On Wed, Jun 20, 2018 at 10:16 AM William A Rowe Jr <[email protected]> wrote: > > On Wed, Jun 20, 2018 at 9:08 AM, Eric Covener <[email protected]> wrote: >> >> On Wed, Jun 20, 2018 at 10:04 AM William A Rowe Jr <[email protected]> >> wrote: >> > >> > Question, what is a server_conf, before httpd's argv[] array >> > has even been processed? >> >> It's NULL, but these (log.c only) uses should be well after. > > > In fact, some of these are argv[] handlers, but as long as we have > server_conf of NULL, then your patch should be a no-op in such cases. > I'd expect any APLOG_STARTUP phase messages to predate the > successful creation of server_conf.
A good # of these are related to e.g. opening logs which comes much later since it is based on the config -- I think APLOG_STARTUP is just misused/misunderstood (the issue with syslog is actually independent of APLOG_STARTUP) > > The exception would be if some of this processing happens after > pconf is torn down during a graceful restart, without resetting > server_conf to NULL, first. That might show up as graceful restart > segv's. I looked at that a bit and the loop seemed pretty tight. At some point I was looking to explicit reset ap_server_conf but then thought it was not necessary -- maybe no hooks called between cleanups and re-assignment? -- Eric Covener [email protected]
