[Buildroot] Qt5WebKit anf gcc 6.x

[Buildroot] Qt5WebKit anf gcc 6.x

[Alexey Brodkin]

Hello,

I'd like to get some suggestions on how to cope with another gcc 6.x related issue.
I'm trying to build Qt5Webkit with gcc 6.x either for ARM (with gcc 6 selected) or
for ARC (where as of now we use gcc 6.2-based tools only).

The first sign of this problem I see on WebKit configuration:
--------------------->8-------------------
Info: creating cache file XXX/.qmake.cache

Running configure tests...

Checking for fontconfig... no
Checking for gccdepends... yes
Checking for glx... no
Checking for gnuld... yes
Checking for icu... no

...

The WebKit build was disabled for the following reasons:?
????* ICU is required.
--------------------->8-------------------

But I knew ICU was a dependency and it was really built so I started to look deeper.
Real breakage happens on ICU check:
--------------------->8-------------------
cd output/build/qt5webkit-b889f460280ad98c89ede179bd3b9ce9cb02002b/Tools/qmake/config.tests/icu
../../../../../../host/usr/bin/qmake?
make
XXX/output/host/usr/bin/arm-buildroot-linux-uclibcgnueabi-g++ -c -Wall -Wextra -Wreturn-type -Wchar-subscripts -Wformat-
security -Wreturn-type -Wno-unused-parameter -Wno-sign-compare -Wno-switch -Wno-switch-enum -Wundef -Wmissing-noreturn
-Winit-self -pipe -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Os --
sysroot=XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot -fno-strict-aliasing -O2 -O3 -std=gnu++0x -fno-
exceptions -fPIC -I. -IXXX/output/build/qt5webkit-b889f460280ad98c89ede179bd3b9ce9cb02002b/Source
-IXXX/output/build/qt5webkit-b889f460280ad98c89ede179bd3b9ce9cb02002b/Source/include -isystem XXX/output/host/usr/arm-
buildroot-linux-uclibcgnueabi/sysroot/usr/include -IXXX/output/host/usr/mkspecs/devices/linux-buildroot-g++ -o
.obj/icu.o icu.cpp
In file included from XXX/output/host/usr/arm-buildroot-linux-
uclibcgnueabi/include/c++/6.2.0/ext/string_conversions.h:41:0,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/include/c++/6.2.0/bits/basic_string.h:5402,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/include/c++/6.2.0/string:52,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/std_string.h:33,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/unistr.h:31,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/strenum.h:14,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/uenum.h:24,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/uloc.h:25,
?????????????????from XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include/unicode/ucol.h:18,
?????????????????from icu.cpp:35:
XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/include/c++/6.2.0/cstdlib:75:25: fatal error: stdlib.h: No such
file or directory
?#include_next <stdlib.h>
?????????????????????????^
compilation terminated.
make: *** [.obj/icu.o] Error 1
--------------------->8-------------------

So there's a known "feature" of gcc 6.x, see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70129

In its turn "-isystem XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include"
appears in ICU's check Makefile because
--------------------->8-------------------
./pkg-config_wrapper.sh --cflags icu-i18n
-IXXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include
--------------------->8-------------------
and that include path matches with one entry from?QMAKE_DEFAULT_INCDIRS.
In that case qmake makes an assumption that it's a system path and adds "-isystem".

At this point I'm not sure how to fix that puzzle so any ideas are much appreciated.

Indeed there's such options like disabling usage of pkg-config in Qt.
And that's quite easy one we just need to add "no-pkg-config" to?QT_CONFIG for
problematic packages but probably there're better approaches here.

Note Qt uses its own pkg-config_wrapper.sh which looks like:
--------------------->8-------------------
#!/bin/sh
PKG_CONFIG_SYSROOT_DIR=XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot
export PKG_CONFIG_SYSROOT_DIR
PKG_CONFIG_LIBDIR=XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/pkgconfig
export PKG_CONFIG_LIBDIR
exec pkg-config "$@"
--------------------->8-------------------

But it then executes Buildroot's pkg-config wrapper which overrides PKG_CONFIG_XXX entries:
--------------------->8-------------------
cat XXX/output/host/usr/bin/pkg-config

#!/bin/sh
PKG_CONFIG_LIBDIR=${PKG_CONFIG_LIBDIR:-XXX/output/host/usr/arm-buildroot-linux-
uclibcgnueabi/sysroot/usr/lib/pkgconfig:XXX/output/host/usr/arm-buildroot-linux-
uclibcgnueabi/sysroot/usr/share/pkgconfig} PKG_CONFIG_SYSROOT_DIR=${PKG_CONFIG_SYSROOT_DIR:-XXX/output/host/usr/arm-
buildroot-linux-uclibcgnueabi/sysroot} $(dirname $0)/pkgconf??$@
--------------------->8-------------------

