istree on Expr

[shashi]

should not be true in my opinion because terms may want to hold Exprs. As we discussed, MT could use _istree(x) = istree(x) || x isa Expr...

cc @0x0f0f0f

[0x0f0f0f]

I think that this issue generalizes to a broader scope. For example, I've encountered a similar problem when fixing a metatheory.jl test case regarding rewriting categorical expressions. I wanted to skip the walk of nodes with a certain specific operation using Rewriters.Postwalk (this can be reduced/converted intuitively to the case of an istree(::T) type), treating those nodes like terminals but only in a certain moment and scope. I still wanted to treat those nodes as trees during other tasks. Maybe a finer degree of control over istree is needed. Wrapping all the MT calls to TermInterface.isexpr seems like a bad hardcoding choice, because if you want to rewrite Symbolics expressions that hold Exprs as non-tree terminals MT would still treat them as trees without any degree of control. Lets chat about potential solutions that are not orthogonal to either packages design. I'm sure there is one that is compatible with both SU being specific on types and MT being as general as possible

cc @ChrisRackauckas and @philzook58 s for any suggestions

[0x0f0f0f]

I would close this since the integration worked out well