checker: allow int i32 to be used interchangeably

[kbkpbot]

This is 3rd PR to support 64bit/32bit int as PR #25236 is the first one and #25298 is the second one.

Before apply v fmt -new_int -w . to the whole v codebase, it is need to ease some type check first.
This PR allow i32 <=> bool and also int <=> i32 type check.

[spytheman]

Why is i32 <=> bool needed?

[JalonSolov]

In a lot of C compilers bool is typedef int bool. On a 32-bit system, an int is i32, so bool will also be i32, and I've seen code where it was simply treated as an int rather than a separate type.

It used to be quite common to see

typedef int bool;
#define true 1
#define false 0

Then your code do

if (mybool == true) // or more often, `if (mybool)`
{
 ...
}

or whatever, which is all i32s.

[kbkpbot]

Alternatively, we can mandate the addition of explicit type casting where necessary in the code, such as int(bool_value) , i32(bool_value) or bool(int_value), to ensure type matching. Such instances are not too many in the code and can be manually modified.

[JalonSolov]

That would be the better thing all around, really, as you cannot guarantee that int == i32 in all C compilers. The only guarantee on sizes is the sizeof(char) == 1. Everything else is "implementation defined".

[kbkpbot]

can we remove these from the checker?

		// allow bool & int to be used interchangeably for C functions
		if (got.idx() == ast.bool_type_idx
			&& expected.idx() in [ast.int_type_idx, ast.int_literal_type_idx, ast.i32_type_idx])
			|| (expected.idx() == ast.bool_type_idx
			&& got.idx() in [ast.int_type_idx, ast.int_literal_type_idx, ast.i32_type_idx]) {
			return
		}
		exp_sym := c.table.sym(expected)
		// unknown C types are set to int, allow int to be used for types like `&C.FILE`
		// eg. `C.fflush(C.stderr)` - error: cannot use `int` as `&C.FILE` in argument 1 to `C.fflush`
		if exp_is_ptr && exp_sym.language == .c && exp_sym.kind in [.placeholder, .struct]
			&& got == ast.int_type_idx {
			return
		}

I suggest remove these checks, and make explicit type casting in V source code.
C.fflush(C.stdout) => C.fflush(&C.FILE(C.stdout))

As int is not a pointer(&C.FILE) at all.

BTW:

pub enum FileBufferMode {
        fully_buffered = C._IOFBF
        line_buffered  = C._IOLBF
        not_buffered   = C._IONBF
}

need change to

pub enum FileBufferMode {
        fully_buffered = i32(C._IOFBF)
        line_buffered  = i32(C._IOLBF)
        not_buffered   = i32(C._IONBF)
}

[kbkpbot]

Or should we introduce something redecalre the C.xx's type in V? This will help integrate C code.

For example:

(C.stdio, C.stderr) = &C.FILE

[spytheman]

Please do focus on the current PR, and avoid discussing things unrelated to it here.

There is https://github.com/vlang/v/discussions and Discord for chatting/discussing other topics.

[spytheman]

  // unknown C types are set to int, allow int to be used for types like `&C.FILE`
  // eg. `C.fflush(C.stderr)` - error: cannot use `int` as `&C.FILE` in argument 1 to `C.fflush`

This should now just apply to i32 .
The C side is not going to change, just because V decided to be more like Go on a whim.

[spytheman]

// allow bool & int to be used interchangeably for C functions

Similarly, this should apply to i32 now, and only for C functions, to allow for the same easy interoperability as before.

On the V side, now bool != int, and in the future it will be still != int and != i32.

[spytheman]

Or should we introduce something redecalre the C.xx's type in V?
This will help integrate C code.

That can be a separate proposal/PR, for C globals/consts .

[spytheman]

Excellent work.
Thank you very much @kbkpbot 🙇🏻.