Regards,
Alexey

[Buildroot] Qt5WebKit anf gcc 6.x

[Thomas Petazzoni]

Hello,

On Wed, 21 Sep 2016 15:28:21 +0000, Alexey Brodkin wrote:

> In its turn "-isystem XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include"
> appears in ICU's check Makefile because
> --------------------->8-------------------  
> ./pkg-config_wrapper.sh --cflags icu-i18n
> -IXXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include
> --------------------->8-------------------  
> and that include path matches with one entry from?QMAKE_DEFAULT_INCDIRS.
> In that case qmake makes an assumption that it's a system path and adds "-isystem".

Are you sure about this? I don't see this happening anywhere in the
qmake source code. I do see some isystem related code around
QMAKE_CFLAGS_ISYSTEM and INCLUDEPATH.

In any case, I believe this is where the problem is: I don't see why
Qt5 should mess up with the addition of -isystem flags.

Best regards,

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

[Buildroot] Qt5WebKit anf gcc 6.x

[Alexey Brodkin]

Hi Thomas,

On Wed, 2016-09-21 at 21:09 +0200, Thomas Petazzoni wrote:
> 
> Hello,
> 
> On Wed, 21 Sep 2016 15:28:21 +0000, Alexey Brodkin wrote:
> 
> > 
> > 
> > In its turn "-isystem XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include"
> > appears in ICU's check Makefile because
> > --------------------->8-------------------??
> > ./pkg-config_wrapper.sh --cflags icu-i18n
> > -IXXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include
> > --------------------->8-------------------??
> > and that include path matches with one entry from?QMAKE_DEFAULT_INCDIRS.
> > In that case qmake makes an assumption that it's a system path and adds "-isystem".
> 
> Are you sure about this? I don't see this happening anywhere in the
> qmake source code. I do see some isystem related code around
> QMAKE_CFLAGS_ISYSTEM and INCLUDEPATH.

Take a look here:?https://github.com/qt/qtbase/blob/dev/qmake/generators/unix/unixmake2.cpp#L179
That's what I mean:
--------------------->8-------------------
? ? t << "####### Compiler, tools and options\n\n";
????t << "CC????????????= " << var("QMAKE_CC") << endl;
????t << "CXX???????????= " << var("QMAKE_CXX") << endl;
????t << "DEFINES???????= "
??????<< varGlue("PRL_EXPORT_DEFINES","-D"," -D"," ")
??????<< varGlue("DEFINES","-D"," -D","") << endl;
????t << "CFLAGS????????= " << var("QMAKE_CFLAGS") << " $(DEFINES)\n";
????t << "CXXFLAGS??????= " << var("QMAKE_CXXFLAGS") << " $(DEFINES)\n";
????t << "INCPATH???????=";
????{
????????QString isystem = var("QMAKE_CFLAGS_ISYSTEM");
????????const ProStringList &incs = project->values("INCLUDEPATH");
????????for(int i = 0; i < incs.size(); ++i) {
????????????const ProString &inc = incs.at(i);
????????????if (inc.isEmpty())
????????????????continue;

????????????if (!isystem.isEmpty() && isSystemInclude(inc.toQString()))
????????????????t << ' ' << isystem << ' ';
????????????else
????????????????t << " -I";
????????????t << escapeFilePath(inc);
????????}
????}
--------------------->8-------------------

For G++ "QMAKE_CFLAGS_ISYSTEM = -isystem", see?https://github.com/qt/qtbase/blob/dev/mkspecs/common/gcc-base.conf#L47

That's how include path gets prefixed with either normal "-I" if path has no matches in
QMAKE_DEFAULT_INCDIRS otherwise "-isystem " prefix is used.

> 
> In any case, I believe this is where the problem is: I don't see why
> Qt5 should mess up with the addition of -isystem flags.

Agree, but see what Qt people say in similar bug reports:
https://bugreports.qt.io/browse/QTBUG-53367
https://bugreports.qt.io/browse/QTBUG-53375

And BTW that commit added all this mess:
http://code.qt.io/cgit/qt/qtbase.git/commit?id=0b144bc76a368ecc6c5c1121a1b51e888a0621ac

-Alexey

[Buildroot] Qt5WebKit anf gcc 6.x

[Thomas Petazzoni]

Hello,

On Wed, 21 Sep 2016 19:22:42 +0000, Alexey Brodkin wrote:

> For G++ "QMAKE_CFLAGS_ISYSTEM = -isystem", see?https://github.com/qt/qtbase/blob/dev/mkspecs/common/gcc-base.conf#L47
> 
> That's how include path gets prefixed with either normal "-I" if path has no matches in
> QMAKE_DEFAULT_INCDIRS otherwise "-isystem " prefix is used.

