On 14/06/11 17:31, Matthias Klose wrote:
On 06/14/2011 06:06 PM, Philip Ashmore wrote:
I've found a way to get treedb et al building again.
Basically I dropped the do-nothing-but-drag-in-dependent libs.
M4, pkg-config and automake can simply push new variable values, so it's almost
seamless.
I triggered a g++-4.6 miscompilation
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=630441
or something triggered in your code by g++-4.6. Do the build logs show some
suspicious warnings, e.g. about strict aliasing?
Nope. I always try to make my code warning free.
Take a look at that bug report and see if you can reproduce this yourself if
you're
interested.
I'll let the good folks at GNU/g++ figure this out.
Like I said in the bug report, there's a line that sets a pointer followed by
a line that uses the pointer.
The pointer isn't being set, hence the segfault.
It works in debug build with g++ 4.6, and in debug and release builds with g++
4.5.
I should point out that g++-4.6.0-10 is the only version of 4.6 I've used.
If I changed a few lines around and the problem went away, then I wouldn't have
a
code example that the GNU/g++ folks could work with to pin down the problem.
One single point of failure is ridiculously easy to miss otherwise.
I'll stick with 4.5 until then.
Philip
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
