search

SciML / ADTypes.jl

Repository for automatic differentiation backend types
https://sciml.github.io/ADTypes.jl/
MIT License
38 stars 11 forks source link

Downstream updates tracker for v1.0 #44

Open gdalle opened 5 months ago

gdalle commented 5 months ago

Which dependents must we upgrade to use ADTypes v1.0? Listed by topological order below:

I only tick the ones for which ADTypes v0.2 is no longer in the compat bound (see my comment below)

ChrisRackauckas commented 5 months ago
gdalle commented 5 months ago

Is it Christmas already?

ChrisRackauckas commented 5 months ago

No, it's a Monday. That the list that needs work to update 😅

gdalle commented 5 months ago

For most of those, the huge breaking change will be bumping Julia compat up to 1.10. If that is accepted, the rest is a bunch of renamings that I can try to contribute little by little

gdalle commented 5 months ago

Since I'm not a member of SciML, I can only create new PRs, not update those from CompatHelper though

ChrisRackauckas commented 5 months ago

Most should already be bumped to v1.10.

gdalle commented 4 months ago

I'm trying to do one of the harder upgrades with https://github.com/SciML/NonlinearSolve.jl/pull/426

Can someone make me a member of the NonlinearSolve repo, so that CI runs automatically on my commits? It's rather tedious to wait for maintainer approval. If you want I can also join the SciML organization if it makes things easier.

@avik-pal @Vaibhavdixit02 @ChrisRackauckas

gdalle commented 4 months ago

The CompatHelper PRs that were merged don't force ADTypes v1, since they still allow v0.2. As a result, most of them don't change anything, and ADTypes v0.2 is still the version picked by CI.

I think the right approach is to force ADTypes v1, so that we can see whether it breaks stuff or not. That's what I did in SparseDiffTools and NonlinearSolve.

However, a first step is to allow ADTypes non-exclusively everywhere so that downstream tests may run

ChrisRackauckas commented 4 months ago

Most cases have very small usage of ADTypes. NonlinearSolve, SparseDiffTools, and BoundaryValueDiffEq are the three that require a deeper dive.