[PATCH] D154007: Reland "Try to implement lambdas with inalloca parameters by forwarding without use of inallocas."

[Reid Kleckner via Phabricator via cfe-commits]

rnk added inline comments.


================
Comment at: clang/lib/CodeGen/CGClass.cpp:3095
+  StringRef CallOpName = CallOpFn->getName();
+  std::string ImplName = ("__impl" + CallOpName).str();
+
----------------
akhuang wrote:
> The old code tried to find the "<lambda_0>" part of the function name and replace the front with "?__impl@", so that the function name was consistent with the mangling scheme.  But apparently there are some cases where when the function name is too long, it uses a hash instead (?), so the attempt to find the <lambda_0> was failing. 
> 
> I couldn't find a good way to name this function through the actual mangling code -- is it fine to just add "__impl" to the front and have it not be mangled correctly? 
Let me suggest one more thing to try here to convince the mangler to do our work for us. You can create a CXXMethodDecl AST node here in CodeGen, and ask the mangler to mangle it for you. If you search CodeGen, you can find other examples of this, notably, for making implicit `this` variable declarations:
```
$ git grep Decl::Create'(' ../clang/lib/CodeGen/
../clang/lib/CodeGen/CGBuiltin.cpp:  Args.push_back(ImplicitParamDecl::Create(
../clang/lib/CodeGen/CGBuiltin.cpp:    Args.push_back(ImplicitParamDecl::Create(
../clang/lib/CodeGen/CGCXXABI.cpp:  auto *ThisDecl = ImplicitParamDecl::Create(
```

Call `CXXMethodDecl::Create`, and copy in all of the attribute from the call operator, but replace the DeclarationNameInfo with an IdentifierInfo of `__impl`.

If that's not working out, let's land this as is and follow up on fixing the mangling.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D154007/new/

https://reviews.llvm.org/D154007