gcc.dg/i386-local2.c breaks testsuite on x86

gcc.dg/i386-local2.c breaks testsuite on x86

[Roger Sayle]

Hi Jan,

It looks like your recent testcase, gcc.dg/i386-local2.c breaks
the gcc testsuite on i686-pc-linux-gnu.  For example, see
http://gcc.gnu.org/ml/gcc-testresults/2003-03/msg00405.html

Specifically this bit of that posting:

FAIL: gcc.dg/i386-local.c (test for excess errors)
ERROR: gcc.dg/i386-local.c: error executing dg-final: couldn't open
"i386-local.s": no such file or directory
UNRESOLVED: gcc.dg/i386-local.c: error executing dg-final: couldn't open
"i386-local.s": no such file or directory
ERROR: (DejaGnu) proc "re" does not exist.

The syntax of the DejaGNU directive at the start of the file, i.e.

/* { dg-final { scan-assembler-not "sub.*[re]sp" } } */

is being misinterpreted, so that the "re" in the regular expression
is considered a Tcl procedure.  This terminates the gcc testsuite
prematurely, effectively creating ~400 regressions on the gcc tests.
You'll notice there's only ~17,000 passes for gcc in the above e-mail
rather then the expected ~21,000.

This might be caused by different versions of DejaGNU, if your
testing can't reproduce it.  I'm currently running Tcl 8.3, expect
5.33.0 and dejagnu 1.4.0.  I tried upgrading to dejagnu 1.4.3 but
that doesn't build/install cleanly on RedHat 7.2.


Given the otehr bootstrap failures caused by this commit, might it
be that you applied an earlier version of your patch?  You'll notice
that there are also non-fatal errors from your gcc.dg/i386-local.c
change.

Many thanks in advance,

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833

Re: gcc.dg/i386-local2.c breaks testsuite on x86

[Jan Hubicka]

> 
> Hi Jan,
> 
> It looks like your recent testcase, gcc.dg/i386-local2.c breaks
> the gcc testsuite on i686-pc-linux-gnu.  For example, see
> http://gcc.gnu.org/ml/gcc-testresults/2003-03/msg00405.html
> 
> Specifically this bit of that posting:
> 
> FAIL: gcc.dg/i386-local.c (test for excess errors)
> ERROR: gcc.dg/i386-local.c: error executing dg-final: couldn't open
> "i386-local.s": no such file or directory
> UNRESOLVED: gcc.dg/i386-local.c: error executing dg-final: couldn't open
> "i386-local.s": no such file or directory
> ERROR: (DejaGnu) proc "re" does not exist.
> 
> The syntax of the DejaGNU directive at the start of the file, i.e.
> 
> /* { dg-final { scan-assembler-not "sub.*[re]sp" } } */
> 
> is being misinterpreted, so that the "re" in the regular expression
> is considered a Tcl procedure.  This terminates the gcc testsuite
> prematurely, effectively creating ~400 regressions on the gcc tests.
> You'll notice there's only ~17,000 passes for gcc in the above e-mail
> rather then the expected ~21,000.

Strange, there is no such problem in my runs.  I will fix it by
replacing re by .

Honza
> 
> This might be caused by different versions of DejaGNU, if your
> testing can't reproduce it.  I'm currently running Tcl 8.3, expect
> 5.33.0 and dejagnu 1.4.0.  I tried upgrading to dejagnu 1.4.3 but
> that doesn't build/install cleanly on RedHat 7.2.
> 
> 
> Given the otehr bootstrap failures caused by this commit, might it
> be that you applied an earlier version of your patch?  You'll notice
> that there are also non-fatal errors from your gcc.dg/i386-local.c
> change.
> 
> Many thanks in advance,
> 
> Roger
> --
> Roger Sayle,                         E-mail: roger@eyesopen.com
> OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
> Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
> Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833