GitLab

Build failure debug tips

See merge request kde/kdesrc-build!12

This is supposed to help the user get started with debugging the build failure.

Such an ordering can be used to help the user getting started debugging failed builds.

Add missing dependency for Fedora 29+ source build failures

See merge request kde/kdesrc-build!10

The CI infrastructure started failing kdesrc-build after commit
864aa9d4, which I suspect is because the
CI builds and installs kdesrc-build without setting up a KDE buildsystem
(i.e. extra-cmake-modules).

This is supported in kdesrc-build, but the commit that broke the build
adds a reference a variable defined in ECM's KDEInstallDirs macro.

Fixed by adding to the existing compat layer machinery in the existing
CMakeLists.txt (at least, I hope!)

add screenshot borrowed from https://www.purinchu.net/wp/2018/04/14/fancy-status-updating-in-kdesrc-build/

Improves support for gpgme, libgpg-error and libdbusmenu-qt

Differential Revision: https://phabricator.kde.org/D22319

As David points out, this bit of advice ("see this page for list of
packages for your distro") is probably not that helpful for a user who
has surely run into having to find a single package on their distro
already.

BUG:409659
FIXED-IN:19.08

libgpg-error currently fails to update with kdesrc-build, ultimately due
to git-rebase complaining that there are local changes that need to be
stashed or reverted first.

These local changes are inherent to the libgpg-error build process, but
even so, kdesrc-build *does* check for local changes and stashes them
automatically.

But it turns out the current check from git of whether there are local
changes (using git-diff) gives a 'no changes' answer with libgpg-error's
modifications. So git-diff thinks there's no change, but git-rebase
thinks there is a change.

git-status --short gives an answer matching git-rebase, so switch to
that instead. git-status also gives answers for both the index and the
working tree at the same time to the updated code is even a bit shorter
now.

It is optional for module-sets to have names, but it's still a good idea
to *have* a name since kdesrc-build will try to provide you information
on which module-set a given module is being built from. If no name is
present then you'll get an ugly name for the set like
"Building qca from <module-set at line 34>".

As long as the user doesn't mind that's fine, but our base configuration
should set a higher standard.

This was the only anonymous module-set in the current base
configuration. Thanks to Laurent for the report.

BUG:409198
FIXED-IN:19.08

This was needed prior to CMake (and for some time as we were porting to
CMake) so that a wedged build system had at least a hope of eventually
being corrected.

This interferes with other use cases however and since CMake is really
good nowadays at generating valid build systems, remove this behavior.
Users can still pass --refresh-build when desired.

Fixes #30.

Differential Revision: https://phabricator.kde.org/D21649

It's been merged into kdepim-runtime and Volker says
the non-building libkolab doesn't build anymore

Summary:
Added '--needed' to the pacman call during initial setup. This will only
download and install missing packages, preventing unnecessary re-installs.

Reviewers: mpyne

Reviewed By: mpyne

Differential Revision: https://phabricator.kde.org/D21377

Summary: This makes the shell include work with the prebuilt Qt from qt.io, saving users with less powerful machines a lot of time.

Test Plan: applications can be compiled against the Qt binaries from qt.io and launch.

Reviewers: #kdesrc-build, mpyne

Reviewed By: #kdesrc-build, mpyne

Subscribers: mpyne

Differential Revision: https://phabricator.kde.org/D21387

Contributed by Jonah Brüchert.

Debian and its derivatives use paths like /usr/lib/$abi-triple, and
these /don't/ seem to also exist on non-Debian distros. So assume that
if a GNU ABI triple path is present that it should be used.

Differential Revision: https://phabricator.kde.org/D19864

Contributed by José JORGE of Mageia based on the existing Fedora package
list (and deliberately falling back to the Fedora install cmds which are
compatible).

I've tested myself by mounting the kdesrc-build directory into a Docker
container running the base Mageia 6 image and verifying that
"kdesrc-build --inital-setup" works. I have not verified that any
substantial packages build from there.

Differential Revision: https://phabricator.kde.org/D20936

New/updated config file options:

* 'configure-flags', reused as the way to pass cmdline options to the
meson setup command.
* 'ninja-options', a new option to pass cmdline options to the `ninja`
command. Note that ninja is mandated by Meson as the underlying build
tool.

Tested with https://github.com/plibither8/2048.cpp

Fixes #27, reviewed in !8.

Test suite passes and I continue to be able to build 2048.cpp. I've also
validated that ninja-options is passed to ninja when building 2048.cpp,
though this was a manual verification.

CCBUG:406268

This fixes #29, where kdesrc-build was aggressively prepending '/bin' to
the PATH if either kdedir or qtdir were unset (and qtdir is unset by
default). This isn't normally an issue but can conflict with user
settings.

Since the desire here is to setup Qt and/or KF5 before system packages,
only perform this prepending if Qt and/or KF5 are actually in use from
kdesrc-build. If those values aren't set then the system should already
be configured to provide in the environment so we shouldn't interfere.

It turns out that both ksb::Module and ksb::BuildContext are already
somewhat testable without a lot of orchestration needed here. I've
verified that the new tests fails without the patch applied, and pass
with this patch applied.

Differential Revision: https://phabricator.kde.org/D20174

See also issue: #26
Depends on Phabricator diff D20279: https://phabricator.kde.org/D20279

See also issue: #26
Depends on Phabricator diff D20279: https://phabricator.kde.org/D20279

When sorting back edges for dependency 'votes' in the new dependency
resolver, dfaure encountered Perl warnings to the effect of:

    > Argument "5/8" isn't numeric in numeric comparison (<=>) at
    > /kdesrc-build/modules/ksb/DependencyResolver.pm line 776.

This turns out to be because converting an entire hash object into a
scalar value doesn't just return the number of key/value pairs, but
instead returns a string describing metadata about the hash object's
content (see https://perlmaven.com/perl-hash-in-scalar-and-list-context)

While this doesn't actually break sorting due to the way Perl constructs
the string, it does cause warnings about using numeric comparisons on
strings.

Fix by applying `scalar` to just the list of keys in the hash containing
votes, which is just a way to determine the number of keys in the hash.