OK.

> > In any case, I believe this is where the problem is: I don't see why
> > Qt5 should mess up with the addition of -isystem flags.  
> 
> Agree, but see what Qt people say in similar bug reports:
> https://bugreports.qt.io/browse/QTBUG-53367
> https://bugreports.qt.io/browse/QTBUG-53375
> 
> And BTW that commit added all this mess:
> http://code.qt.io/cgit/qt/qtbase.git/commit?id=0b144bc76a368ecc6c5c1121a1b51e888a0621ac

Thanks for the pointers. Unfortunately, the commit log is useless: it
says what the commit does, but not *why*. So we have no idea why such a
change was done, and the author is not very helpful in explanations in
the bug reports.

So, I believe there are two possibilities here:

 - Check who is adding
   XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include
   to QMAKE_DEFAULT_INCDIRS, and change it to add it to INCLUDEPATH
   instead, so that it's not considered as a "system include path".
   This is what the developer suggested in
   https://bugreports.qt.io/browse/QTBUG-53367.

 - Change our package/qt5/qt5base/qmake.conf file to set
   QMAKE_CFLAGS_ISYSTEM to the empty value, which will disable this new
   "isystem" logic.

Best regards,

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

[Buildroot] Qt5WebKit anf gcc 6.x

[Alexey Brodkin]

Hi Thomas,

On Wed, 2016-09-21 at 21:09 +0200, Thomas Petazzoni wrote:
> Hello,
> 
> On Wed, 21 Sep 2016 15:28:21 +0000, Alexey Brodkin wrote:
> 
> > 
> > In its turn "-isystem XXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include"
> > appears in ICU's check Makefile because
> > --------------------->8-------------------??
> > ./pkg-config_wrapper.sh --cflags icu-i18n
> > -IXXX/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/include
> > --------------------->8-------------------??
> > and that include path matches with one entry from?QMAKE_DEFAULT_INCDIRS.
> > In that case qmake makes an assumption that it's a system path and adds "-isystem".
> 
> Are you sure about this? I don't see this happening anywhere in the
> qmake source code. I do see some isystem related code around
> QMAKE_CFLAGS_ISYSTEM and INCLUDEPATH.

Take a look here:?https://github.com/qt/qtbase/blob/dev/qmake/generators/unix/unixmake2.cpp#L179
That's what I mean:
--------------------->8-------------------
? ? t << "####### Compiler, tools and options\n\n";
????t << "CC????????????= " << var("QMAKE_CC") << endl;
????t << "CXX???????????= " << var("QMAKE_CXX") << endl;
????t << "DEFINES???????= "
??????<< varGlue("PRL_EXPORT_DEFINES","-D"," -D"," ")
??????<< varGlue("DEFINES","-D"," -D","") << endl;
????t << "CFLAGS????????= " << var("QMAKE_CFLAGS") << " $(DEFINES)\n";
????t << "CXXFLAGS??????= " << var("QMAKE_CXXFLAGS") << " $(DEFINES)\n";
????t << "INCPATH???????=";
????{
????????QString isystem = var("QMAKE_CFLAGS_ISYSTEM");
????????const ProStringList &incs = project->values("INCLUDEPATH");
????????for(int i = 0; i < incs.size(); ++i) {
????????????const ProString &inc = incs.at(i);
????????????if (inc.isEmpty())
????????????????continue;

????????????if (!isystem.isEmpty() && isSystemInclude(inc.toQString()))
????????????????t << ' ' << isystem << ' ';
????????????else
????????????????t << " -I";
????????????t << escapeFilePath(inc);
????????}
????}
--------------------->8-------------------

For G++ "QMAKE_CFLAGS_ISYSTEM = -isystem", see?https://github.com/qt/qtbase/blob/dev/mkspecs/common/gcc-base.conf#L47

That's how include path gets prefixed with either normal "-I" if path has no matches in
QMAKE_DEFAULT_INCDIRS otherwise "-isystem " prefix is used.

> In any case, I believe this is where the problem is: I don't see why
> Qt5 should mess up with the addition of -isystem flags.

Agree, but see what Qt people say in similar bug reports:
https://bugreports.qt.io/browse/QTBUG-53367
https://bugreports.qt.io/browse/QTBUG-53375

And BTW that commit added all this mess:
http://code.qt.io/cgit/qt/qtbase.git/commit?id=0b144bc76a368ecc6c5c1121a1b51e888a0621ac

-Alexey