Re: In tree configure errors since 6116aea9

[BALATON Zoltan]

On Sat, 21 Mar 2020, Laurent Vivier wrote:
> Le 21/03/2020 à 18:29, BALATON Zoltan a écrit :
> > Hello,
> >
> > Since 6116aea99, or actually 4d6a835d (linux-user: introduce parameters
> > to generate syscall_nr.h) but only next commit starts to enable it I get
> > these errors when running configure in source tree:
> >
> > grep: ./.gitlab-ci.d: Is a directory
> > grep: ./scripts/qemu-guest-agent/fsfreeze-hook.d: Is a directory
> >
> > for each entry in that loop over arches. Could this be silenced?
>
> I didn't see that because I always do an out-of-tree build.

Isn't there a test for that or should there be one?

> Could you try this?
>
> --- a/configure
> +++ b/configure
> @@ -1911,6 +1911,7 @@ for arch in alpha hppa m68k xtensa sh4 microblaze
> arm ppc s390x sparc sparc64 \
>     rm -f "${source_path}/linux-user/${arch}/syscall_nr.h"
>     # remove the dependency files
>     find . -name "*.d" \
> +           -type f \
>            -exec grep -q
> "${source_path}/linux-user/${arch}/syscall_nr.h" {} \; \
>            -exec rm {} \;
> done

This gets rid of the errors but seems to be much slower:

with 4d6a835d running my usual configure script:

real    0m5.968s
user    0m4.642s
sys     0m1.402s

with HEAD and above patch:

real    0m20.246s
user    0m14.143s
sys     0m6.152s

Given that configure is rerun when some files change if there's a way to 
get at least the previous speed back might be better if possible.

Regards,
BALATON Zoltan