"Deducing this" mangling

[royjacobson]

Hi,

Has there been a discussion about the mangling scheme for "deducing this" methods from http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p0847r7.html? At least MSVC seem to have different mangling scheme for those new functions, but I couldn't find anything related to Itanium.

[cor3ntin]

The reason we do need a different mangling is for the following case

struct S {
  static void f(S);
  void f(this S);
};```

[jicama]

I'd be inclined to mangle the presence of an explicit object parameter in <nested-name>, like we currently do function cv-quals and ref-quals for implicit object member functions.

[jicama]

Maybe 'B', since 'O' is already a ref-qualifier, and 'E' might create confusion with the end marker. Or perhaps 'X'.

[jicama]

Or we could make the above case an ill-formed overload set and not need to mangle it at all.

[cor3ntin]

Or we could make the above case an ill-formed overload set and not need to mangle it at all.

That would be my preferred solution I think. But we'd need an other round trip through wg21, and i think they might have already discussed that point (not 100% sure)

[jicama]

Or we could make the above case an ill-formed overload set and not need to mangle it at all.

That would be my preferred solution I think. But we'd need an other round trip through wg21, and i think they might have already discussed that point (not 100% sure)

https://cplusplus.github.io/CWG/issues/2687.html

[cor3ntin]

https://isocpp.org/files/papers/P2797R0.html Alas! @jicama I'm not sure how ABI discussions usually go. I'm happy to use either B or O or whatever other scheme :)

[royjacobson]

Doesn't matter too much, but I like 'x' for the eXplicit object parameter

[zygoloid]

Is there any world in which we want an ABI tag marker to appear as part of the cv-qualifier sequence on a function? If not, B seems fine to me. (Otherwise, how about H (tHis)? That seems entirely unused.)

Presumably this marker appears instead of the [\] [\] portion (that is, we never mangle a \ or \ for a function with an explicit object parameter, regardless of the declared type of the explicit object parameter).

[cor3ntin]

@zygoloid I'm happy with either option but we should pick one (Otherwise i might use a vendor extension in clang until this is resolved). Maybe H is the option with the least question so we can go with that

[zygoloid]

I think having NB3foo as an explicit object parameter marker but N1XB3foo as an abi tag would be confusing for human readers of manglings, even if it's unambiguous, so I think we should use H.

[cor3ntin]

@zygoloid thanks. I went with H in clang.

[ldionne]

Should we standardize using H then? We also need to implement this change in the demangler and we're wondering if we want to go with H now or if we need to wait for this to be settled to something else in the official ABI?

[jakubjelinek]

GCC uses H since https://gcc.gnu.org/r14-7222 Would be nice if this was actually turned into the mangling ABI patch and pushed.