On Apr 29, 2017 7:24 AM, "Nick Kew" <n...@apache.org> wrote:
On Fri, 2017-04-28 at 11:35 -0500, William A Rowe Jr wrote: > ./configure --enable-timedlocks > > Right off the bat we find new rpm hokum in our configure; Where does the rpm_share come from? Isn't tar a complete red herring? And grep -R doesn't find init_baselib either. > make test is a bit of a mess, here are just the stderr observations; OK, actual errors are presumably the place to look for what might be actual regressions. In summary: > Failed Tests Total Fail Failed % > =================================================== > testdso 9 8 88.89% > testprocmutex 6 3 50.00% > testsock 16 2 12.50% > The less-than-straightforward build is due to crazy AIX > quirks convincing their xlc/ccs toolchains to emit anything > that is 64 bit on a 64 bit build box; You've evidently worked around it. Is that a fix you can apply in our build (from trunk or otherwise), or is it better left as a release note? > Failed Tests Total Fail Failed % > =================================================== > testdso 9 8 88.89% > testprocmutex 6 3 50.00% > testsock 16 2 12.50% So no change there. Good. > The obvious question, what is new since 1.5.2? New tests, > for one, new misbehavior as well; the prior results from > a recent product build with APR 1.5.2; This is quite a lot to digest. Would it be fair to assume that anything outside those failed tests is unlikely to be an actual regression from 1.5.x? Not certain yet but the newly failing dso tests might be the application of LIBPATH in place of LD_LIBRARY_PATH. The new socket test failures are likely sctp discord. The new procmutex failures and failure messages about locking returning immediately worries me.
