Basilisk forces the use of potentially incompatible Conda binaries

[rekado]

You're using Basilisk to ensure that a suitable Python environment exists in which reticulate can use a number of Python packages. The problem with Basilisk is that it forces the use of Conda, which downloads binaries. These binaries have been linked with libraries that may or may not have a different ABI than those used on the user's system. For example, when using Voltron with Guix the C library very likely differs from what has been linked with the Conda binaries.

It seems to me that the use of Basilisk is not actually required if the user can guarantee a suitable Python environment. In Guix we could take care of installing the required Python packages.

While Basilisk could still be a required package, perhaps its runtime use could be toggled with a configuration option or an environment variable. Would you be open to a pull request that accomplishes this?

[Artur-man]

I forgot to tag the commit but we have the dev branch allowing to specify a path to python (e.g. as.AnnData) which runs reticulate using a certain python executable if exists.

VoltRon/R/conversion.R

Lines 265 to 272 in 99ae8b7

	as.AnnData <- function(object,
	file,
	assay = NULL,
	flip_coordinates = FALSE,
	method = "anndata",
	create.ometiff = FALSE,
	python.path = NULL,
	...) {

The environment necessary for installing the default environment that basilisk is also using can be found at environment.yml.

[rekado]

Is there a single location where a default for python.path could be specified? Or would I need to patch every call of any procedure that might use reticulate?

[Artur-man]

oh you mean if you need to specify the path everytime you call a method that use reticulate ?

Perhaps you may need something like:

options(voltron.python.path = "/path/to/python")
as.AnnData(xenium_data)

as opposed to

as.AnnData(xenium_data, python.path = "/path/to/python")

[Artur-man]

Ok now you have now two options:

data("xenium_data")
h5ad_file <- tempfile(fileext = ".h5ad")
as.AnnData(xenium_data, file = h5ad_file, python.path = "path/to/python")

or

data("xenium_data")
h5ad_file <- tempfile(fileext = ".h5ad")
options(voltron.python.path = "path/to/python")
as.AnnData(xenium_data, file = h5ad_file)

You can run both chunks above,xenium_data is an example data embedded in the package.