Name mangling for possible new Clang type about cxx-abi HOT 12 CLOSED

from cxx-abi.

The current vendor-extended type qualifier for address spaces makes the literal address space part of the <source-name>, e.g. U4AS16. I believe that predated proper support for parameterized qualifiers; if we were doing it now, it would probably be U2ASILi16EE, and that's probably the right pattern to use here.

from cxx-abi.

That is, this should just be "U2ASI" <expression> "E".

We should probably document the mangling of the address_space qualifier, since it doesn't follow the standard rule of using the qualifier's source spelling.

from cxx-abi.

If I am following this correctly then would the mangling function change to the following:

void CXXNameMangler::mangleType(const DependentExtAddressSpaceType *T) {
Out << "U2ASI";
mangleExpression(T->getAddrSpaceExpr());
Out << 'E';
}

from cxx-abi.

Yes, except you need to respect the correct ordering of qualifiers, and you may need to worry about qualifier nesting, i.e. whether you have Const(DependentExtAddressSpace(int)) vs. DependentExtAddressSpace(Const(int)). Please take further discussion of that back to the Clang development lists, however.

from cxx-abi.

@jicama How does GCC mangle address_space? I don't seem to have a copy of GCC for a port that supports a non-generic address space.

from cxx-abi.

from cxx-abi.

FWIW, that's https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69549.

from cxx-abi.

Okay. Do you have any objection to adopting our mangling if you ever decide to implement them? If not, I'll prepare a PR for that.

from cxx-abi.

from cxx-abi.

I've submitted a pull request (#29) documenting the address_space mangling.

from cxx-abi.

Merged that PR, which I believe finishes this issue.

from cxx-abi.