Add missing dependency in pixi.toml

[flferretti]

lxml dependency was missing in the examples'pixi.toml

---

📚 Documentation preview 📚: https://jaxsim--144.org.readthedocs.build//144/

[flferretti]

Actually it would be smarter to unify the two pixi.lock, I'll port the modifications on the new file

fyi @traversaro

[flferretti]

For some reason cuda-cupti (removed in https://github.com/ami-iit/jaxsim/pull/138/files) is still missing from the resulting GPU environment, even if seems to be in the run_exports (https://github.com/conda-forge/cuda-cupti-feedstock/blob/0fa5abe8a5b6475b827c241ccb472c0f4093bff8/recipe/meta.yaml#L81-L93 and conda-forge/cuda-cupti-feedstock#10)

CUDA backend failed to initialize: Unable to load cuPTI. Is it installed? (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

[flferretti]

I needed to move some attribute from setup.cfg to the pyproject.toml as pixi was complaining that some attributes were missing. Apparently this breaks the CIs as it is looking for the dependencies in the pyproject.toml

fyi @diegoferigo

[flferretti]

A solution could be to use pixi for the CI, but this requires to setup 3 additional environments (py310, py311, py312), which would increase the size of the pixi.lock. See https://pixi.sh/latest/advanced/github_actions/

[traversaro]

I needed to move some attribute from setup.cfg to the pyproject.toml as pixi was complaining that some attributes were missing. Apparently this breaks the CIs as it is looking for the dependencies in the pyproject.toml

fyi @diegoferigo

If having the [project] prevents us to use the setup.cfg, then it is a bug in pixi I guess.

[traversaro]

For some reason cuda-cupti (removed in https://github.com/ami-iit/jaxsim/pull/138/files) is still missing from the resulting GPU environment, even if seems to be in the run_exports (https://github.com/conda-forge/cuda-cupti-feedstock/blob/0fa5abe8a5b6475b827c241ccb472c0f4093bff8/recipe/meta.yaml#L81-L93 and conda-forge/cuda-cupti-feedstock#10)

CUDA backend failed to initialize: Unable to load cuPTI. Is it installed? (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

I guess that if jaxlib==0.4.23=*_0 is present in the pixi.lock, pixi will not try to install a newer jaxlib==0.4.23=*_1 unless you delete the pixi.lock and recreate it.

[traversaro]

I needed to move some attribute from setup.cfg to the pyproject.toml as pixi was complaining that some attributes were missing. Apparently this breaks the CIs as it is looking for the dependencies in the pyproject.toml
fyi @diegoferigo

If having the [project] prevents us to use the setup.cfg, then it is a bug in pixi I guess.

I guess this is a regression due to prefix-dev/pixi#1112 .

[flferretti]

For some reason cuda-cupti (removed in https://github.com/ami-iit/jaxsim/pull/138/files) is still missing from the resulting GPU environment, even if seems to be in the run_exports (https://github.com/conda-forge/cuda-cupti-feedstock/blob/0fa5abe8a5b6475b827c241ccb472c0f4093bff8/recipe/meta.yaml#L81-L93 and conda-forge/cuda-cupti-feedstock#10)

CUDA backend failed to initialize: Unable to load cuPTI. Is it installed? (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

I guess that if jaxlib==0.4.23=*_0 is present in the pixi.lock, pixi will not try to install a newer jaxlib==0.4.23=*_1 unless you delete the pixi.lock and recreate it.

I tried with that, but I still get:

CUDA backend failed to initialize: Unable to use CUDA because of the following issues with CUDA components:
Unable to load cuPTI. Is it installed?
Traceback (most recent call last):
  File "/home/flferretti/git/jaxsim/.pixi/envs/test-gpu/lib/python3.12/site-packages/jax/_src/xla_bridge.py", line 322, in _version_check
    version = get_version()
              ^^^^^^^^^^^^^
RuntimeError: jaxlib/cuda/versions_helpers.cc:44: operation cuptiGetVersion(&version) failed: Unknown CUPTI error 999. This probably means that JAX was unable to load cupti.
 (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

[traversaro]

I tried with that, but I still get:

What is the version of jaxlib that gets installed in that case? If for any reason a version diffent from 0.4.23=*_1 is installed, we can just leave cupti for the time being (until at least a new jaxlib release).

[traversaro]

I needed to move some attribute from setup.cfg to the pyproject.toml as pixi was complaining that some attributes were missing. Apparently this breaks the CIs as it is looking for the dependencies in the pyproject.toml
fyi @diegoferigo

If having the [project] prevents us to use the setup.cfg, then it is a bug in pixi I guess.

I guess this is a regression due to prefix-dev/pixi#1112 .

I tried to leave almost everything in the setup.cfg with the following:

[project]
name = "jaxsim"
dynamic = ["version", "dependencies", "optional-dependencies", "urls", "classifiers", "keywords", "authors", "requires-python", "license", "readme"]

Do you think it could work?

[flferretti]

I tried with that, but I still get:

What is the version of jaxlib that gets installed in that case? If for any reason a version diffent from 0.4.23=*_1 is installed, we can just leave cupti for the time being (until at least a new jaxlib release).

Oh yes, it still installs jaxlib-0.4.23. Well then I'll add cupti again

[flferretti]

I tried to leave almost everything in the setup.cfg with the following:

[project]
name = "jaxsim"
dynamic = ["version", "dependencies", "optional-dependencies", "urls", "classifiers", "keywords", "authors", "requires-python", "license", "readme"]

Do you think it could work?

Yes, thanks a lot!

[traversaro]

I tried with that, but I still get:

What is the version of jaxlib that gets installed in that case? If for any reason a version diffent from 0.4.23=*_1 is installed, we can just leave cupti for the time being (until at least a new jaxlib release).

Oh yes, it still installs jaxlib-0.4.23. Well then I'll add cupti again

Which build number?

[flferretti]

Build 1

[diegoferigo]

@flferretti I though we agreed to wait a bit longer before merging to understand how to deal with the two lock files.

[flferretti]

Sorry Diego, apparently I misunderstood that. I merged since I haven't received any updates on this. Do you want to revert or I can open an issue to address this in the future

[diegoferigo]

Reverting a merged PR without leaving traces is not straightforward, let's keep it as it is this time. From now on, in these cases I'll be leaving empty blocking reviews to prevent other similar occurrences.