😃 (and other Unicode characters) in identifiers not allowed by g++

I am 😞 to find that I cannot use 😃 as a valid identifier with g++ 4.7, even with the -fextended-identifiers option enabled:

int main(int argc, const char* argv[])
{
  const char* 😃 = "I'm very happy";
  return 0;
}

main.cpp:3:3: error: stray ‘\360’ in program
main.cpp:3:3: error: stray ‘\237’ in program
main.cpp:3:3: error: stray ‘\230’ in program
main.cpp:3:3: error: stray ‘\203’ in program

After some googling, I discovered that UTF-8 characters are not yet supported in identifiers, but a universal-character-name should work. So I convert my source to:

int main(int argc, const char* argv[])
{
  const char* \U0001F603 = "I'm very happy";
  return 0;
}

main.cpp:3:15: error: universal character \U0001F603 is not valid in an identifier

So apparently 😃 isn't a valid identifier character. However, the standard specifically allows characters from the range 10000-1FFFD in Annex E.1 and doesn't disallow it as an initial character in E.2.

My next effort was to see if any other allowed Unicode characters worked - but none that I tried did. Not even the ever important PILE OF POO (💩) character.

So, for the sake of meaningful and descriptive variable names, what gives? Does -fextended-identifiers do as it advertises or not? Is it only supported in the very latest build? And what kind of support do other compilers have?

Solution 1:

As of 4.8, gcc does not support characters outside of the BMP used as identifiers. It seems to be an unnecessary restriction. Also, gcc only supports a very restricted set of character described in ucnid.tab, based on C99 and C++98 (it is not updated to C11 and C++11 yet, it seems).

As described in the manual, -fextended-identifiers is experimental, so it has a higher chance won't work as expected.

Edit:

GCC supported the C11 character set starting from 4.9.0 (svn r204886 to be precise). So OP's second piece of code using \U0001F603 does work. I still can't get the actual code using 😃 to work even with -finput-charset=UTF-8 with GCC 8.2 on https://gcc.godbolt.org though (You may want to follow this bug report, provided by @DanielWolf).

Meanwhile both pieces of code work on clang 3.3 without any options other than -std=c++11.

Solution 2:

This was a known bug in GCC 9 and before. This has been fixed in GCC 10.

The official changelog for GCC 10 contains this section:

Extended characters in identifiers may now be specified directly in the input encoding (UTF-8, by default), in addition to the UCN syntax (\uNNNN or \UNNNNNNNN) that is already supported:

static const int π = 3;
int get_naïve_pi() {
  return π;
}