Resolve can be slow for specific packages

[fecet]

Checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pixi, using pixi --version.

Reproducible example

name: default
channels:
- conda-forge
- nodefaults
dependencies:
- cmake *
- compilers *
- gcc ==11.4
- pyspark *

save this as environment.yml, then solve it with

pixi init --import environment.yml
pixi project export conda-explicit-spec . -vvvv

Issue description

Solve takes a long time (90s on my machine), as a comparsion, conda-lock takes 55s

Expected behavior

The solve should not be that slow, or at least as fast as conda-lock

[fecet]

@synapticarbors from #1873 to here, maybe related to mamba-org/resolvo#49.

[ruben-arts]

Thanks for the information! This was merged some time ago which has resulted in better results in some cases but we're now seeing some edge cases where it got much slower.

FYI @baszalmstra

Tip

While we're working on improving the speed. You can always add more constrained requirements to help the solver avoid unwanted paths in the traversal of the solution space. A good one to always add is the minimal version of python and in this specific case you can probably also limit cmake.

[lambdaxdotx]

I can confirm that upgrading to 0.30.0 makes the resolving conda stage of a pixi update way, way slower than before. It really takes ages.

Here is my pyproject.toml just edited a little bit to remove private info:

$ cat pyproject.toml 
[project]
authors = [...]
requires-python = ">= 3.10"

[build-system]
build-backend = "hatchling.build"
requires = ["hatchling"]

[tool.pixi.project]
channels = ["conda-forge", "nvidia", "pytorch", "pyg"]
platforms = ["linux-64"]

[tool.pixi.pypi-dependencies]

[tool.pixi.tasks]

[tool.pixi.dependencies]
python = ">=3.10,<3.11"
black = ">=24.8.0,<25"
tqdm = ">=4.64.1,<5"
gitpython = ">=3.1.27,<4"
matplotlib = ">=3.9.2,<4"
seaborn = ">=0.13.2,<0.14"
scikit-learn = ">=1.5.1,<2"
imbalanced-learn = ">=0.12.3,<0.13"
termcolor = ">=2.4.0,<3"
numpy = ">=1.23.2,<2"
pandas = ">=1.4.4,<2"
xgboost = ">=1.6.2,<1.7.0"
pre-commit = ">=3.8.0,<4"
tensorflow = ">=2.9.1,<3"
keras = ">=2.9.0,<3"
clang-11 = ">=11.1.0,<12"
pytorch = { version = ">=2.2.1,<2.3", channel = "pytorch" }
pyg = { version = ">=2.5.1,<2.6", channel = "pyg" }
pytorch-sparse = { version = ">=0.6.18,<0.7", channel = "pyg" }
pytorch-scatter = { version = ">=2.1.2,<2.2", channel = "pyg" }
pytorch-cluster = { version = ">=1.6.3,<1.7", channel = "pyg" }
huggingface_hub = ">=0.21.4,<0.22"
transformers = ">=4.40.1,<4.41"
wandb = ">=0.17.3,<0.18"

HTH

[0xbe7a]

This takes 40minutes to solve with 0.30.0

[project]
channels = ["conda-forge"]
name = "slow-pixi-solve"
platforms = ["osx-arm64"]

[dependencies]
ipython = "*"
awscli = ">=2"

[ruben-arts]

This was fixed in #2162, Thanks for providing examples that helped a lot!