Re: svn commit: rev 2346 - trunk/subversion/include trunk/subversion/libsvn_wc trunk/subversion/libsvn_client trunk/subversion/tests/clients/cmdline

Blair Zajac <blair@orcaware.com> writes:

> Yes, there was a core file in the directory. Below is the stack trace.
>
> Shouldn't the test suite check for core dumps and non-zero exit status
> from svn commands? Or can this not be done in a OS portable way?

Well, when our tests run 'svntest.main.run_svn()', we normally invoke
it like this:

    outlines, errlines = svntest.main.run_svn(.....)
    if errlines:
      return 1

But not all our tests do this all the time. But looking at trans_test
#1, I see even bigger bugs... like the fact that it's calling
run_and_verify_status without checking *any* return value at all!
Ugh.

> #0 svn_io_check_path (path=0x13 <Address 0x13 out of bounds>, kind=0x80574e0,
> pool=0x80574e0) at subversion/libsvn_subr/io.c:59
> 59 if (path[0] == '\0')
> (gdb) bt
> #0 svn_io_check_path (path=0x13 <Address 0x13 out of bounds>, kind=0x80574e0,
> pool=0x80574e0) at subversion/libsvn_subr/io.c:59
> #1 0x400320a3 in svn_wc__load_prop_file (propfile_path=0x13 <Address 0x13 out
> of bounds>, hash=0x8081028, pool=0x80574e0) at subversion/libsvn_wc/props.c:234
> #2 0x40028ba1 in svn_wc_transmit_prop_deltas (path=0x8062510
> "/export/amanda/blair/s/subversion/tests/clients/cmdline/working_copies/trans_tests-1/embd_author_rev_exp",

Oop, yeah, I see the bug. My fault, thanks. I'm possibly using an
uninitialized variable... I'm just lucky that it hasn't yet segfaulted
on my box. One patch coming right up.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Jun 27 18:05:39 2002