ClimateMachine.init() should not parse process ARGS by default?

[jakebolewski]

Description

Currently ClimateMachine parses command line arguments by default upon initialization: https://github.com/CliMA/ClimateMachine.jl/blob/master/src/Driver/Driver.jl#L212.

This has the side effect of exclusively claiming process global ARGS, which doesn't compose well when invoked in a programmatic / sub-procedure context (ex. it makes creating an alternative CLI program that invokes ClimateMachine awkward to write). One example of this behaviour is that it breaks running a subset of tests with the current test harness:

julia --project=@. -e 'using Pkg; Pkg.test(test_args=["Numerics"])

Starting tests for Numerics
Starting tests for Mesh
┌ Warning: Incompatibility detected between CUDA and LLVM 8.0+; disabling debug info emission for CUDA kernels
└ @ CUDAnative ~/.julia/packages/CUDAnative/hfulr/src/CUDAnative.jl:114
too many arguments
usage: <PROGRAM> [--disable-gpu] [--show-updates <interval>]
                 [--diagnostics <interval>] [--vtk <interval>]
                 [--monitor-timestep-duration <interval>]
                 [--monitor-courant-numbers <interval>]
                 [--checkpoint <interval>] [--checkpoint-keep-all]
                 [--checkpoint-at-end] [--checkpoint-dir <path>]
                 [--restart-from-num <number>] [--fix-rng-seed]
                 [--log-level <level>] [--output-dir <path>]
                 [--integration-testing]
ERROR: Package ClimateMachine errored during testing

Proposal

ClimateMachine.int() should be configured through keyword arguments and the public API should not depend on passing ArgParse data structures. Methods with overloads for simple cases like passing a dictionary of values could be added (ex. json / TOML processing). ClimateMachine.cli() would exhibit the current behaviour of parsing process ARGS and then internally invoking ClimateMachine.init().

For CLIMA Developers

• This feature can be added (if it cannot be, explain why in a comment below -- lack of technical expertise, not relevant to the scope of this project, too ambitious)
• There is a timeline for when this feature can be implemented
• The feature has been (or will be) implemented (Please link the PR)

[kpamnany]

Fair point. Will do.

[jakebolewski]

I don't mind creating a PR (if needed), just wanted to get feedback if this is a reasonable proposal as it breaks the current public API.

[kpamnany]

The reason it's done this way is to simplify the experiments which is really the point of the Driver. I'm not sure if that simplicity of use can be retained (do you see a way to do that?) but I don't especially like the current approach either.

[simonbyrne]

We really should move all the tests that use MPI (including those that call CLIMA.init()) to be run wrapped by runmpi: this would allow us to pass custom ARGS to each.

That said, I am still in favor of making the ARGS parsing at least somewhat optional.