Preprocessing numbers with dollar signs are rejected when they are accepted in identifiers

[Halalaluyafail3]

The following program demonstrates the issue:

#define CAT(X,Y,Z)CAT_(X,Y##Z)
#define CAT_(X,Y)X##Y
int main(){
    int CAT(X,0,$);
}

Clang rejects this because it believes 0$ is not a valid preprocessing token. GCC and MSVC accept this code. Considering that the definition of pp-number references identifier-continue, I don't see any reason that this code should be rejected.

This also applies after a digit separator:

#define STR(X)#X
int main(){
    STR(0'$);
}

C23 defines the optional support for dollar signs in identifiers by allowing nondigits to include dollar signs. So the reference to nondigit in pp-number should also allow a dollar sign after a digit separator. MSVC accepts this, GCC doesn't. See also: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123057

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (Halalaluyafail3)

The following program demonstrates the issue: ```c #define CAT(X,Y,Z)CAT_(X,Y##Z) #define CAT_(X,Y)X##Y int main(){ int CAT(X,0,$); } ``` Clang rejects this because it believes `0$` is not a valid preprocessing token. GCC and MSVC accept this code. Considering that the definition of pp-number references identifier-continue, I don't see any reason that this code should be rejected.

This also applies after a digit separator:

#define STR(X)#X
int main(){
    STR(0'$);
}

C23 defines the optional support for dollar signs in identifiers by allowing nondigits to include dollar signs. So the reference to nondigit in pp-number should also allow a dollar sign after a digit separator. MSVC accepts this, GCC doesn't. See also: <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123057>

[shafik]

CC @AaronBallman

[AaronBallman]

The flag is named -fdollars-in-identifiers. A preprocessing number is not an identifier even if it's using identifier-continue in the grammar. What's the use case here? (Nothing in the original discussions in WG14 mention use of this for pp-number, everything is about existing compiler extensions around identifiers. So it's not clear this isn't a bug in the standard as opposed to a bug in Clang.)

[Halalaluyafail3]

What's the use case here? (Nothing in the original discussions in WG14 mention use of this for pp-number, everything is about existing compiler extensions around identifiers. So it's not clear this isn't a bug in the standard as opposed to a bug in Clang.)

Consider the following from the C99 Rationale (Section 6.4.8):

A new feature of C99: C99 replaces nondigit with identifier-nondigit in the grammar to allow
the token pasting operator, ##, to work as expected. Given the code

#define mkident(s) s ## 1m
/* ... */
int mkident(int) = 0;

if an identifier is passed to the mkident macro, then 1m is parsed as a single pp-number, a valid
single identifier is produced by the ## operator, and nothing harmful happens. But consider a
similar construction that might appear using Greek script:

#define µk(p) p ## 1µ
/* ... */
int µk(int) = 0;

For this code to work, 1µ must be parsed as only one pp-token. Restricting pp-numbers to only
the basic letters would break this.

This also applies to dollar signs:

#define $k(p)p##1$
int $k(int)=0;

[AaronBallman]

Thanks, that makes sense to me!