work around FastBroadcast bugfix - Githubissues

SciML / OrdinaryDiffEq.jl

High performance ordinary differential equation (ODE) and differential-algebraic equation (DAE) solvers, including neural ordinary differential equations (neural ODEs) and scientific machine learning (SciML)

https://diffeq.sciml.ai/latest/

Other

521 stars 198 forks source link

work around FastBroadcast bugfix #2243

Closed oscardssmith closed 3 weeks ago

oscardssmith commented 3 weeks ago

hopefully this fixes CI. tests currently in progress locally.

Checklist

[ ] Appropriate tests were added
[ ] Any code changes were done in a way that does not break public API
[ ] All documentation related to code changes were updated
[ ] The new code follows the contributor guidelines, in particular the SciML Style Guide and COLPRAC.
[ ] Any new documentation only uses public API

Additional context

Add any other context about the problem here.

oscardssmith commented 3 weeks ago

oh, things are far more broken than we thought. The problem seems to be real and the fact that it broke when FastBroadcast changed seems to be entirely coincidental????

oscardssmith commented 3 weeks ago

So the problem is that D[:,i] is a length 8 vector, while u is a 4x2 matrix. So presumably at some time D changed from being matrix like to vector like but only in the in place version??

oscardssmith commented 3 weeks ago

Oh, this is very funny. version 0.3.0 of Fastbroadcast fixed a bug that this code was relying on. Previously it was allowing broadcasts between 4x2 matrices and length 8 vectors even though that isn't supposed to work.

oscardssmith commented 3 weeks ago

well this was very dumb. I think it should be fixed now.

oscardssmith commented 3 weeks ago

Let's see if https://github.com/YingboMa/FastBroadcast.jl/pull/62 fixed this

oscardssmith commented 3 weeks ago

Local testing shows it did, but I closed and opened before the registry was actually updated.