[WIP] Conda Recipe for Linux

[HuidaeCho]

https://github.com/HuidaeCho/grass-conda

GRASS Conda Packaging

1. Install Miniconda

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
chmod a+x Miniconda3-latest-Linux-x86_64.sh
./Miniconda3-latest-Linux-x86_64.sh -b -u -p ~/opt/miniconda
~/opt/miniconda/bin/conda init
. ~/.bashrc

2. Setup Conda for GRASS build and packaging

conda config --add channels conda-forge
conda config --set channel_priority strict
conda create -n grass-conda conda-build git
conda activate grass-conda

3. Clone this repository

git clone https://github.com/HuidaeCho/grass-conda.git

4. Build and package GRASS

cd grass-conda
conda-build recipe

5. Find a new GRASS package

ls -al ~/opt/miniconda/envs/grass-conda/conda-bld/linux-64/grass-1.0.0-h3fd9d12_0.conda

6. Create a new test Conda environment and install it

conda create -n grass-conda-test
conda activate grass-conda-test
conda install grass -c ~/opt/miniconda/envs/grass-conda/conda-bld

7. See if GRASS is installed

ls -al ~/opt/miniconda/envs/grass-conda-test/bin/grass

[HuidaeCho]

https://github.com/OSGeo/grass/pull/3621

CMake & Conda TODOs:

• [ ] FHS for Python packages as well; Need to check @nilason's previous work
• [x] Fixing stdout/stderr on Windows
• [x] Complete Conda for Linux without FHS for now

[wenzeslaus]

I got mixed results here. After successful execution of the commands above, I get:

/root/opt/miniconda/envs/grass-conda-test/bin/grass --help
Traceback (most recent call last):
  File "/root/opt/miniconda/envs/grass-conda-test/bin/grass", line 2473, in <module>
    main()
    ~~~~^^
  File "/root/opt/miniconda/envs/grass-conda-test/bin/grass", line 2118, in main
    find_grass_python_package()
    ~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/root/opt/miniconda/envs/grass-conda-test/bin/grass", line 2106, in find_grass_python_package
    raise RuntimeError(msg)
RuntimeError: The grass Python package is missing. Is the installation of GRASS complete?

The setting PYTHONPATH doesn't help:

PYTHONPATH=/root/opt/miniconda/envs/grass-conda-test/lib/grass85/etc/python/ /root/opt/miniconda/envs/grass-conda-test/bin/grass --help

...although executing python command and import with PYTHONPATH works:

PYTHONPATH=/root/opt/miniconda/envs/grass-conda-test/lib/grass85/etc/python/ python
>>> import grass.script as gs

...but it fails later:

>>> gs.create_project("test")
/root/opt/miniconda/envs/grass-conda-test/lib/grass85/bin/g.gisenv: error while loading shared libraries: libgrass_gis.so.8: cannot open shared object file: No such file or directory
/root/opt/miniconda/envs/grass-conda-test/lib/grass85/bin/g.message: error while loading shared libraries: libgrass_gis.so.8: cannot open shared object file: No such file or directory

...suggesting that the automated env setup mechanism can't find /root/opt/miniconda/envs/grass-conda-test/lib/grass85/lib/libgrass_gis.so.8.

I can look at find_grass_python_package later. FHS would obviously help at least with the Python part.

[echoix]

I suspect the path on the build machine is hardcoded in that file like on Windows/OSGeo4W since July 7 https://github.com/OSGeo/grass/issues/6290

[wenzeslaus]

Indeed, the build machine (build time) path is the one which is used as GRASS_PYDIR, set to the following in my case:

/root/opt/miniconda/envs/grass-conda/conda-bld/grass_1758295620319/work/build/output/lib/grass85/etc/python

obtained by:

grep os.path.normpath /root/opt/miniconda/envs/grass-conda-test/bin/grass

[wenzeslaus]

With further investigation, I see that grass.py requires GRASS_PYDIR to be right, but does not take advantage of PYTHONPATH or otherwise correctly set up grass package (namely if FHS system puts with other Python packages). I also found that the build for the package did not replace build variables in the setup.py file under python/grass/script.

