Working on Krita Dependencies ¶

Sometimes it happens that you need to patch/fix some of Krita dependencies. In this case you have to build your own version of the dependency instead of using the prebuilt package.

This manual describes the build process for the Linux host system using the docker environment. The process of building deps on Windows, MacOS and Android is almost the same. If the command differs on other systems, you will see a note about that inline. If you are not familiar with the docker-based build process, please check the docker environment manual: Building Krita with Docker on Linux

Opozorilo

Buillding the Linux deps on the host system without the docker environment is not supported and is not going to be supported. It is almost impossible to do.

Building the Android deps on the host system without using docker is possible, but still not supported. You will need to consult the Dockerfile for Kirta’s Android builder image to set up proper versions of the deps.

Building and modifying a single dependency ¶

Prerequisites ¶

The manual expects that you do already have a working docker environment with the prebuild deps installed. See Building Krita with Docker for details.

Setting up folders ¶

First you need to set up folders structure, initialize Python’s virtual environment and install python modules into that:

mkdir /home/appimage/appimage-workspace/root
cd /home/appimage/appimage-workspace/root
git clone https://invent.kde.org/dkazakov/krita-deps-management.git
git clone https://invent.kde.org/dkazakov/ci-utilities.git krita-deps-management/ci-utilities
python3 -m venv PythonEnv
source ./PythonEnv/bin/activate
pip install -r krita-deps-management/requirements.txt

If you used prebuilt deps, you might want to reuse the packages cache from the host system. Just create a symlink for it:

ln -s ~/persistent/deps/cache /home/appimage/appimage-workspace/root/cache

Setting up the environement ¶

Now you need to create the development environment:

python krita-deps-management/tools/setup-env.py -r ./ -v PythonEnv/ -s /home/appimage/appimage-workspace/deps/usr

This line sets the environment scripts in the current folder (-r ./) and configures the „shared install folder“ to coincide with the folder where prebuilt deps are preinstalled by the container initialization scripts (-s /home/appimage/appimage-workspace/deps/usr)

Opomba

If you are setting up the Android environment, pass --android-abi <targe-arg> option to this script (where <target-arg> is one of x86_64, armeabi-v7a or arm64-v8a)

The script has generated an environment file for you. Just source it now:

source /home/appimage/appimage-workspace/root/base-env

Next time you enter the docker container, just repeat this „source“ line and it will activate the environment for you.

If you are developing for Android or MacOS, you should also set up the toolchain file for the build:

# for Android
export KDECI_EXTRA_CMAKE_ARGS="-DCMAKE_TOOLCHAIN_FILE=/home/appimage/appimage-workspace/root/krita-deps-management/tools/android-toolchain.cmake"

# for MacOS
export KDECI_EXTRA_CMAKE_ARGS="-DCMAKE_TOOLCHAIN_FILE=/home/appimage/appimage-workspace/root/krita-deps-management/tools/macos-toolchain.cmake"

You should manually set this variable every time you enter the container. It is neceesary because the toolchain file for building Krita itself differs. If you need to build Krita in the same environment, you should change the toolchain file name to the one with „-krita“ suffix, i.e. android-toolchain-krita.cmake

Building the dependency ¶

To build a single dependency, just enter its subfolder and issue the build command:

cd krita-deps-management/ext_qt/
python -u ../ci-utilities/run-ci-build.py --project ext_qt --branch master --platform Linux -e env --skip-dependencies-fetch

Make sure you change the „project“ and „platform“ arguments of the command. You can try replacing --project ext_qt with --project $(basename $(pwd)) if you are on Linux.

The possible strings for „platform“ option are:

Linux
Windows
MacOS
Android/x86_64
Android/armeabi-v7a
Android/arm64-v8a

The script will (hopefully) build the dependency for you and leave you two files in the dependency subfolder:

krita-deps-management/ext_qt/env
krita-deps-management/ext_qt/env_deactivate

You can source the first file to activate environment for building this very subproject. When switching to another subproject, don’t forget to deactivate the environment with the second script.

