Incorrect spacing in macro stringizing

[kyuupichan]

#define str(x) #x
#define xstr(x) str(x)
#define f(x) x :
xstr(+f())

With the empty argument to f, the space before the colon somehow gets eaten.
I believe this should expand to "+ :" with a space before the colon which is what GCC does.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Neil Booth (kyuupichan)

#define str(x) #x #define xstr(x) str(x) #define f(x) x : xstr(+f())

With the empty argument to f, the space before the colon somehow gets eaten.
I believe this should expand to "+ :" with a space before the colon which is what GCC does.

[shafik]

gcc/MSVC agree and clang/EDG agree: https://godbolt.org/z/jn7Pse8f7

[shafik]

Also see: https://godbolt.org/z/8Wa9v1bWc

#define str(x) #x
#define xstr(x) str(x)
#define f(x) x :
xstr(f()) // Only MSVC differs
xstr(+ :) // All agree

[shafik]

Also see: https://godbolt.org/z/rcEsEbaea

#define str(x) #x
#define xstr(x) str(x)
#define f(x) x :
+f() // Only clang differs
f() // Only clang differs
+ : // All agree

clang output:

+:
:
+ :

Everyone else:

+ :
 :
+ :

So this looks like the difference is in the behavior after macro expansion. Clang seems the odd one out. (I will wait for more feedback before changing the title)

I feel maybe both behavior are acceptable since whitespace characters separating tokens are no longer significant, I think at this point.

CC @cor3ntin @tahonermann (feel free to update title if you agree w/ me)

[kyuupichan]

Of course, the standard places no requirements on preprocessed output.
However, it does for stringization which is why whitespace tracking matters in a conformance sense.

[zygoloid]

Also see: https://godbolt.org/z/8Wa9v1bWc

#define str(x) #x
#define xstr(x) str(x)
#define f(x) x :
xstr(f()) // Only MSVC differs
xstr(+ :) // All agree

MSVC is wrong here. See [cpp.stringize]/2: "Whitespace before the first preprocessing token and after the last preprocessing token comprising the stringizing argument is deleted." MSVC's new preprocessor (/Zc:preprocessor) gets this right.