Struct initialization and reassignment to data dependent location causes possibly unexpected behavior

[aidanbabo]

Zig Version

0.13.0

Steps to Reproduce and Observed Behavior

Sorry for the lengthy title. Given a snippet of Zig code like below:

const Point = extern struct {
    r: i32,
    c: i32,
};

export fn foo() Point {
    var p: Point = .{ .r = -1, .c = 0 };
    p = .{ .r = p.c, .c = -p.r };
    return p;
}

pub fn main() void {
    const p: Point = foo();
    @import("std").debug.print("({d}, {d})\n", .{ p.r, p.c });
}

the program prints (0, 0) when run.

The struct and function are marked as export because I was working with them in Godbolt, but the problem manifest with Zig's layout/cc as well.

Expected Behavior

Given a similar snipped of C code:

struct point {
    int r;
    int c;
};

struct point foo(void) {
    struct point p = { .r = -1, .c = 0 };
    p = (struct point){ .r = p.c, .c = -p.r };
    return p;
}

#include <stdio.h>

int main(void) {
    struct point p = foo();
    printf("(%d, %d)\n", p.r, p.c);
}

the program prints (0, 1), which was my intention when I wrote the above Zig code.

I'm fairly new to Zig, but the behavior in C feels much more intuitive. I would expect the struct initializer to be fully evaluated before any data is written into p.

[rohlem]

Duplicate of / related to #3696 (or more particularly #12064) and the dozen other issues linked there.

At some point the current behavior was additionally documented in the language reference (langref):

In essence, Zig desugars the assignment foo = .{ .a = x, .b = y } to the two statements foo.a = x; foo.b = y; .

In your example code, p = .{ .r = p.c, .c = -p.r }; becomes p.r = p.c; p.c = -p.r;.
The workaround for now is to specify the type (p = Point{ .r = p.c, .c = -p.r };), or use a temporary const new_p = ...; p = new_p;.