Fix location where math-vector-fortran.h is installed.

  On 2/26/19 10:41 PM, Zack Weinberg wrote:
> On Tue, Feb 26, 2019 at 4:16 PM Florian Weimer <fweimer@redhat.com> wrote:
>> * Martin Liška:
>>> On 2/26/19 7:05 PM, Joseph Myers wrote:
>>>> On Tue, 26 Feb 2019, Martin Liška wrote:
>>>>
>>>>> Hi.
>>>>>
>>>>> This is follow up patch where I fix location where the header
>>>>> is installed. I made a type as I was copying & pasting.
> 
> Do I understand correctly that the desired installation location is
> $(prefix)/include/finclude/math-vector-fortran.h ?

Yes.

> 
>> This causes test suite failures for me:
>>
>> math/check-installed-headers-c.out:
>>
>> In file included from /tmp/cih_test_C67G81.c:8:
>> ../sysdeps/x86/fpu/finclude/math-vector-fortran.h:1:1: error: expected identifie
>> r or ‘(’ before ‘!’ token
>>  ! Platform-specific declarations of SIMD math functions for Fortran. -*- f90 -*

I skipped that header explicitly.

> 
> Moving the file out of bits/ caused it to be newly subject to these tests.
> 
> If $(prefix)/include/finclude/ is intended to contain nothing but
> Fortran headers, we could add finclude to the list of subdirectories
> to skip in scripts/check*headers* ...
> 
> zw
> 

And for the check-wrapper-headers.py, I read a header file and skip files
with 'f90' file type.

Updated patch is attached.

Martin
  

* Martin Liška:

>          is_nonsysdep_header = os.access(header, os.R_OK)
>          if is_nonsysdep_header:
> +            # Skip Fortran header files
> +            if '-*- f90 -*-' in open(header).readlines()[0]:
> +                continue

Please use “with” to close the file promptly, and do not read the entire
file, like this (untested):

    # Skip Fortran header files.
    with open(header) as inp:
        if '-*- f90 -*-' in next(inp):
            continue

Thanks,
Florian
  

On Feb 27 2019, Martin Liška <mliska@suse.cz> wrote:

> 2019-02-26  Martin Liska  <mliska@suse.cz>
>
> 	* math/Makefile: Change location where math-vector-fortran.h is
> 	installed.
> 	* finclude/math-vector-fortran.h: Move from bits/math-vector-fortran.h.

That file resides in math/.

Andreas.
  

On Wed, Feb 27, 2019 at 3:09 AM Martin Liška <mliska@suse.cz> wrote:
> On 2/26/19 10:41 PM, Zack Weinberg wrote:
> > On Tue, Feb 26, 2019 at 4:16 PM Florian Weimer <fweimer@redhat.com> wrote:
> >> * Martin Liška:
> >>> On 2/26/19 7:05 PM, Joseph Myers wrote:
> >>>> On Tue, 26 Feb 2019, Martin Liška wrote:
> >>>>
> >>>>> Hi.
> >>>>>
> >>>>> This is follow up patch where I fix location where the header
> >>>>> is installed. I made a type as I was copying & pasting.
> >
> > Do I understand correctly that the desired installation location is
> > $(prefix)/include/finclude/math-vector-fortran.h ?
>
> Yes.

Will there ever be C headers in $(prefix)/include/finclude ?

> >> In file included from /tmp/cih_test_C67G81.c:8:
> >> ../sysdeps/x86/fpu/finclude/math-vector-fortran.h:1:1: error: expected identifie
> >> r or ‘(’ before ‘!’ token
> >>  ! Platform-specific declarations of SIMD math functions for Fortran. -*- f90 -*
>
> I skipped that header explicitly.
...
> +        # Skip Fortran header
> +        (finclude/math-vector-fortran.h)
> +            continue;;

To put the above question another way, is there a reason why this shouldn't be

+        # Skip Fortran headers
+        (finclude/*)
+            continue;;

?

> And for the check-wrapper-headers.py, I read a header file and skip files
> with 'f90' file type.

If you're going to do it this way, I recommend you copy the more
general Emacs modeline parser from
https://sourceware.org/ml/libc-alpha/2019-02/msg00539.html ... but if
we can just skip finclude/* that might be a better approach for both
scripts.

zw