I opened two PRs:

• #6391 removes the need to fix the build variable substitution
• #6393 takes advantage of standard Python setups before trying GRASS_PYDIR (so here it helps to work around the problem, but does not fix anything since the conda package is not currently using FHS)

[HuidaeCho]

Please test the latest https://github.com/HuidaeCho/grass-conda. It builds and installs the non-FHS version in bin/grass and opt/grass on Linux. Running grass works for me.

[wenzeslaus]

The compilation now fails for me with:

mv: cannot stat '/root/opt/miniconda/envs/grass-conda/conda-bld/grass_1759321665374/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_p/lib64': No such file or directory

Details: https://github.com/HuidaeCho/grass-conda/issues/6

[wenzeslaus]

I tried the previous version (the one I tested before, that is https://github.com/HuidaeCho/grass-conda/commit/d6c353ef09ab06b9b4a4640b33ca0f41718237fa) and that builds as before and the changes from #6391 and #6393 helped some with the package (can be imported and run commands):

PYTHONPATH=/root/opt/miniconda/envs/grass-conda-test/lib/grass85/etc/python/ python                             

>>> import grass.script as gs
>>> gs.create_project("test")
>>> gs.create_project("test2", epsg=3358)
>>> gs.setup.init("test2")
<grass.script.setup.SessionHandle object at 0x749cc2b5c1a0>
>>> from grass.tools import Tools
>>> Tools().g_proj(flags="p")
...

...and with the executable package interface (same setup as the package):

PYTHONPATH=/root/opt/miniconda/envs/grass-conda-test/lib/grass85/etc/python/ python -m grass.app run g.region -p

...and with the main executable (no PYTHONPATH needed):

grass --help

I'm testing in a fresh Docker Ubuntu container (docker run --network host -it ubuntu:24.04).

As a side note, because that might be already fixed in the new version, these still fail:

$ grass --tmp-project XY --exec g.proj -p
FileNotFoundError: [Errno 2] No such file or directory: '/root/opt/miniconda/envs/grass-conda/conda-bld/grass_1759322494942/work/build/output/lib/grass85/etc/lock'
$ grass --config python-path
/root/opt/miniconda/envs/grass-conda/conda-bld/grass_1759322494942/work/build/output/lib/grass85/etc/python

[wenzeslaus]

With (recent) a92a3a018ace6833bc2f13cf0031ae2178d8e5ea (#6469) on main and (older) https://github.com/HuidaeCho/grass-conda/commit/d6c353ef09ab06b9b4a4640b33ca0f41718237fa, the following works:

grass --tmp-project XY --exec g.proj -p
grass --tmp-project EPSG:3358 --exec g.proj -p
grass run --crs XY g.proj -p
grass run --crs EPSG:3358 g.proj -p

Meaning the main GRASS executable is on path and it can find all its files (because of fallbacks implemented in #6391, #6393, and #6469).

The executable now also correctly returns path to Python packages:

$ grass --config python-path
/root/opt/miniconda/envs/grass-conda-test/lib/grass85/etc/python

So that can be used to set up Python path (PYTHONPATH or sys.path) without hardcoding it:

PYTHONPATH="$(grass --config python-path):$PYTHONPATH" python

import grass.script as gs
from grass.tools import Tools
gs.create_project("test1", crs="XY")
gs.create_project("test2", crs="EPSG:3358")
gs.setup.init("test2")
Tools().g_proj(flags="p")

[HuidaeCho]

PR for conda-forge: https://github.com/conda-forge/staged-recipes/pull/31234

[dudaka]

@HuidaeCho

I’ve submitted a PR to publish the GRASS GIS conda package for Linux, supporting Python 3.10–3.13:

🔗 PR: conda-forge/staged-recipes#31234 🔗 Source branch: dudaka/staged-recipes/tree/grass-linux-8.4.1 🔗 CI runs: Pipeline status & logs

All required checks have passed, and the CI pipeline successfully built the package.

