[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Bernd Kuhls]

The build error I saw while bumping ffmpeg to 2.3.1 was caused by uClibc, not
ffmpeg:
http://news.gmane.org/find-root.php?message_id=20140804213955.42443190%40debian

ffmpeg-0001-fminf.patch can be removed in favour of uclibc-0063-fminf.patch.

Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
---
 package/ffmpeg/ffmpeg-0001-fminf.patch          |   47 ---------
 package/uclibc/0.9.33.2/uclibc-0063-fminf.patch |  127 +++++++++++++++++++++++
 2 files changed, 127 insertions(+), 47 deletions(-)
 delete mode 100644 package/ffmpeg/ffmpeg-0001-fminf.patch
 create mode 100644 package/uclibc/0.9.33.2/uclibc-0063-fminf.patch

diff --git a/package/ffmpeg/ffmpeg-0001-fminf.patch b/package/ffmpeg/ffmpeg-0001-fminf.patch
deleted file mode 100644
index 1e08517..0000000
--- a/package/ffmpeg/ffmpeg-0001-fminf.patch
+++ /dev/null
@@ -1,47 +0,0 @@
-Fix build error
-
-CC      libavdevice/alldevices.o
-In file included from ./libavutil/internal.h:167:0,
-                 from ./libavutil/common.h:415,
-                 from ./libavutil/avutil.h:289,
-                 from ./libavutil/log.h:25,
-                 from libavdevice/avdevice.h:46,
-                 from libavdevice/alldevices.c:22:
-./libavutil/libm.h:87:40: error: static declaration of 'fminf' follows non-static declaration
- static av_always_inline av_const float fminf(float x, float y)
-                                        ^
-make: *** [libavdevice/alldevices.o] Error 1
-
-using buildroot-compiled uClibc toolchain, used defconfig:
-
-BR2_x86_64=y
-BR2_x86_corei7=y
-BR2_TOOLCHAIN_BUILDROOT_LARGEFILE=y
-BR2_TOOLCHAIN_BUILDROOT_INET_IPV6=y
-BR2_TOOLCHAIN_BUILDROOT_INET_RPC=y
-BR2_TOOLCHAIN_BUILDROOT_WCHAR=y
-BR2_TOOLCHAIN_BUILDROOT_CXX=y
-
-Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
-
-
-diff -uNr ffmpeg-d028c907004e8a3c0f5161ce595331e4cc57c86c.org/libavutil/libm.h ffmpeg-d028c907004e8a3c0f5161ce595331e4cc57c86c/libavutil/libm.h
---- ffmpeg-d028c907004e8a3c0f5161ce595331e4cc57c86c.org/libavutil/libm.h	2014-07-16 20:37:20.000000000 +0200
-+++ ffmpeg-d028c907004e8a3c0f5161ce595331e4cc57c86c/libavutil/libm.h	2014-07-17 22:58:01.781316442 +0200
-@@ -82,16 +82,6 @@
- #define exp2f(x) ((float)exp2(x))
- #endif /* HAVE_EXP2F */
- 
--#if !HAVE_FMINF
--#undef fminf
--static av_always_inline av_const float fminf(float x, float y)
--{
--    //Note, the NaN special case is needed for C spec compliance, it should be
--    //optimized away if the users compiler is configured to assume no NaN
--    return x > y ? y : (x == x ? x : y);
--}
--#endif
--
- #if !HAVE_ISINF
- static av_always_inline av_const int isinf(float x)
- {
diff --git a/package/uclibc/0.9.33.2/uclibc-0063-fminf.patch b/package/uclibc/0.9.33.2/uclibc-0063-fminf.patch
new file mode 100644
index 0000000..0cb5ca3
--- /dev/null
+++ b/package/uclibc/0.9.33.2/uclibc-0063-fminf.patch
@@ -0,0 +1,127 @@
+To fix the fminf-related build error in ffmpeg a patch for uClibc is necessary,
+taken from upstream patchwork:
+
+http://patchwork.ozlabs.org/patch/392465/
+
+Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
+
+
+From patchwork Tue Sep 23 12:17:25 2014
+Content-Type: text/plain; charset="utf-8"
+MIME-Version: 1.0
+Content-Transfer-Encoding: 7bit
+Subject: Do not define unimplemented functions
+From: Cristian Morales Vega <cristian@samknows.com>
+X-Patchwork-Id: 392465
+Message-Id: <1411474645-19490-1-git-send-email-cristian@samknows.com>
+To: uclibc at uclibc.org
+Date: Tue, 23 Sep 2014 13:17:25 +0100
+
+e.g. fminf() is not implemented, only fmin(), but both are defined.
+
+Signed-off-by: Cristian Morales Vega <cristian@samknows.com>
