PossibleContents is not a lattice

[tlively]

IIUC, we have this structure:

             Type(i32)
            /         \
Global("g1", i32) Global("g2", i32)
           |     X     |
       Literal(1) Literal(2)

There is no unique least upper bound of the two literals and there is no unique greatest lower bound of the two globals. This is a problem because union and intersect are supposed to calculate unique least upper / greatest lower bounds.

I think there is a probably a way to solve this, but I'll have to think on it.

I ran into this while trying to simplify and complete the implementation of intersect as part of the investigation suggested here: https://github.com/WebAssembly/binaryen/pull/7538/files#r2054860951

[kripken]

Isn't it

                             Type(i32)
            /             /               \       \
Global("g1", i32) Global("g2", i32)  Literal(1) Literal(2)
            \             \               /       /
                               None

?

The union of a global and a literal is neither a global or a literal, it is an unknown thing of the shared type, and the intersection of a global and a literal is empty (same as unioning/intersecting two different globals, or two different literals).

[tlively]

If I'm reading the code correctly, haveIntersection and isSubContents will happily report that Literal(1) < Global("g1", i32), and since intersect exits early when isSubContents is true, it has this behavior as well.

[kripken]

Ah, there might be bugs in haveIntersection and isSubContents then. But intersect() explicitly says it isn't ready for Globals yet,

https://github.com/WebAssembly/binaryen/blob/6e7a4a0507986d067550911a3bac59ce76f8453c/src/ir/possible-contents.cpp#L145-L147

edit: I guess it only checks that on other, so there might be a bug there too