On 2009-11-27 at 16:57:23 [+0100], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx> wrote: > Ingo Weinhold <ingo_weinhold@xxxxxx> wrote: > > How about make the call synchronous? The comment suggests that it was > > made > > asynchronous only for debugging the registrar's main thread. Not sure > > I > > understand how this is meant, but it probably concerns debugging in > > the test > > environment, which no-one does anymore. > > Doesn't this also mean that you can't start any application that links > against libbe.so if the registrar's thread already exists, but is > currently debugged? Yes, but that's simply what you get when you debug a vital system server. I'm sure the same happens when you try that with certain app server threads. Anyway, that reminds me, that we should change a lot of the static libbe (and to a lesser extend libroot) initialization to be done lazily instead. There's an awful lot of stuff happening (particularly semaphore creation) for any program just linking against libbe. CU, Ingo