On Sat, May 30, 2009 at 9:39 PM, Sander Temme <sander@temme.net> wrote:

On May 29, 2009, at 1:08 AM, William A. Rowe, Jr. wrote:

Candidates in the usual location, may take up to an hour to sync.
Letting this vote run through midnight Monday morning, my time,
so depending on my weekend I'll either stage-to-mirrors late that
night, or first thing Monday a.m. for announcement about 12 hrs
later after some mirrors have caught up.  Windows .zip's will show
up sometime tomorrow when I have the humor :)

+/-1
[-0]  Release apr 1.3.4 as GA
[+1]  Release apr-util 1.3.5 as GA

Folks, your opinions please.

 
-1 on APR for Darwin -- see below.
 

----------
Darwin Legadema.local 9.7.0 Darwin Kernel Version 9.7.0: Tue Mar 31 22:52:17 PDT 2009; root:xnu-1228.12.14~1/RELEASE_I386 i386

apr-1.3.4:

testpoll            : |/bin/sh: line 1:  1831 Bus error               ./$prog
Programs failed: testall
make[1]: *** [check] Error 138
make: *** [check] Error 2

apr-1.3.3: All tests passed.

That's a regression.  Don't know what it means.

FWIW, "make check" has been causing a system panic for me on at least 10.5.5-10.5.7.  As Paul mentioned earlier (somewhere), this is triggered by our kqueue() use (which has no issues AFAICT on FreeBSD).  I don't recall other people reporting system panics.

After disabling kqueue(), I get different crashes in testpoll.

#0  0x000d5eea in impl_pollcb_remove (pollcb=0x208d00, descriptor=0xbffff7d8) at poll/unix/poll.c:352
352        if (descriptor->desc.s == pollcb->copyset[i]->desc.s) {
(gdb) where
#0  0x000d5eea in impl_pollcb_remove (pollcb=0x208d00, descriptor=0xbffff7d8) at poll/unix/poll.c:352
#1  0x000d6365 in apr_pollcb_remove (pollcb=0x208d00, descriptor=0xbffff7d8) at poll/unix/pollcb.c:152
#2  0x00012b9a in trigger_pollcb (tc=0xbffff814, data=0x0) at testpoll.c:614
#3  0x00001eec in abts_run_test (ts=0x200470, f=0x12a14 <trigger_pollcb>, value=0x0) at abts.c:168
#4  0x0001303f in testpoll (suite=0x200470) at testpoll.c:698
#5  0x00002885 in main (argc=3, argv=0xbffff898) at abts.c:424