Fixing OverLinkingError

The initial build failed during conda-build post-processing with an OverLinkingError, caused by several binaries linking to DSOs not declared in requirements.run:

lib/libgdal.so.37   → provided by libgdal-core
lib/libpdalcpp.so.19 → provided by libpdal-core
lib/libgomp.so.1     → provided by libgomp

Representative errors:

ERROR (grass,grass84/bin/g.proj): Needed DSO lib/libgdal.so.37 … not in reqs/run
ERROR (grass,grass84/bin/r.in.pdal): Needed DSO lib/libpdalcpp.so.19 … not in reqs/run
ERROR (grass,grass84/lib/libgrass_gpde.8.4.so): Needed DSO lib/libgomp.so.1 … not in reqs/run

Fix applied (in meta.yaml → requirements.run):

• ibgdal-core (for lib/libgdal.so.37)
• libpdal-core (for lib/libpdalcpp.so.19)
• libgomp (for lib/libgomp.so.1, OpenMP support)

These were added explicitly since the recipe currently ignores run_exports for GDAL/PDAL. Alternatively, ignore_run_exports could be adjusted to let their run_exports automatically include these split libraries.

Testing the recipe locally

The build-locally.py script used for local testing is available in the conda-forge/staged-recipes repository.

Method 1 — Using build-locally.py (recommended)

# From the root of staged-recipes
python build-locally.py          # choose linux_64.yaml interactively
# or directly:
python build-locally.py linux_64

• Runs inside a Docker container that replicates conda-forge’s CI environment.
• Build artifacts are saved under build_artifacts/

Method 2 — Using pixi (modern approach)

conda install -c conda-forge pixi
pixi run build-linux

Testing via a local channel

# From the repo root
conda create -y -n test-grass -c file://$PWD/build_artifacts -c conda-forge python=3.12 grass=8.4.1

# Basic runtime checks
conda run -n test-grass grass --version
conda run -n test-grass grass --tmp-project EPSG:4326 --exec g.version -rge

# Download sample data for validation
wget https://grass.osgeo.org/sampledata/north_carolina/nc_basic_spm_grass7.zip
unzip nc_basic_spm_grass7.zip
cd nc_basic_spm_grass7

# Run basic GRASS GIS commands
conda run -n test-grass grass --text user1 --exec g.region -p
conda run -n test-grass grass --text user1 --exec r.info elevation
conda run -n test-grass grass --text user1 --exec g.list type=raster

Next Steps

The maintainers will review the PR and may suggest refinements. Once approved and merged, a new GRASS feedstock repository will be automatically created. After acceptance, @HuidaeCho and I’ll help maintain the feedstock — handling version updates, dependency management, and future build fixes.

For platform coverage:

• macOS support will be added for the current 8.4.1 release using the same testing and CI validation methods described above.

• Windows support is planned for the next version (8.5), once environment and dependency adjustments are finalized.

These platforms can be tested using the same approach as above, but note:

• Linux builds use Docker containers (via build-locally.py).

• Windows and macOS builds run natively on the host machine since Docker-based builds are not supported there. The host environment must match the target platform (e.g., Windows host for Windows builds, macOS host for macOS builds).

[wenzeslaus]

The CI was passing on this 2 weeks ago already without a human looking at this, so perhaps polishing the PR and then contacting team is the right thing to do. The contacting etiquette is not completely clear to me. Anyway, I left some comments in the PR.

https://conda-forge.org/docs/maintainer/adding_pkgs/#feedback-and-revision https://conda-forge.org/docs/maintainer/maintainer_faq/#how-can-i-contact-conda-forgecore

[dudaka]

@wenzeslaus I started a discussion on Zulip on October 21 but haven’t received any response yet.

#general > Review request: GRASS GIS recipe PR (#31234)

There’s also another thread for other package, #general > ✔ PR that has been wating a review for a month, but that one has already been resolved.

[petrasovaa]

See https://conda-forge.org/docs/maintainer/infrastructure/#conda-forge-admin-please-ping-team, post the following for the bot:

@conda-forge-admin, please ping @conda-forge/help-python-c

[dudaka]

@petrasovaa Thanks for the link! That page is used after the recipe PR is merged to manage the feedstock. For now, we just wait for the staged-recipes PR to be approved. Once it's merged, we can use those tools.

I have posted another message to follow up on the zulip chat

[petrasovaa]

You need to tag the appropriate team in the PR, read the repo Readme, FAQ 12

@petrasovaa Thanks for the link! That page is used after the recipe PR is merged to manage the feedstock. For now, we just wait for the staged-recipes PR to be approved. Once it's merged, we can use those tools.

I have posted another message to follow up on the zulip chat

[dudaka]

@petrasovaa Thank you!

I posted the comment and the post responded https://github.com/conda-forge/staged-recipes/pull/31234#issuecomment-3469323983