[CMake] Re: CMake test suite - 51% failure rate?! (x86/Solaris 10)

Matthew Woehlke mw_triad at users.sourceforge.net
Mon Mar 26 16:09:16 EST 2007


Bill Hoffman wrote:
> Matthew Woehlke wrote:
>> A while back I "successfully"* built CMake on an x86/Solaris 10 box. 
>> Now, trying to build on a different box, I am getting a 51% failure 
>> rate in the test suite. Huh?!
>>
>> (* the test suite has an odd problem with GNU make 3.81, otherwise it 
>> was clean)
>>
>> I also can't seem to find any output from the failed tests, which were:
>> [snip]
>>
>> Ideas? The most obvious difference I can think of is that a different 
>> compiler is being used in this case ("Sun C 5.8 2005/10/13", also 
>> there is a(n ancient!) gcc 2.95.3 available) versus the box where 
>> CMake tested OK ("Sun C 5.7 2005/01/07", no gcc available).
>>
> If the machine is on a network, you can do:
> make Experimental
> 
> This will send the results to the CMake dashboard.  Then others will be 
> able to see the full results.  If not you can run ctest -V to see more 
> verbose output.

ctest -V was monumentally helpful...

ld.so.1: executable: fatal: libstdc++.so.2.10.0: open failed: No such 
file or directory

...I bet that's the problem in most/all cases. Ok, turns out (well, I 
sort-of knew this before, but) g++ is broken, at least in that binaries 
it produces all have this problem. Which means I seem to have found a 
"bug"; tests that involve compiling C++ programs either a: ignore that I 
was using CC=cc CXX=CC to build cmake, or b: fail to notice that g++ 
produces broken executables. (I'm attaching the full output of 
'VERBOSE=1 bin/ctest -V -I 63,63'.)

After forcing the issue with 'CC=cc CXX=cxx bin/ctest', the test suite 
passes.

On a different note: I thought the verbose output was always written to 
a file somewhere; is this not the case?

FYI test 47 is leaking output as follows:
  47/ 67 Testing wrapping                      sh: 
QT_MOC_EXECUTABLE-NOTFOUND: not found
sh: QT_UIC_EXECUTABLE-NOTFOUND: not found
    Passed

Another note: running test 63 "broken" once made it almost impossible to 
clean things up and re-run with the right compilers... ultimately I did 
'rm -rf Tests' and re-unpacked the tarball to fix it. Is it supposed to 
be that difficult? (I did try nuking just Tests/SubDir/CMakeFiles first, 
to no effect.)

Anyway, thanks for the quick help!

-- 
Matthew
"Stop that, it's silly!" -- Colonel
(Monty Python, "And Now For Something Completely Different")
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ctest-broken-g++.log
Type: text/x-log
Size: 9837 bytes
Desc: not available
Url : http://public.kitware.com/pipermail/cmake/attachments/20070326/dd8821ff/ctest-broken-g-0001.bin


More information about the CMake mailing list