Opomba

Sourcing env and base-env scripts will also set up two special variables for you that will help you determine which environment you are in:

KDECI_ENV_ACTIVATION_SCRIPT
KDECI_ENV_DEACTIVATION_SCRIPT

Hence, to deactivate the current environement you can just call:

source $KDECI_ENV_DEACTIVATION_SCRIPT

After the project is built at least once, you can edit its sources and do incremental rebuilds manually :

cd krita-deps-management/ext_qt

# activate the environment
source ./env

# go to the sources directory
pushd _build/ext_qt-prefix/src/ext_qt

## ... patch the sources as much as you need ...

# go to the build directory
popd
pushd _build/ext_qt-prefix/src/ext_qt-build

# build and install the project as usual
make -j8 install

# on Windows and MacOS you might need to use
# Ninja instead:
## ninja -j8 install

# after the work is done, deactivate the per-project environment
source ./env_deactivate

# alternatively, use the environment variable for that
# source $KDECI_ENV_DEACTIVATION_SCRIPT

Building all packages in isolated environments ¶

When testing dependency build scripts you might need to test if each package can be built independently in an isolated encironment. In such a case, the „shared install folder“ method will not work. Instead, you needs to use „publish to cache“ method. With this method, each subproject is built in an isolated environment, with only minimally necessary depenencies present. Each subproject is then packaged and „uploaded“ into the package cache. By the end of the build process the local cache is „forcefully“ populated with the locally built packages, which are newer than the ones in the registry. These local packages will later be used for building Krita itself.

This method works almost the same as the previous one, with a few minor differences.

Firstly, you need to skip installing the prebuilt dependencies when building the docker image. To do that, pass -s (or --skip-deps) option to build_image script:

./bin/build-image -s

This will create a container image without any prebuilt deps installed.

Then run the container and follow the folders setup process as described above in Setting up folders section.

When folders are set up, generate the envidonment without the -s option. It will disable the „shared install feature“:

python krita-deps-management/tools/setup-env.py -r ./ -v PythonEnv/
source /home/appimage/appimage-workspace/root/base-env

Opomba

If you are setting up the Android environment, don’t forget to pass --android-abi <targe-arg> option to the script (where <target-arg> is one of x86_64, armeabi-v7a or arm64-v8a)

Set up KDECI_EXTRA_CMAKE_ARGS if you are on MacOS or Android.

Make sure your pacakges cache is clean:

rm -rf /home/appimage/appimage-workspace/root/cache/*.{tar,json}

Run the build of all the packages:

cd krita-deps-management
python -u ../ci-utilities/seed-package-registry.py --platform Linux --seed-file latest/krita-deps.yml --publish-to-cache --missing-only

If some build has failed, fix it and then just rerun this command. Thanks to --missing-only the script will skip the subprojects that has been successfully built.

When the build process fails, you can manually rebuild the failing project by running run-ci-build.py script manually in the corresponding folder.

Troubleshooting ¶

Force verbose output during building ¶

When Ninja or GNU Make build your project, they usually hide the actual commands used to build a specific object file. Sometimes, especially when triaging bugs on CI, it can be helpful to see the exact build command lines. Usually it can be done by passing the following options to GNU Make and Ninja:

# for GNU Make
make VERBOSE=1 all

# for Ninja
ninja -v

However these commands work only when you start the build manually. On CI the build process is delegated to CMake or Meson, so we should pass the options to them instead:

# for CMake you just pass CMAKE_VERBOSE_MAKEFILE option,
# works with both, Ninja and GNU Make backends
cmake -DCMAKE_VERBOSE_MAKEFILE=ON ...
cmake --build . -j${SUBMAKE_JOBS}

# for Meson you need to pass `-v` option to the compilation command
meson setup ...
meson compile -v -C . -j${SUBMAKE_JOBS}

When working with Krita dependencies you can activate this options by any of the two methods:

1) Pass -DCMAKE_VERBOSE_MAKEFILE=ON to the toplevel cmake configuration call. Then all external projects will be built in verbose mode.

Set KRITACI_VERBOSE_MAKEFILE environment variable before running the toplevel CMake:
export KRITACI_VERBOSE_MAKEFILE=1

The latter approach is very convenient for fetching the debug output from the WebGUI of our CI system. Just start a job with this variable set and get the full output in the log.