Re: [Discuss-gnuradio] build error: test fails in gr_vmcircbuf_mmap_shm_open_factory

[Eric Blossom]

On Sun, Jun 19, 2005 at 09:18:34PM +0200, Patrick Strasser wrote:
> Hello!
> 
> My build processed
> 
> address@hidden:~/src/gnuradio-cvs/gr-build$ ./for-all-dirs ../buildit 
> 2>&1 | tee make.log
> 
> stopped with this error (tee make.log helped a lot :-)):
> 
> .Testing gr_vmcircbuf_sysv_shm_factory...
> ....... gr_vmcircbuf_sysv_shm_factory: OK
> Testing gr_vmcircbuf_mmap_shm_open_factory...
> /bin/sh: line 1: 29405 Aborted                 ${dir}$tst
> FAIL: test_all
> ===================
> 1 of 1 tests failed
> ===================
> 
> I ran the test alone:
> 
> gr-build/gnuradio-core/src/tests$ ./test_vmcircbuf -v
> 
> [lots of lines...]
> Testing gr_vmcircbuf_mmap_shm_open_factory...
> ... test_a_bunch_1x4KB[0]  OK
> ... test_a_bunch_64x16KB[0]  OK
> ... test_a_bunch_64x16KB[1]  OK
> ... test_a_bunch_64x16KB[2]  OK
> ... test_a_bunch_64x16KB[3]  OK
> . [all the same, index increasing]
> .
> ... test_a_bunch_64x16KB[63]  OK
> Abgebrochen
> 
> $ echo $?
> 134
> 
> This is no meaningfull fail of the test. I guess somthing went 
> completely wrong.
> 

Hi Patrick,

We've seen this problem or something very similar to it in the past.
Mostly we've seen the problem on Debian systems.  What's your hardware
and software platform?

The two (possibly related) problems are a buggy implementation of
shm_open and a compiler bug that our code exercises.  As I recall, this
problem generally occurs on x86 using some version of gcc 3.3

If you run gdb on the problem, or take a look at the code, you'll see
that we ought to have a properly installed signal handler to catch
this, but for some reason the compiler with the bug is not setting it
up correctly.

g++ 3.2 and 3.4 and later are known to work.  If you've got the buggy
compiler there may also be other latent problems in other C++ code
compiled into libraries.  E.g., libcppunit.

Eric