Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
On Thu, 2010-10-28 at 13:21 +0100, Caolán McNamara wrote: Hmm, sort of looks like the underlying cppu stuff, or something it uses, got pulled out from under it during the test. I don't think it could have gotten that far for the uno::sequence dtor (which looks up the type of its argument) to blow up without the type system working earlier. I'll have another blast at this, this time only for Linux and copying the rdb files. I'd like to see if it passes in the buildbot anyway. C. ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
Caolan McNamara wrote: register component 'configmgr.uno.so' in registry '../../unxlngx6.pro/misc/qa_unit/services.rdb' failed! error (CannotRegisterImplementationException): loading component library failed: configmgr.uno.so Can you (git pull bootstrap) and try export DEBUGREGCOMP=1 and run dmake in that dir to get a log of the try and catches Thing was, this did not usually happen when simply building there - only during full builds, so I suspect some race... ...anyhow, after the pull no longer happens, instead there's a crash in the number formatter, see second gdb log (the first has the successful regcomp run, for archival). Smells a bit like incomplete uno type info or something, I recall having similar funny crashes when doing minimal uno runtime fakeage in vcl. HTH, -- Thorsten gdbtrace.log.gz Description: GNU Zip compressed data gdbtrace2.log.gz Description: GNU Zip compressed data pgp6fS71ZvngE.pgp Description: PGP signature ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
On Thu, 2010-10-28 at 11:17 +0200, Thorsten Behrens wrote: Caolan McNamara wrote: register component 'configmgr.uno.so' in registry '../../unxlngx6.pro/misc/qa_unit/services.rdb' failed! error (CannotRegisterImplementationException): loading component library failed: configmgr.uno.so Can you (git pull bootstrap) and try export DEBUGREGCOMP=1 and run dmake in that dir to get a log of the try and catches Thing was, this did not usually happen when simply building there - only during full builds, so I suspect some race... ...anyhow, after the pull no longer happens, instead there's a crash in the number formatter, see second gdb log (the first has the successful regcomp run, for archival). Smells a bit like incomplete uno type info or something, I recall having similar funny crashes when doing minimal uno runtime fakeage in vcl. Hmm, sort of looks like the underlying cppu stuff, or something it uses, got pulled out from under it during the test. I don't think it could have gotten that far for the uno::sequence dtor (which looks up the type of its argument) to blow up without the type system working earlier. So maybe some other build process is modifying the shared .rdb files, poking at the testtools bridge test which noone ever complains about and which is run every time I see it simply copies the .rdb locally, so that might be worth a shot as an experiment, so pushed that for a test now. C. ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
Michael Meeks wrote: On Wed, 2010-10-27 at 01:15 +0200, Thorsten Behrens wrote: I've for the while reverted the commit, will look into this tomorrow. Ah ! - better to add a prefix to the prj/build.lst dep. or something to the makefile.mk so it still build on (only) Linux (?) ;-) Looking forward to having it fixed on mac though ... I breaks on *Linux* here. ;) This is all rather nasty - currently fighting with some resmgr issues... Cheers, -- Thorsten pgpCnRuFWpysv.pgp Description: PGP signature ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
On Mon, 2010-10-25 at 21:58 -0400, Kohei Yoshida wrote: On Wed, 2010-10-20 at 20:44 +0100, Caolán McNamara wrote: So the sc test, while working reliably for me, is apparently not working for Kohei and others So, I came home and did a clean re-build of the latest source tree on my reasonably fast desktop machine, and the unit test now runs with 'OK (1)'. So, feel free to re-enable this bit. Righteo, re-enabled again and lets have another go. Right away I see it detected a new bug introduced with the vos removalstuff, which I've now fixed :-). Tests are good. I think I've got the right syntax for this to be only run on Unix platforms, that's my intent anyway. So all I need now is someone on windows to cd sc/qa/unit and run dmake and see if it a) builds b) passes the regcomp lines c) crashes or says ok C. ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
Caolan McNamara wrote: Righteo, re-enabled again and lets have another go. Right away I see it detected a new bug introduced with the vos removalstuff, which I've now fixed :-). Tests are good. Hi Caolan, sorry - breaks for me on at least two platforms consistently. Error log snippet: rm -f ../../unxlngx6.pro/misc/qa_unit/services.rdb : LD_LIBRARY_PATH=${LD_LIBRARY_PATH+${LD_LIBRARY_PATH}:}/builds/ooobuild/ooo-build-master/build/libreoffice-3.2.99.2/solver/330/unxlngx6.pro/lib /builds/ooobuild/ooo-build-master/build/libreoffice-3.2.99.2/solver/330/unxlngx6.pro/bin/regcomp -register -r ../../unxlngx6.pro/misc/qa_unit/services.rdb -wop \ -c configmgr.uno.so \ -c libfwklx.so configmgr.uno.so libfwklx.so register component 'configmgr.uno.so' in registry '../../unxlngx6.pro/misc/qa_unit/services.rdb' failed! error (CannotRegisterImplementationException): loading component library failed: configmgr.uno.so register component 'libfwklx.so' in registry '../../unxlngx6.pro/misc/qa_unit/services.rdb' successful! dmake: Error code 1, while making '../../unxlngx6.pro/misc/qa_unit/services.rdb' dmake: '../../unxlngx6.pro/misc/qa_unit/services.rdb' removed. (fc11.x86_64 openSUSE 11.2-x86_64) I've for the while reverted the commit, will look into this tomorrow. Cheers, -- Thorsten pgpDVaR4Yt1pg.pgp Description: PGP signature ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
On Wed, 2010-10-20 at 20:44 +0100, Caolán McNamara wrote: So the sc test, while working reliably for me, is apparently not working for Kohei and others So, I came home and did a clean re-build of the latest source tree on my reasonably fast desktop machine, and the unit test now runs with 'OK (1)'. I assume my issue was with the incremental build on my laptop since I never really did a fully clean re-build during my travel since that would've taken ages on my laptop. So, feel free to re-enable this bit. In case I get a problem again I'll be well equipped to look into it on my own now that I'm home. Thanks, Kohei -- Kohei Yoshida, LibreOffice hacker, Calc kyosh...@novell.com ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: [Libreoffice] howto for getting non-trivial cppunit tests up and running
Caolan McNamara wrote: sc/qa/unit had an existing cppunit test to ensure that calc can do =SUM(A1:A2) correctly but it was currently disabled seeing as it didn't work, mostly due to miserable bootstrapping problems. I've fixed that test (I think), so I hope it can now provide a basis for adding more. [...] Hi Caolan - interesting, cool that this is actually possible - I kinda evaded the full-scale bootstrapping problem somehow, mostly doing scaffolding for testing parts of app functionality. There are three patches where I ended up with useful unit tests: lobuild/patches/unittesting/unittesting-more-sw.diff lobuild/patches/dev300/impress-autolayout.diff lobuild/patches/dev300/sd_effects_styles.diff The impress autolayout actually gets away w/o any uno setup; the writer test fakes a uno.ini (see the makefile there) - I guess, there's a fine line between trying to fake a ~whole office setup, vs. testing exactly one class (and doing some scaffolding and/or refactoring to really only test that one class) - for the former, maybe the complextest framework is a closer fit? So, one suggestion I have it to look at stripmining testtool and try and start to *move* the calc tests out of the unreliable and painful testtool and into hopefully reliable cppunit tests which are run at buildtime That would be a *very* useful end result. Cheers, -- Thorsten pgpCUdBcRciNq.pgp Description: PGP signature ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice