[PATCH] D149677: [clang][TypePrinter] Add option to skip over elaborated types

[Eric Li via Phabricator via cfe-commits]

li.zhe.hua added inline comments.


================
Comment at: clang/lib/AST/TypePrinter.cpp:1570
 
+  if (Policy.IgnoreElaboratedQualifiers) {
+    printBefore(T->getNamedType(), OS);
----------------
aaron.ballman wrote:
> So, effectively, the idea here is: you want the ability to skip printing the elaborated keyword and nested name specifier, and you additionally want to skip use of `ElaboratedTypePolicyRAII` when calling `printBefore()`/`printAfter()` because that forces suppression of tags and scopes?
> 
> I think what I'm struggling with a bit is that this is doing three things at once; one is the elaboration keyword (so we no longer print `typename` or `struct` when `IncludeTagDefinition` is false), another is the nested name specifier (we no longer print the leading foo::bar), and the third is that we no longer suppress tags and scopes when printing the underlying type. That makes it tricky to figure out how to name this thing, but the best I could come up with is: `SuppressElaboration` which isn't really different from `IgnoreElaboration` at all. So I think either of those names is "fine", but I'm still a bit uncomfortable about how complex the interactions are becoming.
> another is the nested name specifier (we no longer print the leading foo::bar), and the third is that we no longer suppress tags and scopes when printing the underlying type

I would say that this is the primary desired behavior of the option. More specifically, it's //not// that we don't print the nested name specifier and no longer suppress those options, but rather, we print the underlying type with the options as specified by the policy. `SuppressTagKeyword`, `SuppressScope`, and `FullyQualifiedName` work as expected (from my perspective). The interaction with `IncludeTagDefinition` dropping the keyword wasn't explicitly intended, but it is still consistent with printing the canonical type (which would also drop the keyword).

One last name idea to throw out there, in case it seems worth discussing: `DesugarElaboratedTypes`. In that case, I'd probably also move the check up above L1559 and we treat the entire `ElaboratedType` as if it did not exist.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149677/new/

https://reviews.llvm.org/D149677