Re: [Tinycc-devel] Segmentation Faults and test failures on ARM (Raspbian/Raspberry Pi)

[Thomas Preud'homme]

Le dimanche 27 janvier 2013 02:49:17, vous avez écrit :
> Hoping I picked up the latest and greatest version, as I am not very
> familiar with git:
> 
> git clone git://repo.or.cz/tinycc.git
> git fetch origin -v

The latest version is the mob branch so if you want it, you should do git 
clone -b mob git://repo.or.cz/tinycc.git

Then everytime you want to update to the latest commit just do git pull.

> 
> [hopefully this picks up the version you've been working on so hard
> lately.  If not, please let me know the proper git incantations to get the
> newest stuff]
> 
> Then:
> 
> ./configure
> make
> make test
> 
> At this point, had to make a number of symlinks/copies into /usr/include of
> header files from:
> 
> /usr/include/arm-linux-gnueabihf/ to /usr/include

These shouldn't be necessary. Can you try without the symlink and the mob 
branch?

> 
> [This seems peculiar to the Raspbian distro on the Pi. Never had to do this
> on the BeagleBone - Angstrom distro there]
> 
> make test
> 
> The good news is that the tests get a lot farther than before -- and
> clearly you fixed the problem with 40_stdio.c.  A couple of runs got as far
> as:
> 
> ------------ speedtest ------------
> time ./ex2 1238 2 3 4 10 13 4
> make[1]: time: Command not found
> make[1]: *** [speedtest] Error 127
> make[1]: Leaving directory `/home/pi/tinycc/tests'
> make: *** [test] Error 2
> 
> There is lots of diff output spewed along the way, but the tests do
> complete, which means all the stuff in test2, plus test1, 2 and 3 in
> tests.  Not sure what is up with the speedtest failure, but it is probably
> my environment.
> 
> However the bad news is that other runs still get intermittent Segmentation
> faults, like this:
> 
> Test: 45_empty_for...
> Test: 47_switch_return...
> Test: 48_nested_break...
> Segmentation fault
> make[1]: *** [48_nested_break.test] Error 139
> make[1]: Leaving directory `/home/pi/tinycc/tests2'
> make: *** [test] Error 2
> 
> different places on different runs, of course.

Yeah, I tried to debug this in gdb but of course I can't reproduce it there, 
even when using the batch mode with 100+ run command in the batch file. I 
tried catching the SIGSEGV signal in the 28_strings test to display the 
address of failure but it seems it always fail before the sigaction is put in 
place so I suspect the sigsegv happens in tcc itself. Also, I'm not sure but 
it could be there forever but have been exposed by the long suite of tests in 
tests2. I got the SIGSEGV also in the old testsuite but since there was only 3 
programs run and this seems to happens at tcc -run startup, I guess it was 
just not frequent enough to be noticed.

> 
> So better, but still some problems.  Similar results if I configure
> --with-libgcc as well.
> 
> Will try BeagleBone tomorrow if the VPN to my office is working, otherwise
> will try it Monday AM EST.
> 
> Thanks for all the hard work you and others have put in.

I'm really interested in making the arm port work smoothly so your bug report 
are very valuable.

Thanks

> 
> Regards,
> 
> Jerry

Best regards,

Thomas

signature.asc
Description: This is a digitally signed message part.