CI: Install dvc via uv in the "GMT Tests" workflow #3695
Conversation
seisman
force-pushed
the
python-3.13-dvc
branch
2 times, most recently
from
c398edc to
aa06f58
Compare
seisman
changed the title
seisman
added
maintenance
| - name: Install dvc | ||
| run: | | ||
| uv venv | ||
| source .venv/bin/activate | ||
| uv pip install dvc | ||
| uv pip list | ||
|
|
||
| # Pull baseline image data from dvc remote (DAGsHub) | ||
| - name: Pull baseline image data from dvc remote | ||
| run: dvc pull --no-run-cache --verbose && ls -lhR pygmt/tests/baseline/ | ||
| run: | | ||
| source .venv/bin/activate | ||
| uv run dvc pull --no-run-cache --verbose && ls -lhR pygmt/tests/baseline/ |
There was a problem hiding this comment.
Maybe we can combine the "Install dvc" and "Pull baseline image data from dvc remote" steps into one step "Setup DVC and pull baseline image data from dvc remote", so we don't have to source .venv/bin/activate twice.
There was a problem hiding this comment.
Let's just keep this as is, since I'm assuming this will be a temporary workaround until conda-forge's dvc supports Python 3.13. We might want to revert the changes here after a few months, so best to have two separate steps.
| - name: Install dvc | ||
| run: | | ||
| uv venv | ||
| source .venv/bin/activate | ||
| uv pip install dvc | ||
| uv pip list | ||
|
|
||
| # Pull baseline image data from dvc remote (DAGsHub) | ||
| - name: Pull baseline image data from dvc remote | ||
| run: dvc pull --no-run-cache --verbose && ls -lhR pygmt/tests/baseline/ | ||
| run: | | ||
| source .venv/bin/activate | ||
| uv run dvc pull --no-run-cache --verbose && ls -lhR pygmt/tests/baseline/ |
There was a problem hiding this comment.
Let's just keep this as is, since I'm assuming this will be a temporary workaround until conda-forge's dvc supports Python 3.13. We might want to revert the changes here after a few months, so best to have two separate steps.
dvc is a complex package that has ~40 direct dependencies (https://github.com/iterative/dvc/blob/007041bfe953db08b832ddea2d3adeaeb74eec95/pyproject.toml#L33). The complicated dvc dependency tree sometimes causes failures or version conflicts e.g., #2468, #2338, #1544.
In #3490, we're trying to add the Python 3.13 support to PyGMT. That PR is stale because some dvc dependencies don't have 3.13 support yet (#3490 (comment)), and we may have to wait for a few more days or months.
Since dvc is a dev dependency, we don't have to wait for the conda-forge updates. This PR tries to install dvc in other ways. I've tried the following options:
pippipon Linux/macOS andchoco/wingeton WindowsuvIt turns out that installing dvc on Linux/macOS only takes ~10 seconds, but on Windows, options 1-3 take 60 seconds, except for option 4, which takes ~10 seconds on all platforms.
The CI job runs (https://github.com/GenericMappingTools/pygmt/actions/runs/12389452771/job/34582514750?pr=3695) for commit 83bb2d7 show that, with changes in this PR, we can continue the Python 3.13 support in PR #3490.