John Klos <john%ziaspace.com@localhost> writes: > I recently updated a netbsd-5 macppc system to netbsd-7. I rebuilt > each package in turn until all had been recompiled to use the built-in > OpenSSL. This included Apache 2.4. You said "rebuilt each package in turn", which sounds like make replace. If not please explain. To verify that everything has been rebuilt in order, I would do two things: remove packages that aren't wanted/needed. I suggest installing pkgin (even if you are building from source), and patrolling the list of "pkgin sk" and "pkgin -n ar" until you believe that everything you want is in the "keepable" list (keepable, means the same as manually installed, which corresponds to not having automatic=YES). Then do "pkgin -n ar". This isn't really relevant, but I find that constant maintenance helps reduce confusion. Run "pkg_rolling-replace -uv". If you have rebuilt everything in the right order, it won't do anything. But if you haven't, and there are packages that were rebuilt after something that depends on them was rebuilt, pkg_rr should do rebuilds in the right order. (At its core, it's just tsorting the dependency list, maintaining dirty flags and rebuilding in order.) > However, this Apache serves files which do not report the current > content-length. The files served appear to be padded with 0x0 bytes - > a 7K file got padded to 8192 bytes, and a 3K file to 8150. I don't > understand why. > > What's strange, though, is when I deleted apache-2.4.17, then added > apache-2.4.17 compiled with OpenSSL 1.0.2c with NetBSD 5, it worked > perfectly. > > I'm not serving these files via https, plus I tried both with and > without compression, but both times the NetBSD 7 compiled Apache had > issues. Besides openssl, a lot of other things have changed. Perhaps package versions (is this 2015Q3 up to date before and after? something else?). Also the compiler has changed, and the new one may be buggy. Actually the old one is probably buggy too but in different ways. I would also remove the old openssl package, if the pkgin instructions about didn't do that. That will make any unintended double linking with 2 openssl versions more obvious :-) > Does anyone have a clue about why this might be and/or a good place to > start trying to track down this issue? Did you run the atf tests in /usr/tests? Do they work as well on your system as how others report? (I'm not saying I really suspect your box - symptoms don't make sense - but it's an easy thing to do and this is part of why they are there.) I would try rebuilding packages without optimization, and with -g, to avoid compiler optimization errors and to get you set up for running gdb on apache.
Attachment:
signature.asc
Description: PGP signature