Re: Really obscure C++ mangling error with -fnew-abi

[Alex Samuel <samuel at codesourcery dot com>]

"Martin v. Loewis" <martin@loewis.home.cs.tu-berlin.de> writes:

  Martin> I was confused by the statement

  spec> There are two exceptions that appear to be substitution
  spec> candidates from the grammar, but are explicitly excluded:
  spec> ... function and operator names other than extern "C"
  spec> functions.

I think that comment is simply out-of-date.  It should be removed.

  Martin> The rationale for this exception was that a function name
  Martin> cannot appear later in the mangling, and that it may cause
  Martin> confusion with types that have the same qualifier. So I
  Martin> assumed the same rationale holds for names of template
  Martin> functions.

The function name can appear later in the mangling, for instance in a
template argument, as you point out.  But the name of the function is
not enough to meet the `entities that would appear in a symbol table'
test.  I think that the name of a template is (since you can't
overload based on template arguments... phew!).

  Martin> However, the exception appears to apply to the 
  Martin> <function name> in an <encoding> only.

The grammar in the spec as currently posted is out-of-date too.
Changes were accepted at last week's meeting that haven't made their
way into the document yet.  These changes reflect a different grammar,
as it is implemented in g++.  If you want to see them right away, the
grammar is documented in comments in cp-demangle.c.

Regards
Alex Samuel
CodeSourcery LLC