Tests failing for prerelease

[isaacsas]

We need to investigate this and see if the issue is Catalyst related or not.

[ChrisRackauckas]

I wouldn't worry about it yet. I think it's a bug in the prerelease that is breaking some world age behavior due to some changes in globals. I'll need to talk with @oscardssmith about that.

[isaacsas]

Ok, thanks for letting us know.

[oscardssmith]

yeah it is a change in globals. 1.12 got a tiny bit stricter in ways that pretty much only affect code that was slightly weird/broken which given the amount of code we have, it isn't exactly surprising that we have some of it.

[ChrisRackauckas]

Some of it could be things we need to change, but other things may just be accidental breaking changes in the Julia compiler that we need to notify get fixed during the beta stage 😅 . You can be the arbiter of which is which. Let's walk through that in one of the Monday calls since I assume some of these might need discussion (I see RuntimeGeneratedFunctions popping up)

[ChrisRackauckas]

Let's just keep this to the issue SciML/RuntimeGeneratedFunctions.jl#95 (comment)

[ChrisRackauckas]

It looks like we know what the RGF one is, but the Catalyst one looks like it's just incorrect macro code.

Catalyst.jl/src/reactionsystem.jl

Lines 1187 to 1188 in 005799c

	ex = Catalyst.__unpacksys($(esc(rn)))
	Base.eval($(__module__), ex)

You should never use eval in a macro. The rule is that if you use eval in a macro is probably wrong 😅.

[isaacsas]

That was an internal hack to allow unpacking all the Symbolics in a model, and was never exported or advertised/made api. It really isn’t needed anymore now that Symbol-based indexing generally works, so we should just remove it.

[ChrisRackauckas]

That would be best, because it seems like it's just not going to be easy to support that in Julia v1.12 as it's doing some odd writing to modules at eval time. Working around it seems like a bit more work and is likely not very reliable.