This adds the feature check code found on cppreference [1] and a Jenkins
pipeline to generate a feature status list for all platforms.
Aggregate a summary and upload it to docs.wildfiregames.com
[1] https://en.cppreference.com/w/cpp/experimental/feature_test.html
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The extra space is part of the name and prevents the build tarball to be
excluded from archived artifacts.
Reported-By: Itms <itms@wildfiregames.com>
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The --libdir switch got lost at some point, which is useful for running
tests with a system build.
Further allow to switch the type of output at runtime avoiding an
unnecessary rebuild.
Finally allow to specify an output file, this means there is no need to
redirect stdout which might break the CI in case the tests write
something to stdout like in case of LOGERROR resulting the test result
not being valid xml.
Ref: #7534
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Add a Dockerfile having all needed tooling. Using a container also means
those are the libs that will end up being used in the AppImage.
Then add a script for building pyrogenesis either form git or from
extracted source tarballs and for assembling the AppImage and finally
creating the AppImage.
Add building the AppImage to the "bundles" pipeline.
Also add a wrapper script to use podman to drive the whole process
for convenience for local builds.
Ref: #7577
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
wxWigets 3.0 does stuff that is deprecated in C++20, namely bitwise
operation between different enumeration types. wxWidgets 3.2 has
widespread support and in fact 3.3 was release recently so 3.4 won't be
that far off anymore either. Requiring 3.2 will allow us to do several
cleanups.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Drop reference build as it makes no sense for incremental builds, fix
warnings in test reporting and finally log build warnings per run which
prevents third partly library builds to leak into the report and failing
the CI.
Ref: #7367
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Make use of build matrix, drop reference build as it makes no sense for
incremental builds, fix warnings in test reporting and finally log build
warnings per run which prevents third partly library builds to leak into
the report and failing the CI.
Ref: #7367
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The build.log is created and found, but no warnings are collected.
Redirect stderr so the warnings end up in the log in the first place.
Fixes: #8159
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Extends the current Windows pipeline to also build for AMD64. While at
it use a matrix build.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The new patch-release pipeline takes a git branch originating from a
release tag, applies the changes since the tag onto the nightly build of
the release, and rebuilds the Windows binary.
All resulting changes are sent to the bundling pipeline which builds for
macOS and creates the tarballs and installers.
There will be two instances of the bundles pipeline: the main one will
keep tracking the main branch and providing biweekly bundles of the next
release, while a new patched-bundles pipeline will be manually used to
generate patch releases.
New files in the repository are naively added using `svn add --force`
which adds all unversioned files according to the documentation.
However, this iterates through the entire repository and takes about 12
minutes. Using `svn status` and parsing through the results only takes a
couple seconds.
The call to `svn delete` on deleted files does not work if filenames
contain spaces. This happens in particular when we fix spaces in
filenames, as the pipeline doesn't understand file renames.
Use the single Debian 12 container image and inject CC, CXX and LDFLAGS
instead.
Further do the lfs checkout on the agent instead of inside the container
avoiding full rebuilds due to version mismatch.
Then split the debug and release builds using separate workspaces and
record warnings by build preventing gcc warnings leaking into clang
warnings.
Based on zero warning policy and incremental builds using a reference
build doesn't make sense. Removing that we can also drop an additional
generic workspace.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Don't hardcode the number of parallel jobs in Windows based pipelines
but allow setting the desired number in the environment.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Allow the number of concurrent build jobs in Unix build pipelines be
specified in the environment via the JOBS variable.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The Linux agent now runs Docker in rootless mode for additional
security. Operations inside containers must be run as root to match the
underprivileged user running Docker.
The option --atlas is oddly different than any other feature flag,
change it to be in line with the others and enable it by default. The
new option is --without-atlas.
As the WX binaries aren't precompiled for Windows automagically disable
atlas there for users running the batchfile through filemanger if WX is
missing.
On *nix not providing a valid wx-config whith enabled atlas is
considered an error.
Fixes: #6401
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The migration to git broke the translation of language names in the
credits, as the translators.json, necessary for extracting the language
strings wasn't present anymore when running update_templates.py. This
commit should fix this.
This option is needed to be able to use more than 2GB on 32bit Windows
builds.
Having to explicitly enable this turns out to be a pitfall. Release
builds are done with this option enabled and when then building from
source and running out of memory it's not immediately obvious what the
reason is.
The documented intent is that this behaviour might be useful to find
memory leaks. The large-address-aware enabled build already perfectly
serves this role this days and the memory pressure is an important
reason for the ongoing Windows 64bit build efforts. Also there are
arguably better tools for this purpose.
Refs: commit a8403bd7e5
Refs: issue #1619
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
When a non-incremental build is needed for a PR, the current code cleans
up workspaces and retries, but this leaves the build error in the
console output. This marks the build as unstable, as if warnings were
present.
Instead, let the build fail and instead reschedule a clean build, which
will override the build status in Gitea.
This moves the nightly build to a new Windows autobuilder running a
recent Windows Server version. In this new environment, the Jenkins
agent runs as a normal user connected through SSH. This fixes#7148.
Use the opportunity to use svn directly instead of Jenkins' scm step,
which hangs for 30 min.
- Switched to CMake for configuring and building the doxygen Website.
- Set the minimum requirements to CMake >=3.18.4 and Doxygen >=1.9.1.
- Rewrite the LICENSE.txt and docs README.txt to markdown.
- Add the folder which is used in the README to gitignore.
- Add a custom style to the doxygen website based on the docs webpage.
- Update CI to use the new Doxygen Build process.
Closes#7052
- Generate entity.rng as part of the nightly build, refs #6976.
- Generate docs from a checkout of the nightly build.
- Publish them to docs.wildfiregames.com.
As compile.py only creates shaders for program combinations which don't
have already existing shaders on disk, this removes all SPIR-V shaders
prior to rebuilding them to ensure they actually do get recreated.
