upcoming enhancement requests for GCC testing

[Janis Johnson]

The GCC testsuite support does some really ugly things to support
functionality that is not available in DejaGnu.  For example, it uses
upvar to reference DejaGnu local variables and overrides several DejaGnu
procs.  I'd like to propose several changes to DejaGnu that would
eventually allow GCC to drop its overrides and use that support directly
from DejaGnu, where it would also be available to other projects.  I'll
send separate mail messages suggesting specific changes.  As far as I
know I don't have a copyright assignment for DejaGnu so I won't send
patches.

GCC provides several test directives, including improvements to some
already available from DejaGnu.  If other projects are interested in
those we can look at copying them to DejaGnu.  They're documented at

http://gcc.gnu.org/onlinedocs/gccint/Test-Directives.html#Test-Directives

The niftiest piece is the ability to use expressions for target lists
and the use of effective-target keywords.  This allows a lot of
flexibility for skipping or xfailing tests to clean up test results.

Janis Johnson
GCC testsuite maintainer




_______________________________________________
DejaGnu mailing list
DejaGnu@...
http://lists.gnu.org/mailman/listinfo/dejagnu

[Rob Savoye]

Janis Johnson wrote:

> The GCC testsuite support does some really ugly things to support
> functionality that is not available in DejaGnu.  For example, it uses

  I think that could be called an understatement. :-) In the begging,
all testsuites actually lived in DejaGnu, so there was tighter
integration. This wasn't practical over the long-term, so I moved all
the testsuites under their respective source directories, and added
support for the "check" Makefile target. At that point the GCC
testsuites, (and I'm sure others) have often added their own support for
things instead of adding them to the core.

> upvar to reference DejaGnu local variables and overrides several DejaGnu
> procs.  I'd like to propose several changes to DejaGnu that would
> eventually allow GCC to drop its overrides and use that support directly
> from DejaGnu, where it would also be available to other projects.  I'll
> send separate mail messages suggesting specific changes.  As far as I
> know I don't have a copyright assignment for DejaGnu so I won't send
> patches.

  If you have a GCC copyright assignment already, I'd think doing one
for DejaGnu is easy. :-) But yes, I like your suggestion of breaking
everything into separate email threads, as we'll want to review them as
a group. My gut feeling is many are probably fine, and should be moved
to DejaGnu core, some are likely very GCC test framework specific, and
should stay there.

> The niftiest piece is the ability to use expressions for target lists
> and the use of effective-target keywords.  This allows a lot of
> flexibility for skipping or xfailing tests to clean up test results.

  And xfailing is a feature primarily used for toolchain testing... :-)

        - rob -



_______________________________________________
DejaGnu mailing list
DejaGnu@...
http://lists.gnu.org/mailman/listinfo/dejagnu

Janis Johnson wrote:

> The GCC testsuite support does some really ugly things to support
> functionality that is not available in DejaGnu.  For example, it uses
> upvar to reference DejaGnu local variables and overrides several DejaGnu
> procs.  I'd like to propose several changes to DejaGnu that would
> eventually allow GCC to drop its overrides and use that support directly
> from DejaGnu, where it would also be available to other projects.  I'll
> send separate mail messages suggesting specific changes.  As far as I
> know I don't have a copyright assignment for DejaGnu so I won't send
> patches.
>
> GCC provides several test directives, including improvements to some
> already available from DejaGnu.  If other projects are interested in
> those we can look at copying them to DejaGnu.  They're documented at
>
> http://gcc.gnu.org/onlinedocs/gccint/Test-Directives.html#Test-Directives
>
> The niftiest piece is the ability to use expressions for target lists
> and the use of effective-target keywords.  This allows a lot of
> flexibility for skipping or xfailing tests to clean up test results.
>
>  

Janis will this help address the types of failures we
see when testing RTEMS?  The two biggest categories
are:

+ scan assembly tests when our BSP needs specific
   CPU flags
+ not generating profile output

I know we have talked about this before.

> Janis Johnson
> GCC testsuite maintainer
>
>
>
>
> _______________________________________________
> DejaGnu mailing list
> DejaGnu@...
> http://lists.gnu.org/mailman/listinfo/dejagnu
>  

--
Joel Sherrill, Ph.D.             Director of Research & Development
joel.sherrill@...        On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             [phone]




_______________________________________________
DejaGnu mailing list
DejaGnu@...
http://lists.gnu.org/mailman/listinfo/dejagnu

[Janis Johnson]

On Wed, 2008-12-17 at 15:36 -0600, Joel Sherrill wrote:

> Janis Johnson wrote:
> > The GCC testsuite support does some really ugly things to support
> > functionality that is not available in DejaGnu.  For example, it uses
> > upvar to reference DejaGnu local variables and overrides several DejaGnu
> > procs.  I'd like to propose several changes to DejaGnu that would
> > eventually allow GCC to drop its overrides and use that support directly
> > from DejaGnu, where it would also be available to other projects.  I'll
> > send separate mail messages suggesting specific changes.  As far as I
> > know I don't have a copyright assignment for DejaGnu so I won't send
> > patches.
> >
> > GCC provides several test directives, including improvements to some
> > already available from DejaGnu.  If other projects are interested in
> > those we can look at copying them to DejaGnu.  They're documented at
> >
> > http://gcc.gnu.org/onlinedocs/gccint/Test-Directives.html#Test-Directives
> >
> > The niftiest piece is the ability to use expressions for target lists
> > and the use of effective-target keywords.  This allows a lot of
> > flexibility for skipping or xfailing tests to clean up test results.
> >
> >  
> Janis will this help address the types of failures we
> see when testing RTEMS?  The two biggest categories
> are:
>
> + scan assembly tests when our BSP needs specific
>    CPU flags
> + not generating profile output
>
> I know we have talked about this before.

No, and I don't know enough about those issues to help, but perhaps
if you describe them here then I can learn from the experts, if they
need to be fixed within the GCC testsuite support.

Janis



_______________________________________________
DejaGnu mailing list
DejaGnu@...
http://lists.gnu.org/mailman/listinfo/dejagnu

[Ben Elliston-37]

Hi Janis

> The GCC testsuite support does some really ugly things to support
> functionality that is not available in DejaGnu.  For example, it
> uses upvar to reference DejaGnu local variables and overrides
> several DejaGnu procs.  I'd like to propose several changes to
> DejaGnu that would eventually allow GCC to drop its overrides and
> use that support directly from DejaGnu, where it would also be
> available to other projects.  I'll send separate mail messages
> suggesting specific changes.

Thanks for taking the time to send these!

Ben


_______________________________________________
DejaGnu mailing list
DejaGnu@...
http://lists.gnu.org/mailman/listinfo/dejagnu