On Tue, Nov 22, 2011 at 08:12:16AM -0800, Justin Lindberg wrote: > On 11/21/11 23:04, Antoine Jacoutot wrote: > >On Mon, Nov 21, 2011 at 06:11:48PM -0800, Justin Lindberg wrote: > >>The problem is that the shared object library > >> > >>/usr/local/lib/libgthread-2.0.so.2800.0 > >>or > >>/usr/local/lib/libgthread-2.0.so.2992.0 > >>as it is by now in -CURRENT > >> > >>tries to execute function calls that are implemented in > >> > >>/usr/lib/libpthread.so.13.1 > >> > >>These calls will continue to segfault unless libgthread-2.0 is linked > >>with "-lpthread", or equivalently, "--library=pthread" > >Again, *NO*. > >Link your executable with -pthread and stop trying to add -lpthread to all > >libraries, that is not how it works. > I am not adding -lpthread to all my libraries, and I am not > trying to rebuild anything right now. I am trying to figure out > why the binaries released in 5.0 as well as recent snapshots, > namely > > /usr/local/lib/libgthread-2.0.so.2800.0 > and > /usr/local/lib/libgthread-2.0.so.2992.0 > > are segfaulting when they try to call the pthread functions, > and why manually relinking one of them with -lpthread, > without recompiling it, suddenly makes it work.
libgthread is not the problem. The executable that is using libtgthread is the problem. That executable must be linked to all the libraries it needs. In particular libpthread. In OpenBSD it is the responsibility of the executable file to link all the libraries needed. So the executable must be linked with -pthread to work. That's not a problem. That's the intent. .... Ken > > >>If /usr/lib/libpthread.so.13.1 doesn't show up when you run ldd, > >>then libgthread-2.0 was not linked properly. > >Wrong, libpthread is never recorded in so libs when using -pthread and > >linking with -lpthread is not the proper way on OpenBSD for a whole bunch of > >reasons. > >For the last time, you have to link your executable with -pthread. > > > In that case, either -pthread is trying to link that particular > library statically rather than dynamically, or it's trying to > do something new and wonderful that is broken and not yet very > well documented. > > My particular machine is a two-processor i7 with hyperthreading > that runs four pipelines of instructions. >
