translate-c: `char` always converted to `u8`, even on platforms where `char` is signed

[ehaas]

The following C code on x86_64-macos-gnu does not call abort; but when run via translate-c it does call abort:

#include <stdlib.h>
int main(void) {
    char c = -1;
    if (c != -1) abort();
}

In transType, .Char_S is translated as u8, but it means 'char' for targets where it's signed

Treating .Char_S as signed will break string literal -> char * implicit casts on those platforms; perhaps the proper thing to do is to special-case char * in transType ?

[xackus]

#875

[nektro]

https://www.cs.swarthmore.edu/~newhall/unixhelp/C_chars.html

In C, char values are stored in 1 byte, and are encoded as numbers using the ASCII encoding.

This is perfectly compatible with the u8 number type.

[nektro]

but when run via translate-c it does call abort:

what does translate-c generate for you?

[ehaas]

pub export fn main() c_int {
    var c: u8 = @bitCast(u8, @truncate(i8, -@as(c_int, 1)));
    if (@bitCast(c_int, @as(c_uint, c)) != -@as(c_int, 1)) {
        abort();
    }
    return 0;
}

c is signed in the original C code on my platform but translate-c converts it to unsigned. In the C if statement if (c != -1), c is promoted from a signed char to a signed int; in the Zig code 255 is promoted to an unsigned int and then bitcasted to a signed int. So it compares the c_int values 255 and -1 and ultimately calls abort, which is incorrect behavior.

[Mouvedia]

c_char type is currently always signed.

Concerning signedness detection see the discussion starting with this comment.

[Vexu]

Here's the function Aro uses to determine char signedness:
https://github.com/Vexu/arocc/blob/c1955a474270562a4640217e62c888b52c79d39e/src/target.zig#L7