Use package manager for installation

[jentjr]

Any plans to use conda, or another packaging tool to distribute and install MODFLOW binaries? I would be willing to give this a shot to make the binaries available through conda-forge.

[langevin-usgs]

@jentjr, how would this work? Would the installer just pull them out of this repo? Or would it actually handle the compiling as well?

[jentjr]

@langevin-usgs, my thought was to create recipes using the conda-build compiler toolchains to compile for each platform from source. I’m not sure, but it may be possible to pull them directly from this repo.

[mwtoews]

Having looked at https://github.com/conda-forge/modflow6-feedstock for a few weeks, I can say that the conda-forge ecosystem is not that great, as it has old and unmaintained Fortran compilers. (Only Linux has modern compilers on conda-forge). Until there are better Fortran compilers, I don't think conda is a suitable platform to build and distribute Fortran-based applications for all platforms.

I'm planning to submit a PR for a simple Python script that could install executables using the Releases API, which would replace the manual "choose download", "click download", "unzip to folder" steps in a command-line fashion.

[wpbonelli]

@jentjr the get-modflow tool contributed by @mwtoews is included with FloPy 3.3.6 released today, there is more info here

[jentjr]

Great. Thank you!

[Hugovdberg]

With the advent of pixi it would be interesting to see if it could be viable to expose the executables via a conda feedstock. Pixi is a (very fast!) conda replacement with decent pypi integration, mostly based around conda-forge, but more importantly creates a lock file with the current versions that are installed in a project specific environment. This can then be committed to version control to allow reproducibility of conda environments (also cross platform!). Should these executables be installable via conda-forge this would allow projects to pin the exact set of binaries used, and then requiring only a pixi install to recreate the project with the same binaries.

It is already possible to pin the version in pixi using a task that calls get-modflow to download the executables which is then set as a dependency of the tasks that use the executables, but that feels a little hacky and requires to run that task at least once before the environment can be used.

[wpbonelli]

We have already begun to use pixi internally for MF6 development, indeed. And we have started to investigate putting MF6 on conda forge as well. I am not directly involved in the latter but AFAIK the main challenge is to find a compatible compiler in the conda ecosystem, the GCC available is too old and flang still seems too immature.

[Hugovdberg]

would it be possible to, instead of building on conda forge, to keep the current build train, but make a conda feedstock that exposes the precompiled binaries that currently reside on this repo? The conda build would then only need to download and extract the zip, and wrap it in a conda package such that they are automatically extracted to the correct location in the environment.

[wpbonelli]

@scharlton2

[scharlton2]

@Hugovdberg At first, I thought this might not pass the initial conda-forge submission due to potential malware risks. However, it's similar to the Panoply feedstock (https://github.com/conda-forge/panoply-feedstock), where we add a JDK dependency, download pre-compiled Java classes from NASA (https://www.giss.nasa.gov/tools/panoply/download), and create scripts to run the JAR files. So, I think it could possible. And since you're a member of conda-forge you may be able to approve the initial submission yourself.

[scharlton2]

UPDATE: The latest version of gfortran available on conda-forge (14.2.0) successfully compiles and runs the latest release of MODFLOW 6 (6.6.0) on Windows. However, the default Fortran compiler for Windows specified in conda-forge ({{ compiler('fortran') }}) is flang 5.0.0 (as defined in conda-forge-pinning-feedstock), which is unable to compile MODFLOW 6 (6.6.0).

The current conda-forge build of MODFLOW 6 can be found at modflow6-feedstock.