I've committed the first (and second) iterations on the APR thread migration. There are a couple of things still broken, but I've run out of steam for today.

On Windows, AsyncAppenderTestCase.test3 is deadlocking and is #ifdef'd out. I wasn't able to figure out what was going on and I thought a switch of platform might help. The test fills up the async buffer and enters a wait for a signalled condition, the worker thread blocks on a mutex that should not be locked.

On Linux, static initialization of any ObjectPtr will cause a segmentation fault in apr_atomic_inc32. For a reason I haven't diagnosed, my current version of APR is using mutexes instead of inline assembler. This initialization occurs before APR is initialized which causes the mutex implementation to fault.

It also appears that I let some tabs slip in.  I'll fix those later.



Reply via email to