Re: [Libreoffice] make check fails in i18npool

Hi Regina,

On Mon, 2011-10-17 at 15:53 +0200, Regina Henschel wrote:
> >     R=c:/git&&  O=$R/LO35SEPT/solver/wntmsci12.pro&&  W=
> > $R/LO35SEPT/workdir/wntmsci12.pro&&  S=$R/LO35SEPT&&   mkdir -p
> > $W/CppunitTest/&&  (PATH="`cygpath -u $S/solver/wntmsci12.pro`/bin:
> > ${PATH}" STAR_RESOURCEPATH=$O/bin/   $O/bin/cppunit/cppunittester.exe
> > $W/LinkTarget/CppunitTest/test_i18npool_test_breakiterator.dll
> > --protector unoexceptionprotector.dll unoexceptionprotector
> > "-env:UNO_TYPES= file:///$O/bin/udkapi.rdb  file:///$O/bin/types.rdb"
> > "-env:UNO_SERVICES=file:///$O/xml/ure/services.rdb
> > file:///$W/RdbTarget/i18npool_test_breakiterator.rdb"
> > -env:URE_INTERNAL_LIB_DIR=file:///$O/bin
> > -env:OOO_BASE_DIR=file:///$O/bin  -env:BRAND_BASE_DIR=file:///$O/bin
> >
> >     is the one to re-run, preferably inserting a debugger before the
> > cppunittester.exe - to see if we can find what went wrong, a stack trace
> > would be really helpful from wherever it fails there. Perhaps some in
> > some static instance destructor / atexit tear-down code.
>
> I do not understand, what you expect me to do.

        Ah - so the command I paste above (extracted from your log) should be
re-run under cygwin.

        In theory if you paste the above line into your cygwin shell - you will
run just that one unit test, and it will fail for you. If that is the
case, then we need to run it under a debugger.

        So - we need to insert a debugger command (is there some MSVC debugger
you can run, or perhaps just gdb - I'm no windows expert) before the
cppunittester.exe so 

-...STAR_RESOURCEPATH=$O/bin/   $O/bin/cppunit/cppunittester.exe ...
+...STAR_RESOURCEPATH=$O/bin/ gdb $O/bin/cppunit/cppunittester.exe ...

        (at least for linux, for me) - and then run the unit test inside the
debugger. Perhaps Tor / Fridrich could help wrt. the debugger to use.

> >     May have something useful in it - could you poke in there ?
>
> It is only one line:
> OK (4)

        Great - thanks, further confirming that this is some odd at exit
problem.

> I wish, the MSVC&Cygwin problems could be solved. Me last working build 
> is from 2011-08-30.

        Yes; one good piece of news is that Kris did some work to make LibO
compiler under LLVM, which has some very precise error / warning support
- so (hopefully) we can run a tinderbox that compiles rapidly under
Linux, and emulates the Windows warnings: so we can more rapidly close
some of the cross-platform holes. Thanks for being patient with the
problem.

        I suppose (since you're not doing huge global changes); would it be a
solution to have a note or tag on the last-known-good windows compile
that you could checkout and work on top of ? perhaps we could do that.

        Thanks !

                Michael.

-- 
michael.meeks@suse.com  <><, Pseudo Engineer, itinerant idiot