Another placeholder for me... similar to bug 60964. After the configtest has run and the child processes are spawning, the default log file points to stderr (fd 2), but that descriptor has been redirected to /dev/null for child processes. Any ap_log_* calls in a pre_config hook will go to the bit bucket, because we haven't come up with a new log destination yet. This currently prevents my attempt at a graceful warning for users affected by bug 60947. Launching in one-process mode (-X) seems to work around the problem.