math: Enable <bits/math-finite.h> sysdeps override

Message ID 87k1iajiep.fsf@oldenburg2.str.redhat.com
State New
Headers show
Series
  • math: Enable <bits/math-finite.h> sysdeps override
Related show

Commit Message

Florian Weimer Feb. 8, 2019, 5:31 p.m.
Joseph,

I think the patch below is needed.  Is this how things are supposed to
be done?

Compile-tested with build-many-glibcs.py on ia64-linux-gnu.

Thanks,
Florian

math: Enable <bits/math-finite.h> sysdeps override

sysdeps/ia64/fpu/bits/math-finite.h exists and will be installed on
ia64, but during the build, the default math/bits/math-finite.h file
is used, which is wrong.

2019-02-08  Florian Weimer  <fweimer@redhat.com>

	math: Enable <bits/math-finite.h> sysdeps override, as used on ia64.
	* math/bits/math-finite.h: Move to ...
	* sysdeps/generic/bits/math-finite.h: ... here.

Comments

Joseph Myers Feb. 8, 2019, 5:38 p.m. | #1
On Fri, 8 Feb 2019, Florian Weimer wrote:

> Joseph,
> 
> I think the patch below is needed.  Is this how things are supposed to
> be done?

The existing practice is to use top-level bits/ instead of 
sysdeps/generic/bits/ (see 
<https://sourceware.org/ml/libc-alpha/2016-05/msg00189.html>).  Moving to 
top-level bits/ would seem correct for this header (for now, anyway, 
modulo any future integration of math-finite.h into mathcalls.h to fix bug 
23292).
Florian Weimer Feb. 8, 2019, 5:49 p.m. | #2
* Joseph Myers:

> On Fri, 8 Feb 2019, Florian Weimer wrote:
>
>> Joseph,
>> 
>> I think the patch below is needed.  Is this how things are supposed to
>> be done?
>
> The existing practice is to use top-level bits/ instead of 
> sysdeps/generic/bits/ (see 
> <https://sourceware.org/ml/libc-alpha/2016-05/msg00189.html>).  Moving to 
> top-level bits/ would seem correct for this header (for now, anyway, 
> modulo any future integration of math-finite.h into mathcalls.h to fix bug 
> 23292).

Oh, I didn't realize that bits/ would work for this.  I assumed I had to
create sysdeps/generic/bits to make this work. 8-( Should I move the
file sysdeps/generic/bits/unistd_ext.h to bits/unistd_ext.h?

Do you agree that moving math/bits/math-finite.h to bits/math-finite.h
is the right thing to do?

Thanks,
Florian
Joseph Myers Feb. 8, 2019, 5:55 p.m. | #3
On Fri, 8 Feb 2019, Florian Weimer wrote:

> Oh, I didn't realize that bits/ would work for this.  I assumed I had to
> create sysdeps/generic/bits to make this work. 8-( Should I move the
> file sysdeps/generic/bits/unistd_ext.h to bits/unistd_ext.h?

Yes.  (I'm not sure of the rationale for the past move from 
sysdeps/generic/bits/ to bits/, but it was definitely a deliberate change 
establishing the convention that bits/ is the place to use, and I'm not 
aware of any *problems* with using bits/ for this.)

> Do you agree that moving math/bits/math-finite.h to bits/math-finite.h
> is the right thing to do?

Yes.

Patch

diff --git a/math/bits/math-finite.h b/sysdeps/generic/bits/math-finite.h
similarity index 100%
rename from math/bits/math-finite.h
rename to sysdeps/generic/bits/math-finite.h