[all-commits] [llvm/llvm-project] 815446: [clang][NFC] Standard substitution checking cleanup

Nathan Sidwell via All-commits all-commits at lists.llvm.org
Thu Feb 10 04:45:21 PST 2022


  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 815446cd3e164b8be010864648d4104ad178b129
      https://github.com/llvm/llvm-project/commit/815446cd3e164b8be010864648d4104ad178b129
  Author: Nathan Sidwell <nathan at acm.org>
  Date:   2022-02-10 (Thu, 10 Feb 2022)

  Changed paths:
    M clang/lib/AST/ItaniumMangle.cpp

  Log Message:
  -----------
  [clang][NFC] Standard substitution checking cleanup

In preparing for module mangling changes I noticed some issues with
the way we check for std::basic_string instantiations and friends.

*) there's a single routine for std::basic_{i,o,io}stream but it is
 templatized on the length of the name.  Really?  just use a
 StringRef, rather than clone the entire routine just for
 'basic_iostream'.

*) We have a helper routine to check for char type, and call it from
 several places.  But given all the instantiations are of the form
 TPL<char, Other<char> ...> we could just check the first arg is char
 and the later templated args are instantiating that same type.  A
 simpler type comparison.

*) Because basic_string has a third allocator parameter, it is open
 coded, which I found a little confusing.  But otherwise it's exactly
 the same pattern as the iostream ones.  Just tell that checker about
 whether there's an expected allocator argument.[*]

*) We may as well return in each block of mangleStandardSubstitution
 once we determine it is not one of the entities of interest -- it
 certainly cannot be one of the other kinds of entities.

FWIW this shaves about 500 bytes off the executable.

[*] I suppose we could also have this routine a tri-value, with one to
indicat 'it is this name, but it's not the one you're looking for', to
avoid later calls trying different names?

Reviewd By: ChuanqiXu

Differential Revision: https://reviews.llvm.org/D119333




More information about the All-commits mailing list