With the update to Qt 5.9.3, the configure script and qmake expect an mkspec
name, not an absolute path, which is why r458293 switched to using
${QMAKESPEC:T}.
However, the :T modifier breaks things when CXX is set to an absolute path
instead of just "c++", QMAKE_COMPILER is a shell string that will be evaluated
only after make invokes `configure' with CONFIGURE_ARGS. In other words, we end
up turning something like
$$(ccver="$$(/usr/bin/c++ --version)"; case "$$ccver" in *clang*) echo clang ;; *) echo g++ ;; esac)
into
/c++ --version)"; case "$$ccver" in *clang*) echo clang ;; *) echo g++ ;; esac)
which is obviously invalid.
We now just avoid being too smart and set a separate variable called
QMAKESPECNAME, which contains only the mkspec name and that we use both when
setting CONFIGURE_ARGS as well as to create QMAKESPEC.
PR: 224971
qtmultimedia now uses a configure.json file to describe configuration options
and checks that qmake should perform. On the one hand it means
extrapatch-no-gstreamer no longer applies (and neither does the TBR_DEPENDS
hack in the Makefile), on the other the configuration process has been
streamlined: we only need to pass the right options via QMAKE_CONFIGURE_ARGS to
enable and disable options.
While here, stop setting QT_DEFINES altogether in the Makefile, as none of them
are really necessary at all:
- XVIDEO is a Qt4 thing;
- ALSA, OPENAL and PULSEAUDIO are handled by qmake's configure.json
infrastructure, which sets a QT_NO_<OPTION> macro in qtmultimedia-config.h
when they are not enabled.
- There is no QT_{NO_}GSTREAMER upstream, so we're basically defining some
macros that no code is going to use.
Reviewed by: tcberner (earler version without the QT_DEFINES changes)
- Sort fts traversal in repo creation
- Fix floating pint exception when packaging packages with weird elf files
- More test cases for config merge
- Fix running the testsuite on linux and osx
- Fix pkg version regarding flavors
- Add support for creating packages with a hash in their names
- Accept DT_SONAME anywhere in .dynamic
- make pkg version -r imply -R
- Some fixes for the progression display
- Rename some variables in security periodic scripts.
- Parse FreeBSD OS version and forbid to install too new packages
before each commit and I'll whether I can create a pre-commit script that
checks this for me.
Reported by: Walter Schwarzenfeld <w.schwarzenfeld@utanet.at>
Approved by: antoine (implicit)
This took quite a lot of time because Qt's own build system underwent
several changes in 5.8.0 that took a while to adapt to.
And, of course, qt5-webengine is a behemoth that we need to patch like crazy
due to its bundling of Chromium. In fact, most of the Chromium patches in
qt5-webengine have been imported with no changes from www/chromium@433510
("www/chromium: update to 56.0.2924.87").
New port: accessibility/qt5-speech
Bigger changes to Qt5 ports we had to make:
- Qt now allows using a configure.json file to define configuration options
and specify configuration checks that can be done when qmake is invoked.
However, configure.json checks done in a subdirectory only propagates to
subdirectories, and checks elsewhere will fail if all .pro files are being
parsed at once (i.e. qmake -recursive), so several ports had to switch to
USES=qmake:norecursive along with manual additional qmake invocations in
subdirectories in order to work. It's been mentioned in a few places such
as Qt's bug tracker that qmake's recursive mode is pretty much deprecated,
so we might switch to non-recursive mode by default in the future.
- Uses/qmake.mk: Introduce QMAKE_CONFIGURE_ARGS. qmake now accepts
arbitrary options such as '-foo' and '-no-bar' at the end of the
command-line. They can be specified in QMAKE_CONFIGURE_ARGS.
- graphics/qt5-wayland: The port can only be built if graphics/mesa-libs is
built with the WAYLAND option, so a corresponding option (off by default)
was added to the port.
- misc/qt5-doc: Switch to a pre-built documentation tarball. The existing
port was not working with Qt 5.9. Instead of trying to fix it, switch to
what Gentoo does and fetch a tarball that already contains all
documentation so that we do not have to build anything at all. The
tarball's name and location in download.qt.io look a bit weird, but it
seems to work fine.
- www/qt5-webengine: Use binutils from ports, Chromium's GN build system
generates a build.ninja that uses ar(1) with the @file syntax that is not
supported by BSD ar, so we need to use GNU ar from binutils.
- x11-toolkits/qt5-declarative-render2d: This port was merged into the main
Qt Declarative repository upstream, and into x11-toolkits/qt5-quick in the
ports tree.
Changes to other ports we had to make:
- biology/ugene: Drop a '#define point "."' that is not present in more
recent versions of the port. Defining a macro with such a common name
causes build issues with Qt 5.9, which uses |point| as an argument name in
methods.
- cad/qelectrotech: Fix plist with Qt 5.9. Directories are no longer
installed with `cp -f -R', but rather `qmake install qinstall', which does
not install
%%DATADIR%%/elements/10_electric/20_manufacturers_articles/bosch_rexroth/.directory
That's a local file that should not even have been part of the tarball
anyway.
- chinese/gcin-qt5: Add additional private Qt directories (which should not
be used in the first place) to get the port to build with Qt 5.9.
- devel/qtcreator: Fix plist with Qt 5.9. Something changed in qdoc and some
test classes no longer generate documentation files.
- security/keepassx-devel: Import a patch sent upstream almost a year ago to
fix the build with Qt 5.9.
Thanks to antoine for the exp-run, and tcberner and Laurent Cimon
<laurent@nuxi.ca> for landing changes in our qt-5.9 branch.
PR: 224849
This release adds support for sleeping on condition variables (and
mutexes and rwlocks) with relative timeouts. This functionality is
needed by the recently added pthread_cond_timedwait_relative_np() and
some other functions.
Reviewed by: bapt
Differential Revision: https://reviews.freebsd.org/D13786
