bulbazord added a comment.

In D128366#3603943 <https://reviews.llvm.org/D128366#3603943>, @labath wrote:

> How hot is this code? Parsing a demangled name isn't entirely cheap, and the 
> fact that the objc++ class calls back into the c++ version means that the c++ 
> name will be parsed twice.

Not sure, I'll measure this. Also, at least on my machine, the list of language 
plugins includes at least 3 C++ language plugins, so it happens *at least* 
twice...

> Could we move this pruning elsewhere? These values come from the symbol file 
> plugins anyway, and they can do a better job at determining which language 
> does a particular name belong to.
> (OK, they can an also come from the symtab, but there I guess we could infer 
> something from the mangling scheme).

We're specifically pruning the results from a name lookup, so I'm not sure 
where would be a better place to move it. That being said, based on your 
observation, we could try to do a better job of inferring the language a name 
belongs to by looking at the mangling scheme before we look at every single 
language plugin.



================
Comment at: lldb/include/lldb/Target/Language.h:320
+  virtual bool NamesAreEquivalentWithContext(
+      const ConstString &user_provided_name,
+      const ConstString &full_name_with_context) const {
----------------
labath wrote:
> I think we're passing ConstStrings by value these days...
I see. Is there a reason we do that instead of passing by reference?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D128366/new/

https://reviews.llvm.org/D128366

_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits

Reply via email to