zig translate-c Type Propagation Error for Nullable Pointer ([*c]) Dereferencing and Address Calculation in Zig

[Yeaseen]

Zig Version

0.14.0-dev.2064+b5cafe223

Steps to Reproduce and Observed Behavior

Input C Code

typedef int (*myarr)[2];

void main() {
    // Define an array and set `a` to point to it
    int array[2] = {10, 20};
    int (*a)[2] = &array;  // `a` points to `array`

    // Set `b` to point to the same array, casted to `myarr`
    myarr b = (myarr)a;

    // Modify array elements through `b`
    (*b)[0] = 30;
}

Translated zig code

pub const myarr = [*c][2]c_int;
pub export fn main() void {
    var array: [2]c_int = [2]c_int{
        10,
        20,
    };
    _ = &array;
    var a: [*c][2]c_int = &array;
    _ = &a;
    var b: myarr = @as(myarr, @ptrCast(@alignCast(a)));
    _ = &b;
    b.*[@as(c_uint, @intCast(@as(c_int, 0)))] = 30;
}

zig build failure

runner.zig:68:49: error: expected type '[2]c_int', found 'comptime_int'
    b.*[@as(c_uint, @intCast(@as(c_int, 0)))] = 30;
                                                ^~

The generated zig code incorrectly uses a cast that results in a comptime_int type mismatch, making it impossible to modify the array elements as intended.

Expected Behavior

The zig translate-c tool should generate the correct zig code that allows accessing and modifying array elements through a pointer. For example, the output zig code should correctly handle the array pointer and allow the assignment.

[rohlem]

I don't see the issue in the generated code, rather something in the type check / access pattern of b.*[n] = x might be wrong in the compiler:

• @TypeOf(b.*[0]) == @TypeOf(b.*[1]) == c_int (imo correct)
• @TypeOf(&(b.*[1])) == *[2]c_int (imo wrong)
  It seems like taking the address of the double-dereference [*c]-Pointer access somehow moves us up two layers / to the root?
  If myarray is declared as *[2]c_int, the issue doesn't happen.
• @TypeOf(&(b.*[0])) == *allowzero [2]c_int, which I also can't explain - maybe we're incorrectly reinterpreting (uninitialized?) memory in some way?
  This also doesn't happen for myarray = *[2]c_int.

[Yeaseen]

Thank you for your detailed analysis @rohlem

It does seem zig translate-c is correctly mimicking C code's intent here. I updated the title. The issue indeed seems to stem from an inconsistency in how the Zig compiler handles [*c] pointers during dereferencing and address-of operations.

The issue with @TypeOf(&(b.*[0])) producing *allowzero [2]c_int instead of *c_int seems to require a deeper analysis of how the Zig compiler handles the following:

1. Dereferencing Nullable Pointers ([*c]):

   • Specifically, how the compiler propagates types during dereferencing and indexing for nullable pointers pointing to arrays.

2. Array Indexing and Address Calculation:

   • The unexpected type suggests the compiler might be treating the operation as taking the address of the entire array rather than a specific element. This could be related to how indexing (b.*[n]) interacts with nullable pointer semantics.

3. Role of allowzero:

   • The presence of allowzero in the result type might indicate additional complexity in how nullable pointers are being interpreted. This could involve an incorrect reinterpretation of memory during address-of operations.

Further investigation into these areas could help clarify why the compiler produces *allowzero [2]c_int and resolve the unexpected type propagation.

[flyfish30]

It is not a problem with array indexing and address calculation. The type of b.*[n] is c_int, which is correct, but the type of &b.*[n] is [2]c_int, which is incorrect. The result type of the address_of b.*[n] is wrong. In addition, the type of b.* is [2]c_int, so I think it is probably because the effects of operators & and [] cancel each other out, so the result of &b.*[n] is [2]c_int.

[Vexu]

This does look like a Sema bug:

pub export fn entry() void {
    var a: [*c][2]c_int = undefined;
    _ = &a;
    // a.*[0] = 30; // incorrect type
    a[0][0] = 30;

    var b: [*][2]c_int = undefined;
    _ = &b;
    // b.*[0] = 30; // dereference not allowed
    b[0][0] = 30;
}

[flyfish30]

I have fixed this issue on my zig 0.14 source code, and can now successfully compile zig code that translated from the C source code in this issue, and run the built executable successfully.