the language should honor alignment decaying in type coercion

[andrewrk]

export fn entry() void {
    var buf: [10]u8 align(4096) = undefined;
    const x: []u8 = &buf;
    @compileLog(@TypeOf(x));
}

This outputs:

[nix-shell:~/Downloads/zig/build]$ ./zig build-obj foo.zig 
| []align(4096) u8
./foo.zig:4:5: error: found compile log statement
    @compileLog(@TypeOf(x));
    ^

I believe this strange behavior is the cause of the bug that is surfacing in #7482 (comment)

The language should not ignore the programmer's type that they specify.

Related: #1106

Labeling as a bug because if this proposal is accepted I believe it will fix a bug, if this proposal is rejected then this issue will be the bug report for that behavior. The bug is that generic functions which accept slices/pointers get incorrectly generated with respect to alignment. See that linked issue comment above for more details.

[LemonBoy]

The same thing happens with @ptrCast, ir_analyze_ptr_cast has a keep_bigger_alignment parameter that may hint this is a wanted behaviour?

[andrewrk]

I did it intentionally but I think that may have been a bad idea. I'm pretty sure we should change the language spec to accept this proposal but I just want to put a bit of thought into @ptrCast because it would be nice to keep extra alignment when doing pointer casting.