[Feature] Gather more feedback to document support for conda-forge environments

[tombisho]

Describe the bug
When operating in a TRE environment with access to conda-forge, acro_init fails because uv cannot be downloaded (it is blocked by the firewall)

To Reproduce
In a TRE environment, follow the install guide using conda.

Expected behavior
acro to be initialised

Screenshots

Desktop (please complete the following information):
TRE with access to conda-forge

[tombisho]

I have made some progress by installing uv in my r-acro conda environment. However this leads to the next stumbling block because acro tries to install things using pip. Our TRE doesn't allow access to PyPi so this all fails:

[rpreen]

Currently, the R helper function acro_init() calls reticulate::virtualenv_create() and reticulate::py_install() behind the scenes. Thus it requires pip installation and wont work with conda.

To support using conda-forge, ACRO-R needs to add a parameter to acro_init() that switches between PyPI and conda; if using conda then it could instead use the conda tools reticulate::conda_create() and reticulate::conda_install().

---

As an immediate workaround, you should also be able to do this by hand - that is do not call acro_init() but use the reticulate functions to setup a conda environment and install acro.

I have not tested this yet, but something along the lines of the following from R:

# Create a conda environment with a specific Python version
reticulate::conda_create(envname = "r-acro-0.4.11", packages = "python=3.12", channel = "conda-forge")

# Install the Python ACRO backend
reticulate::conda_install(envname = "r-acro-0.4.11", packages = "acro==0.4.11", channel = "conda-forge")

# Activate the environment
reticulate::use_condaenv("r-acro-0.4.11", required = TRUE)

# Import and use ACRO (also set whether you want automatic suppression)
acro <- reticulate::import("acro", delay_load = TRUE)
acroEnv$ac <- acro$ACRO(suppress = FALSE)

After this, the helper functions (like acro_lm() should work as normal.

[jim-smith]

@rpreen to get the R CMD check to run in preparation for submission of v0.4.16 to CRAN I had to change the name of the vent created from r-acro-0.4.11 to the more generic r-acro, otherwise it was having endless problems failing to find acro when it ran the test suite.

[jim-smith]

@tombisho which TRE is that please? I'm wondering if you have local cached copies of pypi and conda, in which case the pip installer may just need to be pointed at those rather than trying to access the 'main' version on the web.

[tombisho]

@rpreen thank you for these pointers and for adding conda support as a potential feature. I will give the workaround a try.

@jim-smith this is for the Cambridge Univeristy Secure Compute Research Platform (https://www.hpc.cam.ac.uk/secure-research-computing). I would be very happy if we did have pypi and conda mirrors with local copies but unfortunately this is not currently available. Instead we have conda-forge (and CRAN) whitelisted as these repositories do not allow packages (and hence data) to be uploaded. Pypi allows you to develop a package and immediately upload this (and hence data) so it is not whitelisted. We do have many Python packages available as modules that can be loaded, and new packages can be requested to be made available but this is a manual process.

[tombisho]

Thank you for the helpful suggestion to bypass acro_init() with the steps described above. Unfortunately it results in an error about scipy not being able to find the libstdc++ library.

I have tried various other things such as creating the conda environment outside R and then activating it within R, all resulting in the same error. This is such a shame as I had no problems at all with a previous version of acro and this has stopped our testing.

I'll do a little more experimentation but I have already spent several days on this so I will need to put it to one side soon.

[jim-smith]

@tombisho thanks for the further feedback, Richard is on leave but I will try to pick this up later this week or next and get back to you properly