[P4Testgen] Add a compiler pass to resolve Type_Name in StructExpressions.

[fruffy]

In some cases, one may want to create a struct expression, which contains the relevant type directly, without having to use a Type_Name as indirection. This PR relaxes the constraints around the creation of struct expressions to make this possible. We simply add a new parameter that allows us to toggle whether or not to accept IR::Type_StructLike types.

When resolving struct expressions, simply preserve Type_Name for structType but assign the resolved type to type. This works for all dependent PRs and preserves functionality of passes such as toP4 etc.

[fruffy]

@mihaibudiu I updated the constructor to include an optional structOkay boolean.

[mihaibudiu]

this can't be the whole PR, there is no use of this flag.

[fruffy]

The flag is used in #4231 and #4255. And an extension which is not part of the main repository here: https://github.com/fruffy/flay

I factored out the change to make the PR more readable.

I could add a compiler pass which resolves Type_Name for StructExpressions. It would make use of this flag. It might be useful for this PR, too: #4225 (comment)

[mihaibudiu]

The pass which resolves the name is called ResolveReferences, it's perhaps the oldest pass in the compiler.
If a struct has no name there is nothing to resolve.

[mihaibudiu]

why do you want this?
There is a difference between a struct declaration (Type_Struct) and a reference to a type by name (Type_Name).
It is more annoying to have to resolve the type everytime, but it is the right distinction.
Without this many passes would break.
Perhaps the type field holds the information you want.

[fruffy]

The pass which resolves the name is called ResolveReferences, it's perhaps the oldest pass in the compiler.
If a struct has no name there is nothing to resolve.

Thanks!

The idea is to resolve all Type_Name past the mid end since they serve no purpose after the passes have completed. This way we can access any type of any expression directly without requiring a map access. Any IR::Expression will have a fully usable type.

[mihaibudiu]

Various backends execute additional transformation passes after the mid-end.
Moreover, this may break to toP4 pass.

[fruffy]

@mihaibudiu I believe I found a more palatable solution which does not affect the ir or the rest of the compiler. When resolving struct expressions, I simply preserve Type_Name for structType but assign the resolved type to type. This works for all dependent PRs and preserves functionality of passes such as toP4 etc.

It essentially exploits the inconsistency of type and structType in StructExpressions.