On 21.01.2014 10:18, Branko Čibej wrote: > On 21.01.2014 03:01, Jeff Trawick wrote: >> On Mon, Jan 20, 2014 at 8:23 PM, Branko Čibej <[email protected] >> <mailto:[email protected]>> wrote: >> >> On 21.01.2014 02:18, Branko Čibej wrote: >> > On 21.01.2014 01:55, Branko Čibej wrote: >> >> I've added a test case to testdir.c to exercise this code in >> parallel. >> > r1559873, original patch (with slightly tweaked comments) and >> test case. >> > Bert, can you please double-check that this works on Windows? >> >> Also r1559878 (1.5.x) and r1559879 (1.4.x). >> >> -- Brane >> >> >> I see intermittent failures testing the 1.5.x branch on Linux (4 >> cores, 8 threads) after picking up your new testcase. The two >> variations seen the most are >> >> testdir : |Line 84: expected <0>, but saw <2> >> testdir : \Line 87: expected <0>, but saw <2> >> >> I've also seen a single failure at line 85 as well as this one that >> needs to be recovered from manually: >> testdir : -Line 199: expected <0>, but saw <39> (39 is >> "Directory not empty") >> >> Current state as of the unrecoverable failure: >> $ find data/prll/ >> data/prll/ >> data/prll/11 >> data/prll/11/12 >> data/prll/11/12/13 >> data/prll/11/12/13/14 >> data/prll/11/12/13/14/15data >> data/prll/11/12/13/14/15data/prll >> data/prll/11/12/13/14/15data/prll/11 >> data/prll/11/12/13/14/15data/prll/11/12 > > "Interesting". I'll take a look.
Heh, it was a bug in the test case; I had it using the same pool from 4 threads, and sometimes apr_pstrdup ended up copying half-baked strings. I have no idea why I didn't see the bug on the Mac; probably different timing and scheduling. >> Also, can you add a CHANGES entry in the 1.4.x and 1.5.x branches? > > Will do, thanks for the reminder. Done. -- Brane
