cmpil...@apache.org writes:

> Author: cmpilato
> Date: Sat Dec 17 00:56:00 2011
> New Revision: 1215374
>
> URL: http://svn.apache.org/viewvc?rev=1215374&view=rev
> Log:
> Followup to r1215260, just tightening up some loose logic.
>
> * subversion/libsvn_ra_neon/log.c
>   (log_start_element): Reduce the scope of the check for an encoding
>     to just those tags for which it makes sense to have one.  While
>     here, fix a long-standing potential segfault (apr_pstrdup on NULL
>     input).
>   (maybe_decode_log_cdata): Lose 'pool' parameter, and just store in
>     the result in lb->subpool.
>   (log_end_element): Update call to maybe_decode_log_cdata().

This causes log_tests.py 25 to fail with neon when built with -O2 on
my Linux box, the test passes when built -O0.  The compiler warns

../src/subversion/libsvn_ra_neon/log.c: In function ‘log_end_element’:
../src/subversion/libsvn_ra_neon/log.c:293: warning: ‘decoded_cdata’ may be 
used uninitialized in this function

but that may not be the cause since the patch below doesn't fix the
fail:

Index: subversion/libsvn_ra_neon/log.c
===================================================================
--- subversion/libsvn_ra_neon/log.c     (revision 1227106)
+++ subversion/libsvn_ra_neon/log.c     (working copy)
@@ -292,8 +292,7 @@
   struct log_baton *lb = baton;
   const svn_string_t *decoded_cdata;
 
-  if (lb->want_cdata)
-    SVN_ERR(maybe_decode_log_cdata(&decoded_cdata, lb));
+  SVN_ERR(maybe_decode_log_cdata(&decoded_cdata, lb));
 
   switch (state)
     {

-- 
Philip

Reply via email to