[Bioc-devel] basilisk falling back to pip and breaking environment setup

[Aaron Lun]

Basilisk recently switched to a conda-less approach to setting up Python
environments, see discussion in:

- https://github.com/LTLA/basilisk/pull/49, you should have been mentioned
in the list of affected packages
- https://github.com/LTLA/basilisk/pull/52, which provides some guidance on
conda-only packages
- https://github.com/LTLA/basilisk/issues/48, for an overview of the
motivation
- https://github.com/LTLA/basilisk/issues/20, for a typical GLIBCXX error

As mentioned in #52, if your required Python dependencies are only present
in conda (igblast seems to be the main culprit), you can use the repurposed
basilisk.utils package to auto-provision a conda environment for you.

That said: whether this conda environment can be used successfully with
reticulate is another matter. It seems that, with the passage of time, the
conda binaries will increasingly depend on more recent versions of system
libraries. This will result in a greater rate of linker failures when
attempting to use reticulate to load the associated Python packages into an
R session linked against older versions of those libraries. It's probably
safest to just run the conda tools via system2, like what some packages do
for Python 2 environments (which aren't even supported by reticulate
anymore).

-A


On Wed, Jun 11, 2025 at 2:48 AM Rodrigo Arcoverde Cerveira <
rodrigo.arcoverdi using gmail.com> wrote:

> Hi basilisk team,
>
> First, thank you for your continued work on basilisk and for the fixes and
> improvements in recent versions. Also thanks to Vincent for the helpful
> replies to my previous questions.
>
> I’ve recently run into an issue where basilisk seems to fall back to using
> pip (via reticulate:::pip_install(python, packages)) instead of setting up
> the environment with conda, even when all packages are available via conda.
> This is happening in packages like scifer, MOFA2, and pirat, which
> previously built fine. Now, builds fail with errors when trying to install
> packages that are supposed to install via conda but fall on pip.
>
> The basilisk environment definitions still specify conda packages and
> channels (bioconda, conda-forge), and nothing has changed there. Has
> something changed in how basilisk sets up environments or handles fallbacks?
>
> Would appreciate any guidance on how to make sure conda is used as expecte
>
>
> Best,
> Rodrigo
>
> --
> Rodrigo Arcoverde Cerveira
> PhD student in Karolinska Institutet, Sweden
> Karin Loré’s research group
> CMM L8:00, Visionsgatan 18
>
>
>
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioc-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel
>

	[[alternative HTML version deleted]]