Pointer arithmetic affects the alignment factor

[LemonBoy]

Closes #1528

[andrewrk]

This case regressed:

test "hi" {
    var ptr: [*]align(8) u32 = undefined;

    const ptr2 = ptr + 1;
    @compileLog(@TypeOf(ptr2));
}

expected: [*]u32
found: [*]align(1) u32

[LemonBoy]

This case regressed:

Hm? You're increasing ptr address by 4 (since it's an array of u32) and that makes it not aligned to 8 anymore

[andrewrk]

Yes but it should still be aligned to 4, e.g. it should not be underaligned. Maybe this is a better example:

test "extra aligned ptr" {
    var x: [2]u32 align(8) = [_]u32{1, 2};

    const result = addTwoNumbers(&x);
    @import("std").testing.expect(result == 3);
}

fn addTwoNumbers(ptr: [*]align(8) u32) u32 {
    return fetch(ptr + 0) + fetch(ptr + 1);
}

fn fetch(ptr: [*]u32) u32 {
    return ptr[0];
}

This correctly passes in master branch.

[LemonBoy]

Yes but it should still be aligned to 4, e.g. it should not be underaligned.

Gotcha, I've added some code to do so ;)

[andrewrk]

👏 👏 👏

much sophisticated, very wow 🐶

[andrewrk]

Wait, this should still be aligned to u32. x is runtime known but the number of bytes the pointer marches forward is @sizeOf(u32) * x which is guaranteed to be divisible by @sizeOf(u32).

[LemonBoy]

Yeah, I focused too much on the comptime-known branch and forgot the other one, now it should be fine!