`local variable is never mutated` regression for field and array accesses

[squeek502]

Zig Version

0.16.0-dev.699+529aa9f27

Steps to Reproduce and Observed Behavior

I've run into this a few times now and thought I was going crazy. Locally, everything would be fine but CI would fail immediately with local variable is never mutated. One reproduction:

fn foo() void {
    var target_path_w: []const u8 = "foo";
    _ = target_path_w.len;
}

With zig1, it hits this error:

> zig1.exe ..\lib build-obj -ofmt=c -Mroot=test.zig
test.zig:2:9: error: local variable is never mutated
    var target_path_w: []const u8 = "foo";
        ^~~~~~~~~~~~~
test.zig:2:9: note: consider using 'const'

With a stage3 compiler, it does not:

> zig build-obj -ofmt=c -Mroot=test.zig

Note that it's not related to the function being unused or something like that. If you add a test block:

test "foo" {
    foo();
}

it passes:

> zig test test.zig
All 1 tests passed.

Somewhat surprisingly, zig1 seems to have the more correct behavior, as in each case I've hit, the error seems applicable.

Expected Behavior

zig1 and the stage3 compiler to agree on when things are never mutated.

[squeek502]

This is actually #25154 (review), so it regressed in #25154 and zig1.wasm hasn't been updated since.

If there was a follow-up issue made, then this should be closed as a duplicate of that. If not, then I guess this can serve as that follow-up issue.

cc @mlugg

[mlugg]

Ah yes, thank you @squeek502; indeed I forgot to make that issue, so this can serve as it.

[mlugg]

Migrated to https://codeberg.org/ziglang/zig/issues/31049