From: "William A. Rowe, Jr." <[EMAIL PROTECTED]> Sent: Monday, October 22, 2001 4:31 PM
> From: "Greg Ames" <[EMAIL PROTECTED]> > Sent: Monday, October 22, 2001 4:12 PM > > > > [EMAIL PROTECTED] wrote: > > > > > > wrowe 01/10/21 21:36:46 > > > > > > Modified: server core.c > > > Log: > > > Fix two typos in the last patch... compiles/tests clean... Greg, please > > > run this against Apache.org until the first core and post the results. > > > > So far, so good. No more seg faults on log replays anyway. Thank > > goodness...I wasn't having any success at all trying to trap it. > > > > When/if the load dies down tonight, I'll try it live. > > You wouldn't. I mean to way... you wouldn't have had success in trapping it ;) Please feel free to test. If we see the same (similar) segfaults... > Now that I had a chance to sleep on it several days, restudy the code and > contemplate my navel, I'm convinced it was an optimization bug. > > Apparently the memcpy sufficiently confused the optimizing engine that we > had broken binaries, while the code itself was sound. > > If you look at the core.c to merge per-dir sections, you will notice the > lhs/rhs args to memcpy are casted. Any bets that's how this broke? Try ripping off those casts and see if it magically 'disappears'.