Thanks, For getpid, this is my mistake, sorry, I must check before if NPTL getpid is conform to the posix requirements (one PID for all threads in one process). It seems to be Posix compliant : <a href="http://linux.die.net/man/7/pthreads" target="_blank">http://linux.die.net/man/7/pthreads</a> I will check and send you a patch for this. I want to add some annotations like format_arg and noreturn (or an equivalent for MSVC, CLang...) specially for trace, assert, abort and info. Is it useful ? Best regards <div class="gmail_quote">2011/11/21 Michael Meeks <<a href="mailto:michael.meeks@suse.com">michael.meeks@suse.com</a>> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"> Hi there, <div class="im"> On Fri, 2011-11-18 at 20:15 +0100, Arnaud Versini wrote: > The first patch replace the usage of gethostbyname by the reentrant > version. </div> Looks nice, and functional for all unix's that I can see :-) <div class="im"> > The second and third patch removes internal gethostbyname_r, backtrace > and backtrace_symbols_fd. I'm not sure about these two patches, those > functions are in the libc on all current OS and symbols are not > exported. </div> The second looks fine; after all we have this osl_ variant that we are actually using - lets hope that _osl_gethostbyname_r wasn't using this impl. on some obscure platform. I reverted the getpid() section from it - sadly, getpid does (or did) do odd things on Linux around threads (as the comment says) and it's hard to see what that might break. The backtrace patch at worst seems to undo a fix for a non-functiuoning Linux/sparc backtrace method: <a href="https://issues.apache.org/ooo/show_bug.cgi?id=18052" target="_blank">https://issues.apache.org/ooo/show_bug.cgi?id=18052</a> that we can expect is fixed now, and since we don't submit backtraces via the crash-reporter anyway ... ;-) So some nice cleanups; thanks ... One thing I noticed there was the under-usage of osl_sempahore and related bits. It might be worth checking the rest of the code to see if in fact we need this at all - the Condition stuff is much more widely used. If we find it is not / hardly ever used, we could use the generic: #ifdef NO_PTHREAD_SEMAPHORES implementation directly in-lined in the semaphore impl. and schedule Semaphore for ultimate removal in LibreOffice 4: <a href="http://wiki.documentfoundation.org/Development/LibreOffice4" target="_blank">http://wiki.documentfoundation.org/Development/LibreOffice4</a> But perhaps I'm missing something there. Anyhow - thanks for that ! :-) ATB, Michael. -- <a href="mailto:michael.meeks@suse.com">michael.meeks@suse.com</a> <><, Pseudo Engineer, itinerant idiot </blockquote></div> -- Arnaud Versini