Can `Var` have an `id` field?

[cscherrer]

For code rewriting, it can be really useful to have more to distinguish a variable than just its name. For example, we might want to do things like "rename all occurrences of this x to y" or even "convert this AST to SSA style".

This could be very easy if JuliaVariables could be extended so Var has an additional id field. Do you have how difficult that would be to add? If it would be a major rewrite I could try to find other ways of getting to this.

[thautwarm]

It's possible, as SymRef should be its object identity. We can assign id to Var at

JuliaVariables.jl/src/JuliaVariables.jl

Line 479 in 67091c1

function from_symref(s::SymRef)

the same SymRef creates the same id in Var.

NOTE that the global variables and keyword arguments cannot be refactored by JuliaVariables.

[cscherrer]

Ok, I have at least a start on this. I'll PR soon to see what you think.

[cscherrer]

Hmm, maybe unique_id should be the same as name for globals, and we only increment the counter for local variables?

[thautwarm]

Sorry, SymRef is not the object identity of the resolved symbols. We should handle LocalVar and GlobalVar(Symbol).

I think your modification to local_var_to_var might not work for some cases. Same symbols in different scopes might result in different id.

Fortunately, LocalVar.is_mutable or LocalVar.is_shared are references, they can be the real object identity of the symbol. NameResolution.jl recognizes the same LocalVar for their different occurrences.

https://github.com/JuliaStaging/NameResolution.jl/blob/463ad09584ce823514a58bbc1391c6db51035a0b/src/Variable.jl#L2

You can use either is_mutable and is_shared as the reference. I'd suggest such a structure for Var:

struct Var
    name::Symbol
    unique_id:: Union{Ref{Bool}, Nothing}  # nothing when it's a global variable
    is_mutable::Bool
    is_shared::Bool
    is_global::Bool
end

[cscherrer]

Ok, I think I understand, but just to be sure... For two LocalVars, are the LocalVar.is_mutables the same (===) exactly when the LocalVar.is_shareds are the same?

[thautwarm]

@cscherrer Yes.. Sorry for such redundancy, Maybe I should make LocalVar a mutable struct then keeping a LocalVar is sufficent..

[cscherrer]

That would make sense, but it's not such a big deal either way. I think I'd refactor it this way if you think you'll be extending it more, since that could be easier to build on. But if it's basically done I'd maybe not worry about it. Up to you of course :)

[cscherrer]

Ok, I've updated the PR.

Also, what do you think of allowing another method passing an id_function? I've prototyped the idea here:
cscherrer#1