The previous code using substituteInPlace was broken for both 5.9 and
5.11 (latter was broken after a minor release bump). Using patches
should make things fail loudly and prevent modules/qtwebengine.nix from
being littered with version checks.
The patch for 5.12 is only tested to apply.
This introduces usingAnnulenWebkitFork variable. Using version
check for qtbase.version is misleading because annulen's fork is not
tied to a specific qt5 release.
The impure library was introduced in
775531c1e3 ('qt58: extend darwin compatibility')
cmake-based build can use a non-apple ICU if -DMACOS_USE_SYSTEM_ICU=OFF
is set.
In Qt-5.12, the order of the dependency between these two packages
flipped.
A symptom of the problem is an error like, `module
"QtQuick.XmlListModel" is not installed`.
The upstream changes that this reflects are in qtxmlpatterns
<8c6e24329e>
and qtdeclarative <0477a057fd>
Undefined symbols for architecture x86_64:
"_OBJC_CLASS_$_NSData", referenced from:
objc-class-ref in qmacfunctions.o
"_OBJC_CLASS_$_NSMutableArray", referenced from:
objc-class-ref in qmactoolbar.o
objc-class-ref in qmactoolbardelegate.o
ld: symbol(s) not found for architecture x86_64
Why is qtmultimedia only needed on Darwin? Why does it only fix 5.9, not
5.11? These things I do not know. What I do know is that, for some
reason, this makes qt59.qtwebkit build on Darwin.
I think the reason it hasn't also fixed 5.11 might be something to do
with the version of qtmultimedia, but I don't know enough about Qt or
cmake to figure it out. The error when trying to build qt511.qtwebkit
(with or without these changes) is:
CMake Error at Source/cmake/OptionsQt.cmake:739 (find_package):
Could not find a package configuration file provided by "Qt5Multimedia"
(requested version 5.2.0) with any of the following names:
Qt5MultimediaConfig.cmake
qt5multimedia-config.cmake
Add the installation prefix of "Qt5Multimedia" to CMAKE_PREFIX_PATH or set
"Qt5Multimedia_DIR" to a directory containing one of the above files. If
"Qt5Multimedia" provides a separate development package or SDK, be sure it
has been installed.
Call Stack (most recent call first):
Source/cmake/WebKitCommon.cmake:50 (include)
CMakeLists.txt:137 (include)
-- Configuring incomplete, errors occurred!
See also "/tmp/nix-build-qtwebkit-5.212-alpha-01-26-2018.drv-0/source/build/CMakeFiles/CMakeOutput.log".
See also "/tmp/nix-build-qtwebkit-5.212-alpha-01-26-2018.drv-0/source/build/CMakeFiles/CMakeError.log".
b785d4813e introduced breakage in Qt
modules for 5.6 and 5.9, especially visible is Qt Webkit.
This was manifested by having a non-sensical build log where it is using
the top-level `src` attribute as source instead of Qt Webkit's own
source.
Were it not for the `src` top-level attribute (which is a legit
package), the error would have been made obvious by passing `null` to
`src`.
This partily reverts newly introduced way `src` can be passed to a
qtModule, instead relying on extending the `srcs` attrset.
For ZHF #45960
On aarch64, linking against the vendored ffmpeg fails. Including ffmpeg
as a dependency and passing -system-ffmpeg to qmake fixes this.
Slightly odd conditional in qmakeFlags to avoid altering the list on
non-arm platforms, so that the change doesn't trigger an unneccessary
rebuild.
This is the first time since 5.9 that we also update `qtwebkit`.
`qtwebkit` is not maintained by Qt anymore and thus, we switch to the
community port as for example arch has done. To prevent pulling in
single patches, we just stick to the latest git version.
This commit was too aggressive in adding configure flags. Only qtbsea
needs the -no-framework flag to work correctly. Qmake will handle
everything else for us.
Before this change:
$ readelf --notes /nix/store/zf5yja02g8n8dzgs25pqfd8w3myfzgzc-qtbase-5.10.1/lib/libQt5Core.so
Displaying notes found at file offset 0x004a7778 with length 0x00000020:
Owner Data size Description
GNU 0x00000010 NT_GNU_ABI_TAG (ABI version tag)
OS: Linux, ABI: 3.17.0
After:
$ readelf --notes /nix/store/sg1s9hdw0b7p6h0dwg09g4lxy1acq7y6-qtbase-5.10.1/lib/libQt5Core.so
Displaying notes found at file offset 0x004a7dcc with length 0x00000020:
Owner Data size Description
GNU 0x00000010 NT_GNU_ABI_TAG (ABI version tag)
OS: Linux, ABI: 2.6.28
-----------
The above paths were before rebasing the commit onto staging,
and it'd probably be good to have someone confirm the same happens
when built on a hydra builder or other non-dtzWill machine :).
[dezgeg: added comments]
Fixes#39334
and Fixes#36729
Also, Fixes#36932 (hopefully it's still building)
(This might be able to be able to be cherry-picked to 18.03 but be
careful - chromium is extremely touchy. Probably best to wait for it
to be built successfully on Hydra a few times.)
A dependency (boost) makes use of `std::auto_ptr`, which is no longer
supported in C++17 in Clang. This change re-enables `std::auto_ptr`
capabilities.
First, closure size is reduced by including the static libraries in $out instead
of trying to move them to $dev. The Qt build system cannot handle libraries
being split between different prefixes. Previously, the static libraries were
moved into $dev and the shared libraries were symlinked from $out to $dev to
fool the build system. However, this causes $dev to be retained at
runtime. Instead, we now keep the static libraries in $out. Fortunately, the
static libraries are not very large anyway.
Second, we build with QT_NO_DEBUG defined unless debugging is enabled. This
causes some assertions to be removed; when assertions are included, they pull
paths from $dev into the runtime closure by using the __FILE__ macro. We also
now patch qtbase to remove even more assertions when QT_NO_DEBUG is defined.