+---
+ include/math.h                             |  2 ++
+ libc/sysdeps/linux/common/bits/mathcalls.h | 18 ++++++++++++++++++
+ 2 files changed, 20 insertions(+)
+
+diff --git a/include/math.h b/include/math.h
+index ecb9aa6..40dd90e 100644
+--- a/include/math.h
++++ b/include/math.h
+@@ -118,6 +118,7 @@ __BEGIN_DECLS
+ #  define _Mfloat_		float
+ # endif
+ # define _Mdouble_		_Mfloat_
++# define _Mdouble_is_float_
+ # ifdef __STDC__
+ #  define __MATH_PRECNAME(name,r) name##f##r
+ # else
+@@ -126,6 +127,7 @@ __BEGIN_DECLS
+ # define _Mdouble_BEGIN_NAMESPACE __BEGIN_NAMESPACE_C99
+ # define _Mdouble_END_NAMESPACE   __END_NAMESPACE_C99
+ # include <bits/mathcalls.h>
++# undef _Mdouble_is_float_
+ # undef	_Mdouble_
+ # undef _Mdouble_BEGIN_NAMESPACE
+ # undef _Mdouble_END_NAMESPACE
+diff --git a/libc/sysdeps/linux/common/bits/mathcalls.h b/libc/sysdeps/linux/common/bits/mathcalls.h
+index 84b793c..4270273 100644
+--- a/libc/sysdeps/linux/common/bits/mathcalls.h
++++ b/libc/sysdeps/linux/common/bits/mathcalls.h
+@@ -271,7 +271,9 @@ __END_NAMESPACE_C99
+ #ifdef __USE_ISOC99
+ __BEGIN_NAMESPACE_C99
+ /* True gamma function.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (tgamma,, (_Mdouble_))
++# endif
+ __END_NAMESPACE_C99
+ #endif
+ 
+@@ -299,7 +301,9 @@ __MATHCALLI (rint,, (_Mdouble_ __x))
+ /* Return X + epsilon if X < Y, X - epsilon if X > Y.  */
+ __MATHCALLX (nextafter,, (_Mdouble_ __x, _Mdouble_ __y), (__const__))
+ # if defined __USE_ISOC99 && !defined __LDBL_COMPAT
++#  ifndef _Mdouble_is_float_
+ __MATHCALLX (nexttoward,, (_Mdouble_ __x, long double __y), (__const__))
++#  endif
+ # endif
+ 
+ /* Return the remainder of integer divison X / Y with infinite precision.  */
+@@ -316,11 +320,15 @@ __MATHDECLI (int,ilogb,, (_Mdouble_ __x))
+ 
+ #ifdef __USE_ISOC99
+ /* Return X times (2 to the Nth power).  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (scalbln,, (_Mdouble_ __x, long int __n))
++# endif
+ 
+ /* Round X to integral value in floating-point format using current
+    rounding direction, but do not raise inexact exception.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (nearbyint,, (_Mdouble_ __x))
++# endif
+ 
+ /* Round X to nearest integral value, rounding halfway cases away from
+    zero.  */
+@@ -333,7 +341,9 @@ __MATHCALLX (trunc,, (_Mdouble_ __x), (__const__))
+ /* Compute remainder of X and Y and put in *QUO a value with sign of x/y
+    and magnitude congruent `mod 2^n' to the magnitude of the integral
+    quotient x/y, with n >= 3.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (remquo,, (_Mdouble_ __x, _Mdouble_ __y, int *__quo))
++# endif
+ 
+ 
+ /* Conversion functions.  */
+@@ -350,13 +360,19 @@ __MATHDECLI (long long int,llround,, (_Mdouble_ __x))
+ 
+ 
+ /* Return positive difference between X and Y.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (fdim,, (_Mdouble_ __x, _Mdouble_ __y))
++# endif
+ 
+ /* Return maximum numeric value from X and Y.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (fmax,, (_Mdouble_ __x, _Mdouble_ __y))
++# endif
+ 
+ /* Return minimum numeric value from X and Y.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (fmin,, (_Mdouble_ __x, _Mdouble_ __y))
++# endif
+ 
+ 
+ /* Classify given number.  */
+@@ -367,7 +383,9 @@ __MATHDECL_PRIV (int, signbit,, (_Mdouble_ __value), (__const__))
+ 
+ 
+ /* Multiply-add function computed as a ternary operation.  */
++# ifndef _Mdouble_is_float_
+ __MATHCALLI (fma,, (_Mdouble_ __x, _Mdouble_ __y, _Mdouble_ __z))
++# endif
+ #endif /* Use ISO C99.  */
+ 
+ #if defined __USE_MISC || defined __USE_XOPEN_EXTENDED || defined __USE_ISOC99
-- 
1.7.10.4

