Re: Fwd: Possible bug in subversion

2015-01-23 Thread Martin Wam 
Philip Martin  wandisco.com> writes:

> 
> Martin Wam  esito.no> writes:
> 
> >> That looks as if it could be a problem with APR's mmap support on 
AIX.
> >> Which version of APR are you using?  Did you build the APR binaries
> >> yourself?  Did whoever built the binaries run the APR regression 
> >> tests?
> >
> > We are using apr version 1.5.1-1 downloaded from perzl.org[1]. An 
> > earlier version, 1.4.8-1, gives the same results. We have also tried 
> > building the binaries ourself.
> >
> > We do get a few test failures running them locally, however they do 
not 
> > seem to be related to the above issue.
> >
> > [1]: http://www.perzl.org/aix/index.php?n=Main.Apr
> 
> A stack trace with debug information should provides more information.
> The devel packages might have debug symbols, or if you build your own
> binaries you could build debug versions.
> 

I'm having some issues with dbx, but was able to extract the following 
stacktrace:

Segmentation fault in apr_palloc at line 681 in file 
"memory/unix/apr_pools.c" ($t1)
  681   active = pool->active;
(dbx) where
apr_palloc(pool = 0xfffeee88, in_size = 20), line 681 in "apr_pools.c"
apr_mmap_create(new = 0x2ff21f34, file = 0x30097660, offset = 2, size = 
1, flag = ??, cont = 0xfffeee88), line 100 in "mmap.c"
serf_bucket_file_create() at 0xd259502c
create_update_report_body() at 0xd256ddfc
util.setup_request() at 0xd255f5a0
setup_request_cb() at 0xd255f51c
write_to_connection() at 0xd258ef10
serf__process_connection() at 0xd258fd68
serf_event_trigger() at 0xd258e278
serf_context_run() at 0xd258e154
update.finish_report() at 0xd256d838
svn_wc_crawl_revisions5() at 0xd23d9c20
update_internal() at 0xd23887ac
svn_client__update_internal() at 0xd2388fd4
svn_client__checkout_internal() at 0xd2387f2c
svn_client_checkout3() at 0xd238805c

Re: Fwd: Possible bug in subversion

2015-01-23 Thread Philip Martin 
Martin Wam  writes:

> I'm having some issues with dbx, but was able to extract the following 
> stacktrace:
>
> Segmentation fault in apr_palloc at line 681 in file 
> "memory/unix/apr_pools.c" ($t1)
>   681   active = pool->active;
> (dbx) where
> apr_palloc(pool = 0xfffeee88, in_size = 20), line 681 in "apr_pools.c"
> apr_mmap_create(new = 0x2ff21f34, file = 0x30097660, offset = 2, size = 
> 1, flag = ??, cont = 0xfffeee88), line 100 in "mmap.c"

The code that calls apr_mmap_create() passes always passes a zero offset
here, so seeing offset=2 shows that something is wrong with the stack or
with dbx.  size=1 is possible but suspiciously small.

If you build Subversion with apxs support you could try running the
Subversion HTTP regression tests.  It might be worth building serf as
well.  I know AIX is supposed to be binary compatible but a local build
might work better than a binary built on some other AIX system.

Another option would be to build APR without mmap support, and then
rebuild serf and Subversion using that APR.  It might just crash
somewhere else but it will not crash in apr_mmap_create.

> serf_bucket_file_create() at 0xd259502c
> create_update_report_body() at 0xd256ddfc
> util.setup_request() at 0xd255f5a0
> setup_request_cb() at 0xd255f51c
> write_to_connection() at 0xd258ef10
> serf__process_connection() at 0xd258fd68
> serf_event_trigger() at 0xd258e278
> serf_context_run() at 0xd258e154
> update.finish_report() at 0xd256d838
> svn_wc_crawl_revisions5() at 0xd23d9c20
> update_internal() at 0xd23887ac
> svn_client__update_internal() at 0xd2388fd4
> svn_client__checkout_internal() at 0xd2387f2c
> svn_client_checkout3() at 0xd238805c
>

-- 
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*

Re: your mail

2015-01-23 Thread Greg Stein 
IGNORE. sorry, folks, accidentally moderated it thru :-(

On Fri, Jan 23, 2015 at 08:35:57AM -0800, Val Pangerl wrote:
> val