inferred error sets have strippable identity (missing documentation?)

[rohlem]

Zig Version

0.14.0-dev.617+208baa37c

Steps to Reproduce and Observed Behavior

Repro:

fn foo() !void {}
comptime {
    const A = @typeInfo(@TypeOf(foo())).ErrorUnion.error_set;
    const B = error{};
    @compileLog(A == B); //false
    @compileLog((A || error{}) == B); //true
    @compileLog(@Type(@typeInfo(A)) == B); //true
}

Expected Behavior

I didn't know that inferred error sets have identity that makes them compare unequal to all other error sets - it's not currently mentioned in their langref section.

When looking for previous issues on this I found that all error sets used to have identity, which changed with implementing #11022 .
A comment there documents that "we're experimenting" with inferred error sets having identity "to ease the implementation of the language specification. It's problematic to make type equality force error set resolution."
If that's still the current stance on the issue, I can accept that it just needs to be documented.

[mlugg]

Yep, status quo is intended, and this is just missing documentation.