Allows dropping a workaround in qutebrowser. And yes, wrt -w, the line
we are changing literally has "Gentoo" in an upstream comment.
Also move qt6-build_src_prepare up given modifying files before
patches tend to be messy.
No revbump given use case is rather uncommon.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
ebuild needs more real refactoring (will do in a separate revbump),
but at least address pkgcheck's warnings in this one
wrt src_unpack (which triggers indirect git-r3 usage), not quite
sure what the debug check was doing (repeated) there, and (currently)
no patchset to unpack so can just drop the whole thing. For patchsets
this may be better handled in qt6-build given it is the one inheriting
git-r3.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
* fix building tests with gcc-13 (backport)
* set both TZ and LC_TIME needed by a few tests (p.m. vs pm)
* skip a few tests with extra needs for simplicity
* skip flaky tests depending on environment/compiler that do not
appear to be real issues (e.g. checks mounted filesystems)
* skip a few tests failing on x86 chroot that do not "seem" to have
consequences, needs more looking into but ultimately want more of
a baseline rather than pursue everything right now
* skip building a test with -Werror=strict-aliasing issues, not
important but saves tinderbox noise (ideally should fix, but well)
* skip tests broken without qtbase already being installed
(these also have issues with out-of-source build, but given
they are broken either way no reason to use IN_SOURCE=1)
* delete bunch of test binaries and files being installed
when tests are enabled (unfortunately not aware of a better
way, upstream may be installing first and using standalone
tests or so)
* set required use, could alternatively skip a few tests based
on USE but this would get further troublesome to maintain
* do not try to use docker-compose if found
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Albeit was almost tempted to just remove the USE as this
will likely cause more problems down the line.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
* use -j1 (similarly to when I tried to use python-xdist on
qutebrowser which uses qtwebengine, parallel tests don't
seem to work so well here either)
* skip tests that need network
* skip cert test, not looked too closely but likely missing
something or (maybe) expirations are causing issues
* export various variables to stop it from using the system's
qtwebengine (otherwise fails if not installed)
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Mostly had this figured out but the real issue is when
qtdeclarative is not installed, and this is a far worse
case than qttools or qtbase's few tests to skip.
QML imports are in general pretty messy to run from
tests as-is.
May revisit if the situation improves, albeit given believe
upstream installs before running tests it may not.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Insists on running tools from the system (fails if qttools
is not installed), and even if force the paths (using qt.conf,
or seds), it still has problem finding support files and most
tests fail. Getting all this to work properly would likely
be too messy.
There's also an annoyance with FetchContent, albeit it'd be
easy to handle.
wrt clang, qdoc seems to produce correct output with >=clang-16
but the ordering is slightly altered (believed to be due to
.cpp parsing being bit different) and cause some tests to
fail simple diff comparison -- may not be worth worrying about.
Originally was skipping a lot of tests, but it was utltimately
not meaningful anymore (aka tools were not really tested).
May revisit if the situation improves, albeit given believe
upstream installs before running tests it may not.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
seatv4 is perhaps due to newer libwayland, or the use of offscreen
rendering. But do not believe this would be used with any of the
mostly modern compositors we have packaged if I understand this
right, and so have not looked into this further (tst_seatv7 does pass).
Not fully looked into but the two others seem to be actually trying
to use a compositor (also opengl surfaces). We could run tinywl but
given the majority of other tests do not need that it does not seem
all that worth it (plus would likely be flaky).
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Attempts to import a test .dae file but that requires
collada which is hard-disabled in our system copy.
May need revisiting in assimp if someone needs this.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
If someone knows a sane way to stop Qt6 from installing
these (this happens with almost every components, but
eclass deletes tests/ by default at least), please tell me.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Imagine may be more that will need skipping. In my case one
test tried to use NVENC/CUDA but there may be more on systems
that support e.g. vaapi (unlike mine).
