| Summary: | testatomic: Line 908: Unexpected value | ||
|---|---|---|---|
| Product: | APR | Reporter: | Lubos Uhliarik <luhliari> |
| Component: | APR test | Assignee: | Apache Portable Runtime bugs mailinglist <bugs> |
| Status: | NEW --- | ||
| Severity: | normal | ||
| Priority: | P2 | ||
| Version: | 1.7.2 | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Attachments: |
build.log from i686 machine
Config.log apr_private.h |
||
Could you please attach the config.log file? And "include/arch/unix/apr_private.h" too if possible. Created attachment 38495 [details]
Config.log
Hi Yann, I'm attaching config.log and apr_private.h
Created attachment 38496 [details]
apr_private.h
Hi Yann, I can confirm that your patch fixes test failure on i686. Thanks! Lubos, I made some changes in r1907988 to possibly enable atomic builtins on i586 and i686 (gcc for instance provides them for these cpus), and since you are using -march=i686 I wonder if it's a suitable change for you or if -march=i686 should mean --enable-nonportable-atomics=no like before for your (and/or redhat's) case. Since a real i686 cpu is quite unlikely to be used/found these days, what's the point of -march=i686 (wouldn't only -m32 be enough)? Some thoughts/opinions about this change? Hello Yann, We don't ship i686 packages in Fedora/RHEL, we only build them if it is not explicitly turned off. I only wanted to report this test failure, since I thought it might be broken on i686 and I didn't want to turn off i686 builds. If I understand to the change correctly, usage of -march=i686 will set --enable-nonportable-atomics=no and therefore atomics buildins will be disabled. If so I think it is fine by me and as I mentioned before, we only build it on i686 to test if it is build-able on this arch. (In reply to Lubos Uhliarik from comment #8) > If I understand to the change correctly, usage of -march=i686 will set > --enable-nonportable-atomics=no and therefore atomics buildins will be > disabled. The other way around, -march=i686 was implicitely forcing --enable-nonportable-atomics=no unconditionally before this change (thus always using the generic/mutex implementation), now if the compiler supports atomic builtins for i686 they will be used instead (they are more efficient after all). I don't think it matters much because i686 atomic builtins will also work on newer Intel CPUs if APR finally runs on a >i686, but if -march=i686 was used (for whatever reasons) to disable builtins it will not be the case anymore (an explicit --enable-nonportable-atomics=no will still do that of course). My reasoning is that if -march=i686 is usually set as a large Intel 32bit CPUs compatibility toggle then there is no reason to forcibly disable builtins thus existing optimizations (should performance matter on 32bit systems..). PS: I'm talking about 32bit atomics only here, 64bit ones will always use the generic/mutex implementation on real or -march/forced 32bit CPUs. Hello Yann, OK, it makes sense to me. As I previously mention, we don't use/ship i686 packages, so it won't have any impact on our packages performance. I have created a new build including your new config change which forces to use atomic buildins and I can confirm that it builds and passes all tests. Thanks a lot for your time and looking into this! |
Created attachment 38492 [details] build.log from i686 machine After downloading apr-1.7.2 and trying to build it on i686 (on all other platforms it was OK), I'm getting following error: make[1]: Leaving directory '/builddir/build/BUILD/apr-1.7.2/test' + ./testall -v -q testatomic : Line 908: Unexpected value FAILED 1 of 33 testdir : SUCCESS testdso : SUCCESS testdup : SUCCESS testencode : SUCCESS testenv : SUCCESS testescape : SUCCESS testfile : abcdefSUCCESS testfilecopy : SUCCESS testfileinfo : SUCCESS testflock : SUCCESS testfmt : SUCCESS testfnmatch : SUCCESS testargs : SUCCESS testhash : SUCCESS testipsub : SUCCESS testlock : SUCCESS testcond : SUCCESS testlfs : SUCCESS testmmap : SUCCESS testnames : SUCCESS testoc : SUCCESS testpath : SUCCESS testpipe : SUCCESS testpoll : SUCCESS testpools : SUCCESS testproc : SUCCESS testprocmutex : SUCCESS testrand : SUCCESS testsleep : SUCCESS testshm : SUCCESS testsock : SUCCESS testsockets : SUCCESS testsockopt : SUCCESS teststr : SUCCESS teststrnatcmp : SUCCESS testtable : SUCCESS testtemp : SUCCESS testthread : SUCCESS testtime : SUCCESS testud : SUCCESS testuser : SUCCESS testvsn : SUCCESS testskiplist : SUCCESS Failed Tests Total Fail Failed % =================================================== testatomic 33 1 3.03% I'm also attaching whole build log from the i686 machine