Front-end support for header_unions #606
Conversation
| cstring fieldName = name + "." + f->name; | ||
| auto sl = create(f->type, fieldName); | ||
| if (globalValid != nullptr) | ||
| dynamic_cast<StructLocation*>(sl)->replaceField( |
There was a problem hiding this comment.
Is there a reason that dynamic_cast is used here instead of a to method?
There was a problem hiding this comment.
I would guess because the to methods currently all return const pointers. We should probably overload them to return non-const pointers when called on non-const objects.
There was a problem hiding this comment.
This is the single call of this method.
We have to rewrite this pass at some point: #151.
| // For header unions we will model all of the valid fields | ||
| // for all components as a single shared field. The | ||
| // reason is that updating one of may change all of the | ||
| // other ones. |
There was a problem hiding this comment.
Expanding this overview to what is implemented below would be helpful. (Could be done later.)
frontends/p4/def_use.h
Outdated
| || type-is<IR::Type_Error>() || type->is<IR::Type_Var>(), | ||
| { BUG_CHECK(type->is<IR::Type_Bits>() || type->is<IR::Type_Enum>() || | ||
| type->is<IR::Type_Boolean>() || type->is<IR::Type_Var>() || | ||
| type->is<IR::Type_Tuple>() || type-is<IR::Type_Error>(), |
There was a problem hiding this comment.
How is this not a syntax error (missing > in type-is<...)
There was a problem hiding this comment.
Isn't C++ beautiful? 'is' is a method on 'this', and this is just pointer arithmetic!
Sick, I will push a fix. Note that the bug has been there for a longer time.
And I was counting on the C++ type-checker...
I renamed the IR node to header_union, in case we ever get a real union. Most changes are due to this renaming.
Def-use analysis now handles unions; not many changes were required.
Also, changed the grammar to allow empty list expressions - good for initializing empty structs.