, Jan 13, 2016 at 01:47:18PM -0500, Jeff King wrote:
> > After rebasing, we call "gc --auto" to clean up if we
> > created a lot of loose objects. However, we do so inside an
> > &&-chain. If "gc --auto" fails (e.g., because a previous
> > background gc blocked us by leaving "gc.log" in place),
> > then:
> > 
> >   1. We will fail to clean up the state directory, leaving
> >      the user stuck in the rebase forever (even "git am
> >      --abort" doesn't work, because it calls "gc --auto"!).
> 
> Just curious, but do you mean "git rebase --abort" ?

Yeah, sorry, I did mean "git rebase --abort".

-Peff