Largely insufficient diagnostics with undefined decls in generics

[vadim-za]

Zig Version

0.13.0

Steps to Reproduce and Observed Output

Compile the following code. Use -freference-trace is you like - it doesn't seem to change the results.

const std = @import("std");

const ContainerA = struct {
    //const Item = struct {};
};

fn View(comptime Container: type) type {
    //if (comptime !@hasDecl(Container, "Item"))
    //    @compileError("Container must define Item");
    return struct {
        item: ?*Container.Item = null,
    };
}

fn ViewHost(comptime Container: type) type {
    return struct {
        view: View(Container) = .{},
    };
}

pub fn main() void {
    _ = ViewHost(ContainerA){};
}

The diagnostics says:

src\main.zig:11:26: error: struct 'main.ContainerA' has no member named 'Item'
        item: ?*Container.Item = null,
                ~~~~~~~~~^~~~~
src\main.zig:3:20: note: struct declared here
const ContainerA = struct {
                   ^~~~~~

It is absolutely unclear what caused ContainerA to be used as a parameter to View(). In a large or even medium codebase it might be impossible to find (unless the statement is literally View(ContainerA), which is not necessarily the case).

Now we can attempt to improve the diagnostics by uncommenting the two lines in the View function and compile again. The diagnostics now says

src\main.zig:9:9: error: Container must define Item
        @compileError("Container must define Item");
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
src\main.zig:17:19: note: called from here
        view: View(Container) = .{},
              ~~~~^~~~~~~~~~~

We are one step further, but still have no information on where ViewHost(ContainerA) was instantiated.

Expected Output

The error message's reference trace is expected to go all the way to the "root cause" position, which in both cases is the line in the main function.