Suggestions for regression testing candidates

[penelopeysm]

#1087 (or #1090, which it got merged into) introduces a CI workflow to check that JuliaFormatter can format a given repo and that formatting is idempotent.

I'm looking for good test candidates for this. The only real criterion is that the repo must have large amounts of Julia code and should be at least somewhat syntactically "interesting" (to hopefully improve coverage). However, bonus points are awarded if your repo currently uses JuliaFormatter with a non-default style.

Please note that, for v2, there is still no guarantee that formatting is stable across versions. Formatted text may change across versions of JuliaFormatter as explained in the Project Status section of the docs. I'd like for this to eventually change, but there are still too many bugs for me to be able to commit to this.

[araujoms]

Not my package, but https://github.com/jump-dev/Hypatia.jl/ is pretty large, has "interesting" syntax, and uses JuliaFormatter with a non-default style.

[JoshuaLampert]

I and @svchb already mentioned this in another issue, but Trixi.jl (and TrixiParticles.jl?) is a good candidate, too. It's pretty big and uses the SciMLStyle. It currently still uses JuliaFormatter.jl v1, but once JuliaFormatter.jl v2 gives at least as good formatting as v1, we consider switching to v2.

[svchb]

yes TrixiParticles.jl currently uses 2.1.1 with SciMLStyle

[gdalle]

I switched DifferentiationInterface.jl (25k lines of Julia) from JuliaFormatter to Runic cause I was starting to struggle with JuliaFormatter v2 and I was fearing the discontinued maintenance. That being said, I like BlueStyle better than Runic so I wouldn't be against switching back. In any case, it's a large enough test

[dcelisgarza]

Not sure if it qualifies but my package uses non-default JuliaFormatter and has a variety of syntax and a lot of docs.

https://github.com/dcelisgarza/PortfolioOptimisers.jl/

[eschnett]

You are asking for large Julia code bases: For the CHORD radio telescope we are writing the low-level GPU kernels in Julia. Actually, there is a code-generator (written in Julia) which produces Julia code that is compiled to PTX. These Julia kernels are large (and repetitive), and we had to disable JuliaFormatter version 2 because it errored out (haven't checked recently).

Anyway, here is a pointer to one of the source files frb_chord_U128.jl. This is straight-line code that would be wrapped in a thin wrapper, then passed to @cuda.

Of course we don't need to format the code to generate PTX code. We DO need to format it so that we can inspect it; it is often easier to read than PTX code.

Thanks for your interest!

[haakon-e]

we're currently using juliaformatter v1 across most packages in the https://github.com/CliMA ecosystem (>200k LOC total). Recently, we're trying to adopt v2, see: CliMA/ClimaAtmos.jl#4572. This is our formatting file:

margin = 92
indent = 4
always_for_in = true
whitespace_typedefs = true
whitespace_ops_in_indices = true
join_lines_based_on_source = true
format_docstrings = true