threads / discuss / 44984

Intermittent failure of t1700-split-index.sh

Subject: Intermittent failure of t1700-split-index.sh

## tl;dr

3 messages between Jan 27, 2017 and Jan 27, 2017.

replies: 2people: 3as markdown or json

Ramsay Jones· Jan 27, 2017, 02:45 UTC · lore

Hi Christian,

I noticed the intermittent failure of t1700-split-index.sh (tests #17 and #18) yesterday. It failed in a full test-suite run, but would not fail when run by hand, until I ran it like so:

    $ cd t
    $ for i in `seq 100`; do
    > echo "== $i =="
    > ./t1700-split-index.sh -i -v || break
    > done

... when it failed on the 82nd run!

I only had a brief look in the trash directory, but I could see that the 'twelve' file did not exist, 'eleven' did exist and was in the index (well, git ls-files showed it), and that there were only two '.git/sharedindex.*' files and that their timestamps had not been changed.

I can't devote any time to looking at this further tonight (it's 2-45am here, I'm off to bed!). Can you reproduce the problem, or is it just me? :)

ATB, Ramsay Jones

Jeff King· Jan 27, 2017, 03:58 UTC · re: Ramsay Jones · lore

Re: Intermittent failure of t1700-split-index.sh

On Fri, Jan 27, 2017 at 02:45:15AM +0000, Ramsay Jones wrote:

> I can't devote any time to looking at this further tonight
> (it's 2-45am here, I'm off to bed!). Can you reproduce the
> problem, or is it just me? :)

I can reproduce here with 'pu'. t1700.17 seems to fail reliably with my stress script after 5-10 seconds.

It bisects to ff97d10f57c2f99b6d86da8f07c16659979b2780, but take that with a moderate grain of salt, as I may have marked a bad commit as "good" if it simply got lucky and didn't fail as quickly.

-Peff

Christian Couder· Jan 27, 2017, 06:38 UTC · re: Jeff King · lore

Re: Intermittent failure of t1700-split-index.sh

On Fri, Jan 27, 2017 at 4:58 AM, Jeff King <peff@peff.net> wrote:

Show 12 quoted lines

> On Fri, Jan 27, 2017 at 02:45:15AM +0000, Ramsay Jones wrote:
>
>> I can't devote any time to looking at this further tonight
>> (it's 2-45am here, I'm off to bed!). Can you reproduce the
>> problem, or is it just me? :)
>
> I can reproduce here with 'pu'. t1700.17 seems to fail reliably with my
> stress script after 5-10 seconds.
>
> It bisects to ff97d10f57c2f99b6d86da8f07c16659979b2780, but take that
> with a moderate grain of salt, as I may have marked a bad commit as
> "good" if it simply got lucky and didn't fail as quickly.

Thanks both for your tests and your report. I will take a look.