[CMake] random BAD_COMMAND error on ctest

[BILODEAU Andre]

I agree this is probably a cygwin specific issue. The same tests pass without problem on my UNIX computer. Moreover, from the mailing lists, we do not see such a behavior for other platforms.

I put a "sleep 1" command just before exit of the driver, then both at entry and exit. Not better.

After having installed cmake 2.8.4, with no improvement, I reverted to 2.8.10.2 and simplified tests. Here are the results, with a dummy test driver written in korne shell:

1- When the driver contained only shell built-in commands, the tests ran correctly.
2- If the driver contained a non built-in command, such as "grep" (which runs a sub-process), there were some errors, although less than with the real driver (which contains more such commands).
3- If the driver ran an executable, even a trivial one which performs nothing significant, there were also some errors.

As a conclusion, for the moment, the simplest test which reproduces the error consists in running a sub-process from the test program.

-------------
If you think the problem is bad now, just wait until we've solved it.
    --Arthur Kasspe
-------------

-----Message d'origine-----
De : Eric Noulard [mailto:eric.noulard at gmail.com] 
Envoyé : mercredi 9 janvier 2013 08:21
À : BILODEAU Andre
Cc : CMake ML
Objet : Re: [CMake] random BAD_COMMAND error on ctest

Andre please do not drop the ML address.

2013/1/9 BILODEAU Andre <andre.bilodeau at safran-engineering.com>:
> New test:
> I just replaced my test driver by a simple stub, which is run for each of the tests from test_01 to test_17:
>
> #!/bin/ksh
> exit 0
>
> All the tests run without error.
>
> That seems strange because the test driver that the stub replaces works well on autotools or in stand-alone mode, either under cygwin or Mac OS X. The real test driver activate programs, reads and writes files etc., but cmake should only "see" that it exits with status 0.

Yes this weird but this may be some time dependent race, may be you can try to put a 'sleep' in the previous stub.
Moreover from your report and the message from Allen it reallly seems to be a cygwin specific issue.

May be if one of you can craft a repeatable testcase which exhibit the issue it would be nice to file this to the bug tracker.

I won't be able to help you for that since I'm no cygwin user and I have no time to work on this.

Good luck though.
--
Erk
Le gouvernement représentatif n'est pas la démocratie -- http://www.le-message.org