qt5: Disable pkg-config usage for cross-building
diff mbox

Message ID 1474573700-22730-1-git-send-email-abrodkin@synopsys.com
State Rejected
Headers show

Commit Message

Alexey Brodkin Sept. 22, 2016, 7:48 p.m. UTC
Some Qt sub-projects use pkg-config to recover compiler and linker
flags. Among other things include paths will be pulled in CFLAGS.

What's interesting Qt tries to play smart and convert include
paths from simple "-I" to "-isystem" if they match ones returned by [1]
---------------------------->8---------------------------
XXX/output/host/usr/bin/yyy-buildroot-linux-g++
--sysroot=XXX/output/host/usr/yyy-buildroot-linux/sysroot
-pipe -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64
-Os  -xc++ -E -v - < /dev/null
---------------------------->8---------------------------

Most probably that [2] was done to mute tons of warnings generated for
"system" headers. But commit message gives no hint here :(

So Qt's subprojects [and 3-rd party apps written on Qt] may end-up
having "-system ZZZ" in their Makefile. That's all right until an
application wants to use an STL header that includes for example
"stdlib.h" with gcc 6.x, see what happens then [3].

That fixes building of Qt5Webkit on ARM with gcc 6.x, see [4].

[1] http://code.qt.io/cgit/qt/qtbase.git/tree/configure?h=5.6.1#n3660
[2] http://code.qt.io/cgit/qt/qtbase.git/commit?id=0b144bc76a368ecc6c5c1121a1b51e888a0621ac
[3] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70129
[4] http://lists.busybox.net/pipermail/buildroot/2016-September/172666.html

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Peter Seiderer <ps.report@gmx.net>
Cc: Julien Corjon <corjon.j@ecagroup.com>
Cc: "Yann E. MORIN" <yann.morin.1998@free.fr>
---
 package/qt5/qt5base/qt5base.mk | 1 +
 1 file changed, 1 insertion(+)

Comments

Yann E. MORIN Sept. 22, 2016, 8:58 p.m. UTC | #1
Alexey, All,

On 2016-09-22 22:48 +0300, Alexey Brodkin spake thusly:
> Some Qt sub-projects use pkg-config to recover compiler and linker
> flags. Among other things include paths will be pulled in CFLAGS.
> 
> What's interesting Qt tries to play smart and convert include
> paths from simple "-I" to "-isystem" if they match ones returned by [1]
> ---------------------------->8---------------------------
> XXX/output/host/usr/bin/yyy-buildroot-linux-g++
> --sysroot=XXX/output/host/usr/yyy-buildroot-linux/sysroot
> -pipe -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64
> -Os  -xc++ -E -v - < /dev/null
> ---------------------------->8---------------------------
> 
> Most probably that [2] was done to mute tons of warnings generated for
> "system" headers. But commit message gives no hint here :(
> 
> So Qt's subprojects [and 3-rd party apps written on Qt] may end-up
> having "-system ZZZ" in their Makefile. That's all right until an
> application wants to use an STL header that includes for example
> "stdlib.h" with gcc 6.x, see what happens then [3].
> 
> That fixes building of Qt5Webkit on ARM with gcc 6.x, see [4].
> 
> [1] http://code.qt.io/cgit/qt/qtbase.git/tree/configure?h=5.6.1#n3660
> [2] http://code.qt.io/cgit/qt/qtbase.git/commit?id=0b144bc76a368ecc6c5c1121a1b51e888a0621ac
> [3] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70129
> [4] http://lists.busybox.net/pipermail/buildroot/2016-September/172666.html
> 
> Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
> Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
> Cc: Peter Seiderer <ps.report@gmx.net>
> Cc: Julien Corjon <corjon.j@ecagroup.com>
> Cc: "Yann E. MORIN" <yann.morin.1998@free.fr>
> ---
>  package/qt5/qt5base/qt5base.mk | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/package/qt5/qt5base/qt5base.mk b/package/qt5/qt5base/qt5base.mk
> index bf541b0..2c52e2c 100644
> --- a/package/qt5/qt5base/qt5base.mk
> +++ b/package/qt5/qt5base/qt5base.mk
> @@ -26,6 +26,7 @@ QT5BASE_CONFIGURE_OPTS += \
>  	-system-zlib \
>  	-system-pcre \
>  	-no-pch \
> +	-no-pkg-config \

I have not paid much attention to the previous discussion on the topic.

However, I think disabling use of pkg-config is wrong. That is qmake
that is at fault here because, from what I understood from the previous
discussion, it emits -isystem where it should not.

So qmake needs to be fixed to not emit -isystem in those conditions.

Removing the use of pkg-config is just papering over the real problem.

Usually, using pkg-config solves more issues than not using it.

So, I am not very happy with this patch, to say the least... :-(

Yes, I undestand this is a complex issue. Yes, I understand it will be
hard to fix, and even if we can come up with something, it will be hard
to get something upstream that reverts this change.

At worst, let's see if we can at least try to revert that patch from
upstream and see if it fixes the issues.

Regards,
Yann E. MORIN.

