in C double and _Float64 mist be destinct types (and float and _Float32)

[llvmbot]

Bugzilla Link	50766
Version	12.0
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@zygoloid

Extended Description

In C double and _Float64 are destinct types always.

In C float and _Float32 are destinct types always.

In _Float32 must NOT promote to double.

I have the definitive response from the sc22wg14 mailing list confirming
above statements.

Since the mailing list is not public I cannot sent the answer here.

x86-64 clang 12.0.0, icc 19.01 and tcc (unknown version) fail to compile
following program, but gcc 9.2.1 succeeds:

#include <stdio.h>
#include <math.h>

#define PGFN(name) _Generic((name),
float: printf("%s():" #name "=%f\n",func,name),
double: printf("%s():" #name "=%f\n",func,name),
default: printf("%s():" #name "=[unknown type]\n",func),
_Float32:printf("%s():" #name "=%f\n",func,((double)name)),
_Float64:printf("%s():" #name "=%f\n",func,name)
)

int main(int argc,char *argv[]){
float f=4.0;
double d=8.0;
_Float32 f32=32.0;
_Float64 f64=64.0;
(void)argc,(void)argv;
PGFN(f);
PGFN(d);
PGFN(f32);
PGFN(f64);
return 0;
}

[AaronBallman]

According to TS 18661-3 X.2.1p3: "Each interchange floating type is not compatible with any other type." so I can confirm that this is incorrect in Clang.

[AaronBallman]

Whoops, I confirmed too soon! Clang itself has no support for _Float32 or _Float64 yet (though we do have support for _Float16 from TS18661-3). These are coming from glibc: https://sourceware.org/git/?p=glibc.git;a=blob;f=bits/floatn-common.h;h=92982d64601bc562733ffd64f9e88854407c16c7;hb=refs/heads/master#l214 (Note: #define __HAVE_DISTINCT_FLOAT32 0 is also specified by glibc.)

So this is not actually a Clang bug, we don't claim to support TS 18661-3 (we don't define __STDC_IEC_60559_TYPES). You can verify this yourself by doing a -E compilation and noting where the typedef comes from. For me, that's:

# 214 "/usr/include/x86_64-linux-gnu/bits/floatn-common.h" 3 4
typedef float _Float32;
# 251 "/usr/include/x86_64-linux-gnu/bits/floatn-common.h" 3 4
typedef double _Float64;

Closing as invalid since this is a glibc behavior.