Unfortunately have not found a way to tell qtmultimedia to
use a dummy audio input/output wrt the other tests.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Broken test could be worked around with a symlink,
but feel it is not really worth extra ebuild churn.
wrt test files, if someone knows a sane way to stop Qt6 from
installing these (this happens with almost every components,
but eclass deletes tests/ by default at least), please tell me.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Misc minor adjustments / sorting, but more commonly:
* use same RDEPEND + DEPEND ordering everywhere, bit of both
order is used all across and sometime inverted in Qt5's so
let's match skel.ebuild to avoid confusion
* use explicit :6 slots, not necessary with =PV but this is more
about normalizing usage when multiple slots exist (plus *cough*
qa-vdb won't complaint anymore)
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
As-is, no changes in this commit which is mostly the same
as the current in-tree ebuilds.
Maintaining these in two different repos feels like just a hassle
(to me), and would rather have everything in one place so I can
change live and release ebuilds simultaneously as needed, plus not
have to sync metadata or eclass changes either (plus chiitoo has
::gentoo commit access now).
May move packages if I happen to work on them, albeit I have no
intention to really touch Qt5 or LXQt (anyone working on these
are free to the same if they want, or keep current workflow).
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Skip on qttranslation given it is the only that compiles
nothing, and so gives a unused option warning.
Closes: https://bugs.gentoo.org/911822
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Been using LTO by default when testing packages and, while
working on Qt6 test support, just kept running into more and
more components hitting test failures only with LTO.
Formerly added this directly in ebuilds, but at this point
feel this warrent a global filter until can figure out what's
wrong.
Technically needs revbumps but Qt 6.5.3 is not that far away
and that will take care of rebuilds.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
qtbase ebuild currently sets /mkspecs itself instead, chances are may
never need the variable but there's no need to have an odd one out
TESTSDIR was unused, and currently installs to /usr/tests which
we are deleting anyway -- and would give QA if that failed (note
changing the path can be messy given qtbase will keep telling
other packages to use old paths until it is itself updated).
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Qt5's may be a lost cause at this point wrt bug #457182, but can
do this fairly easily for Qt6 going forward (or at least for most
components, qtbase, qttools, and qtdeclarative are messier).
About src_install, it was defined but not exported. Not an issue
before but now we need it for cleanups.
Bug: https://bugs.gentoo.org/457182
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Essentially a wrapper to cmake.eclass and may as well advertise
the various things from it as usable (currently already using e.g.
"mycmakeargs" rather than some private "myconf" like qt5-build).
Should be no reasons to call phases directly, but CMAKE_SKIP_TESTS
and BUILD_DIR (rather than the removed QT6_BUILD_DIR) can be useful.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Currently unused given ebuilds were not wired for tests,
and hardly needed either way with QT_QPA_PLATFORM=offscreen.
If need arise, ebuild should inherit virtualx directly.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
All these are currently unused by both the eclass and consumers.
* IUSE=debug:
This was used by qt5-build but is no-op here. There's a commented
out debug feature in qtwebengine:6, but if that is ever used it
should just add its own IUSE.
* QT6_VERSION_MINOR:
Undocumented and assume may be a unused left-over from writing
the case ${PV}. Dropping saves a $(ver_cut) fork in global scope.
Could add properly if ever felt needed, but for now...
* QT6_BUILD_DIR:
cmake.eclass' BUILD_DIR should be used instead, this variable is
essentially just guessing what cmake will use and could be wrong
if an ebuild changes anything from cmake defaults.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
No functional changes here unless (at most) consumers were misusing
internals, so kept as a single commit for convenience.
Rough overview:
* streamline use of quotes
* minor descriptions adjustments
* replace MY_P with QT6_P (MY_* tend to be reserved for ebuilds),
and then unset it to avoid pkgcheck undocumented variables warnings
* use readonly keyword with assignments to avoid maintaining two
lists of variables
* re-order qt6_prepare_env, and keep separate blocks for QT6_*
variables referenced by other variables to avoid mis-ordering
* avoid calling $(get_libdir) twice for nothing by re-using LIBDIR
* rename internal qt6_prepare_env to _qt6-build_prepare_env
* swap SLOT from ver_cut to ${PV%.*}, both to be consistent with usage
in SRC_URI but also to optimize (less subshells in global scope)
Note have no intention to keep in-sync with qt5-build, considering
it legacy and Qt5 just needs to keep working until it can go
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
This is an upstream snapshot that does compile against openssl-3 but
also has known defects. Currently masked for testing.
Signed-off-by: Hans de Graaff <graaff@gentoo.org>