>  	-shared
>  
>  # Uses libgbm from mesa3d
> -- 
> 2.7.4
>
Thomas Petazzoni Sept. 23, 2016, 5:57 a.m. UTC | #2
Hello,

On Thu, 22 Sep 2016 22:58:17 +0200, Yann E. MORIN wrote:

> I have not paid much attention to the previous discussion on the topic.
> 
> However, I think disabling use of pkg-config is wrong. That is qmake
> that is at fault here because, from what I understood from the previous
> discussion, it emits -isystem where it should not.
> 
> So qmake needs to be fixed to not emit -isystem in those conditions.
> 
> Removing the use of pkg-config is just papering over the real problem.
> 
> Usually, using pkg-config solves more issues than not using it.
> 
> So, I am not very happy with this patch, to say the least... :-(

I entirely agree with Yann. Disabling pkg-config is clearly not the
solution we want to use.

Alexey, have you tried the other approaches I outlined in the thread
where the initial discussion took place?

Thanks,

Thomas
Alexey Brodkin Sept. 23, 2016, 6:12 a.m. UTC | #3
Hi Thomas, Yann,

On Fri, 2016-09-23 at 07:57 +0200, Thomas Petazzoni wrote:
> Hello,

> 

> On Thu, 22 Sep 2016 22:58:17 +0200, Yann E. MORIN wrote:

> 

> > 

> > I have not paid much attention to the previous discussion on the topic.

> > 

> > However, I think disabling use of pkg-config is wrong. That is qmake

> > that is at fault here because, from what I understood from the previous

> > discussion, it emits -isystem where it should not.

> > 

> > So qmake needs to be fixed to not emit -isystem in those conditions.

> > 

> > Removing the use of pkg-config is just papering over the real problem.

> > 

> > Usually, using pkg-config solves more issues than not using it.

> > 

> > So, I am not very happy with this patch, to say the least... :-(

> 

> I entirely agree with Yann. Disabling pkg-config is clearly not the

> solution we want to use.

>

> Alexey, have you tried the other approaches I outlined in the thread

> where the initial discussion took place?


Well I don't like removal of pkg-config usage either but...
Another approach with setting QMAKE_CFLAGS_ISYSTEM to "" may work as well
(even thought I haven't tried it yet) but that will mean people using
buildroot-built toolchain for building their Qt5 apps will see those
compiler warnings about "wrong" stuff in system headers. In the end that's why
this "-isystem" was introduced in the first place.

In other words disabling pkg-config in Qt5 affects only Qt5 itself while
adding "QMAKE_CFLAGS_ISYSTEM = ''" to "package/qt5/qt5base/qmake.conf"
affects both Qt5 itself and apps that are written on Qt.

-Alexey
Thomas Petazzoni Sept. 23, 2016, 6:19 a.m. UTC | #4
Hello,

On Fri, 23 Sep 2016 06:12:44 +0000, Alexey Brodkin wrote:

> > Alexey, have you tried the other approaches I outlined in the thread
> > where the initial discussion took place?  
> 
> Well I don't like removal of pkg-config usage either but...
> Another approach with setting QMAKE_CFLAGS_ISYSTEM to "" may work as well
> (even thought I haven't tried it yet) but that will mean people using
> buildroot-built toolchain for building their Qt5 apps will see those
> compiler warnings about "wrong" stuff in system headers. In the end that's why
> this "-isystem" was introduced in the first place.

Which warning will they see? I don't quite get why they would see a
warning, and still fail to understand why qmake uses -isystem.

Thomas
Alexey Brodkin Sept. 23, 2016, 7:28 a.m. UTC | #5
Hi Thomas,

On Fri, 2016-09-23 at 08:19 +0200, Thomas Petazzoni wrote:
> Hello,

> 

> On Fri, 23 Sep 2016 06:12:44 +0000, Alexey Brodkin wrote:

> 

> > 

> > > 

> > > Alexey, have you tried the other approaches I outlined in the thread

> > > where the initial discussion took place?  

> > 

> > Well I don't like removal of pkg-config usage either but...

> > Another approach with setting QMAKE_CFLAGS_ISYSTEM to "" may work as well

> > (even thought I haven't tried it yet) but that will mean people using

> > buildroot-built toolchain for building their Qt5 apps will see those

> > compiler warnings about "wrong" stuff in system headers. In the end that's why

> > this "-isystem" was introduced in the first place.

> 

> Which warning will they see? I don't quite get why they would see a

> warning,


I don't have an example handy but maybe this sheds some light on that
https://gcc.gnu.org/onlinedocs/cpp/System-Headers.h
tml?

> and still fail to understand why qmake uses -isystem.


Again I would assume that's to mute warnings that have nothing to
do with Qt itself or user's code.

-Alexey

Patch
diff mbox

diff --git a/package/qt5/qt5base/qt5base.mk b/package/qt5/qt5base/qt5base.mk
index bf541b0..2c52e2c 100644
--- a/package/qt5/qt5base/qt5base.mk
+++ b/package/qt5/qt5base/qt5base.mk
@@ -26,6 +26,7 @@  QT5BASE_CONFIGURE_OPTS += \
 	-system-zlib \
 	-system-pcre \
 	-no-pch \
+	-no-pkg-config \
 	-shared
 
 # Uses libgbm from mesa3d