We do have a problem across platforms; we skip the code in the ssl_cmd_SSLMutex after the first go around. The old code used mc->pPool as a permanent pool for the szMutexFile, but my previous patch used cfg->pool. So this new patch uses the full-blown cmd->server->process->pool so the pstrdup'ed name hangs around for us.
That's not my biggest problem - see the patch to ssl_mutex_init - it turns out we were creating a new mutex on every graceful restart, if I read things correctly. So *no* platforms had a persistent mutex for the lifetime of the server. Verified against 2.0.44 as well, this has been around a while. The new patch creates the mutex itself in the s->process->pool as well so it's truly persistent. This patch is now tested on Win32, users on other platforms are asked to test this code. Comments? Bill
