Re: [PATCH] Move bench target into benchtests

["Tulio Magno Quites Machado Filho" <tuliom at linux dot vnet dot ibm dot com>]

Siddhesh Poyarekar <siddhesh@redhat.com> writes:

> On Tue, Apr 09, 2013 at 06:42:50PM -0400, Carlos O'Donell wrote:
> > On 04/09/2013 09:44 AM, Siddhesh Poyarekar wrote:
> > > On Tue, Apr 09, 2013 at 09:38:27AM -0400, Carlos O'Donell 
> > > wrote:
> > >>> +bench: $(binaries-bench)
> > >>> +	for run in $^; do \
> > >>> +	  echo "Running $${run}"; \
> > >>> +	  eval $(run-bench) >>  $(objpfx)bench.out-tmp; \
> > >>
> > >> Do you need `eval' here? Schwab caught this in my 
> > >> implementation also,
> > >> and I didn't need it because all the variables were resolved 
> > >> by the time
> > >> the command is evaluated.
> > >>
> > > 
> > > Tulio had added it IIRC to get it to work on powerpc.  I'll 
> > > check if
> > > it is in fact needed and if not, I'll just remove it.
> >
> > Did he use a cross-test wrapper?
>
> I don't know.  Tulio, could you please clarify why you had 
> needed the
> additional 'eval' above?

In my tests, without eval ${run} isn't evaluated, which causes a 
segmentation
fault as there isn't an executable called "${run}".

I tested this with both GNU Make 3.81 and 3.82, using latest glibc 
master.

And I'm building this on ppc64 for ppc64 so, no cross-test.

Any ideas why I'm noticing a different behavior?

--
Tulio Magno