[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Thomas Petazzoni]

Dear Bernd Kuhls,

On Tue, 23 Sep 2014 15:41:39 +0200, Bernd Kuhls wrote:
> The build error I saw while bumping ffmpeg to 2.3.1 was caused by uClibc, not
> ffmpeg:
> http://news.gmane.org/find-root.php?message_id=20140804213955.42443190%40debian
> 
> ffmpeg-0001-fminf.patch can be removed in favour of uclibc-0063-fminf.patch.
> 
> Signed-off-by: Bernd Kuhls <bernd.kuhls@t-online.de>
> ---
>  package/ffmpeg/ffmpeg-0001-fminf.patch          |   47 ---------
>  package/uclibc/0.9.33.2/uclibc-0063-fminf.patch |  127 +++++++++++++++++++++++
>  2 files changed, 127 insertions(+), 47 deletions(-)
>  delete mode 100644 package/ffmpeg/ffmpeg-0001-fminf.patch
>  create mode 100644 package/uclibc/0.9.33.2/uclibc-0063-fminf.patch

Even though this looks like the right fix, it is a bit annoying,
because it means that ffmpeg will no longer build with a large number
of external uClibc toolchains (built with Buildroot versions older than
the one that would have this commit included, built with Crosstool-NG,
made available by third-parties, etc.).

Do you think we could have a solution that combines keeping the ffmpeg
fminf() patch, and adding the uClibc patch? Having the uClibc patch is
good because it fixes a real issue in uClibc. Keeping the ffmpeg
fminf() patch ensures that our ffmpeg package builds with other uClibc
toolchains than just the latest Buildroot uClibc toolchain.

What do you think?

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Bernd Kuhls]

[posted and mailed]

Thomas Petazzoni <thomas.petazzoni-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8
@public.gmane.org> wrote in news:20141005234010.17ab13c5 at free-electrons.com:

> Do you think we could have a solution that combines keeping the ffmpeg
> fminf() patch, and adding the uClibc patch?

Hi,

I will investigate, but it will take several days for me to do so. Meanwhile 
I will send the next ffmpeg version bump now ;)

Regards, Bernd

[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Thomas Petazzoni]

Dear Bernd Kuhls,

On Mon, 06 Oct 2014 21:27:38 +0200, Bernd Kuhls wrote:
> [posted and mailed]
> 
> Thomas Petazzoni <thomas.petazzoni-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8
> @public.gmane.org> wrote in news:20141005234010.17ab13c5 at free-electrons.com:
> 
> > Do you think we could have a solution that combines keeping the ffmpeg
> > fminf() patch, and adding the uClibc patch?
> 
> Hi,
> 
> I will investigate, but it will take several days for me to do so. Meanwhile 
> I will send the next ffmpeg version bump now ;)

Any progress on this issue?

Thanks!

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH 1/1] uclibc/ffmpeg: Fix fminf-related build error

[Bernd Kuhls]

[posted and mailed]

Thomas Petazzoni
<thomas.petazzoni@free-electrons.com> wrote
in news:20141030230535.4d6673bd at free-electrons.com: 

> Dear Bernd Kuhls,
> 
> On Mon, 06 Oct 2014 21:27:38 +0200, Bernd Kuhls wrote:
>> [posted and mailed]
>> 
>> Thomas Petazzoni <thomas.petazzoni-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8
>> @public.gmane.org> wrote in
>> news:20141005234010.17ab13c5-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8 at public.gma
>> ne.org: 
>> 
>> > Do you think we could have a solution that combines keeping the
>> > ffmpeg fminf() patch, and adding the uClibc patch?
>> 
>> Hi,
>> 
>> I will investigate, but it will take several days for me to do so.
>> Meanwhile I will send the next ffmpeg version bump now ;)
> 
> Any progress on this issue?

Hi Thomas,

yes, upstream fixed the issue yesterday!
http://patchwork.ozlabs.org/patch/408557/

Regards, Bernd