search

SciML / Catalyst.jl

Chemical reaction network and systems biology interface for scientific machine learning (SciML). High performance, GPU-parallelized, and O(1) solvers in open source software.
https://docs.sciml.ai/Catalyst/stable/
Other
460 stars 77 forks source link

DO NOT MERGE - test master #1056

Closed isaacsas closed 1 week ago

isaacsas commented 1 week ago

@TorkelE tests seem to be failing from a test that is relying on Symbolics.jl internals.

isaacsas commented 1 week ago

Looking at this more carefully it seems to me like this is a bug and that variables arising from observables are not being created correctly by the DSL. They should be completely equivalent to a normal variable, even generating the same code when using @macroexpand, but when I look at what the DSL generates it is not producing the same thing.

TorkelE commented 1 week ago

Yeah, there is a comment on it

# Declares observables implicitly/explicitly.
# Cannot test `isequal(rn1, rn2)` because the two sets of observables have some obscure Symbolics
# substructure that is different.

Last time when i had a look at it, I could not figure it out though, e.g.

isequal(rn1.X, rn2.X) # false
isequal(Symbolics.unwrap(rn1.X), Symbolics.unwrap(rn2.X)) # false
isequal(Symbolics.unwrap(rn1.X).metadata, Symbolics.unwrap(rn2.X).metadata) # true

However, the remaining fields checked with

println(fieldnames(typeof(Symbolics.unwrap(rn1.X))))

is nothing like normal Julia types

(:metadata, Symbol("###simplified###2"), Symbol("###Any###3"), Symbol("###Any###4"), Symbol("###Any###5"), Symbol("###Any###6"), Symbol("###Any###7"), Symbol("###tag###8"))

and I'd really need a Symbolics expert to explain to me what is going on and what these weird stuff are.