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

[Jerry Reed]

Thank you so much for all your work, Thomas.

On the BeagleBone briefly.  When I configure with 

./configure

Then the tests fail like this:

Test: 11_precedence...

tcc: error: undefined symbol '__aeabi_idiv'

make[1]: *** [11_precedence.test] Error 255

make[1]: Leaving directory `/home/root/tinycc/tests2'

make: *** [test] Error 2

An error I remember seeing before.  So I rebuilt with

./configure --with-libgcc

Later tests spew lots of diff's related to math/floating point, but I haven't seen a Segmentation fault in 30 runs or so on the BeagleBone.  I'll leave the test looping, but it does seem that the build would have Seg faulted on the Pi by now.

Building on the Pi with-libgcc still gives Segmentation fault however, so perhaps the chip/environment on the Pi makes the problem worse either with libtcc or with libgcc?

If you have a suggestion on how I might get past the error above when building with libtcc on the BeagleBone, just let me know and I'll give it a try.

Regards,

Jerry

On Mon, Jan 28, 2013 at 12:02 AM, Thomas Preud'homme <address@hidden> wrote:

Le dimanche 27 janvier 2013 15:59:48, Thomas Preud'homme a écrit :
> Le dimanche 27 janvier 2013 15:52:55, vous avez écrit :
> > I can get a core file,but it is not very informative, at least to me. The
> > backtrace is only 3 entries, in hex, then a message duplicate
> > entries,possible stack corruption. This is what I got last night, need to
> > try again after the clone -mob.  Anything I can do to get symbolic or
> > additional info into the core file when the seg fault occurs?
>
> Got the same thing here and I don't understand why there is no debugging
> info. Probably because it happens in the program, contrary to what I
> thought. I added a printf just before the main of the program compiled is
> run and it's systematically displayed but any printf in the test itself is
> not displayed. So the crash happens in the very first instructions of the
> compiled program. I'll try to dump the main into a file and maybe I will
> be able to see a problem. What puzzles me is that it's random while the
> code generated should be always the same. I thought about an alignment
> problem but all run are aligned the same way (on 4 bytes while it should
> be 8 bytes).

Ok, I think I have something. It seems sometime the relocation gets wrong and
thus calling a printf leads in the middle of nowhere, hence the segfault. I
can't continue now but if I'm right, expect a fix tomorrow night or the day
after (I'm not sure I'll have internect connectivity tomorrow).

>
> I have to leave now but I'll keep you informed. About the symlinks, I'll
> download a rasbian image and look how the FS is done.

Not done yet but it should be easy.

Best regards,

Thomas

-- 
Interested in bats?  Check out my blog at: http://www.karaokebats.com/