Invalid attribute accepted dealing with # and ## operators

[pinskia]

Forwarded from https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120978 because clang is the only major compiler which accepts this (EDG, GCC and MSVC all reject this for having # and ## operators outside of macros).

[[ignored::ignored(##)]]int main(){}

The way I read the standard is during phase 4, this should be diagnostic as invalid. Which is exactly how GCC, EDG and MSVC all do as ## operator is invalid outside of the function like macros.

But I could be reading the spec wrong but with clang being the odd one out though; it looks more likely a clang issue.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Andrew Pinski (pinskia)

Forwarded from https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120978 because clang is the only major compiler which accepts this (EDG, GCC and MSVC all reject this for having # and ## operators outside of macros).

[[ignored::ignored(##)]]int main(){}

The way I read the standard is during phase 4, this should be diagnostic as invalid. Which is exactly how GCC, EDG and MSVC all do as ## operator is invalid outside of the function like macros.

But I could be reading the spec wrong but with clang being the odd one out though; it looks more likely a clang issue.

[cor3ntin]

I think clang is wrong, the grammar of token does not include #

https://eel.is/c++draft/lex.token#nt:token
https://eel.is/c++draft/dcl.attr#nt:balanced-token

[erichkeane]

C considers # to be a token, as well as ##. Taken from N3088:

6.4.6 Punctuators

punctuator: one of
[ ] ( ) { } . ->
++ -- & * + - ~ !
/ % << >> < > <= >= == != ^ | && ||
? : :: ; ...
= *= /= %= += -= <<= >>= &= ^= |=
, # ##
<: :> <% %> %: %:%:

6.4 Lexical elements:

token:
keyword
identifier
constant
string-literal
punctuator

preprocessing-token:
header-name
identifier
pp-number
character-constant
string-literal
punctuator
each universal-character-name that cannot be one of the above
each non-white-space character that cannot be one of the above

So it seems to me that Clang is 'right', at least as far as C is concerned?

[zygoloid]

The behavior in C++ doesn't seem especially clear to me. We seem to reach translation phase 7's "Each preprocessing token is converted into a token", which doesn't specify what mapping to perform, but also has no allowance for this mapping to fail.

See CWG1901; the drafting I suggested there would make this ill-formed, but I think the status quo is that it's undefined behavior.