[eq-dev] Compiling with _GLIBCXX_DEBUG

[Juan Hernando]

Dear all,
I want to debug a random segfault that I'm having inside a third party
library and I want to get the a compilation with the highest number of
runtime debug checks possible. The reason is that valgrind has failed to
diagnose my problem, the crash is random and the backtrace not very
helpful, so I'm a bit lost.

I've already compiled the other C++ libraries with _GLIBCXX_DEBUG to add
the STL debug code, but I didn't manage to do it with Equalizer. After
some naïve tries I've decided to ask for some advice.

I'm using the 0.6 stable release and g++ 4.2.3 in ubuntu. The problem is
related to the hash_map extension. I attach the compiler command line
and output.
Does anybody have any suggestion about how to solve the problem or
workaround it? I don't mind lacking debug support for the hash_table
because my bug is not there, but I need a consistent build.

Of course, this only happens with _GLIBCXX_DEBUG, and that makes me
think there's something awkward inside stdlibc++ but I'd prefer not
messing there.

Thanks in advance,
Juan

[Juan Hernando]

You can forget about this problem.
I was able to replace __gnu_cxx::hash_map with boost::unordered_map and
compile cleanly.
Anyways, it would be a good idea to move to std::tr1 because the
ext/hash_xxx headers are already deprecated in gcc 4.3 and are likely to
be dropped in a near future. Given that boost::unordered_map compiles
just renaming some types and namespaces it shouldn't be hard.

Cheers,
Juan

Juan Hernando escribió:

> ------------------------------------------------------------------------
>
> _______________________________________________
> eq-dev mailing list
> eq-...@equalizergraphics.com
> http://www.equalizergraphics.com/cgi-bin/mailman/listinfo/eq-dev
> http://www.equalizergraphics.com

_______________________________________________
eq-dev mailing list
eq-...@equalizergraphics.com
http://www.equalizergraphics.com/cgi-bin/mailman/listinfo/eq-dev
http://www.equalizergraphics.com

[Stefan Eilemann]

Hi Juan,

On 6. Feb 2009, at 17:22, Juan Hernando (via Nabble) wrote:

> Anyways, it would be a good idea to move to std::tr1 because the
> ext/hash_xxx headers are already deprecated in gcc 4.3 and are  
> likely to
> be dropped in a near future. Given that boost::unordered_map compiles
> just renaming some types and namespaces it shouldn't be hard.

We can do the switch once all major platform support tr1.  
Unfortunately this means VS2005 on Win32, which I guess doesn't  
support it.

I'm generally open to use new features and technologies, but  
unfortunately cross-platform support and minimizing dependencies often  
dictates that it will take years to be able to use new stuff.



Cheers,

Stefan.

---

View this message in context: Re: [eq-dev] Compiling with _GLIBCXX_DEBUG
Sent from the Equalizer - Parallel Rendering mailing list archive at Nabble.com.