Proposal: throw away diag.rs module and use proc-macro-error crate about auto_impl HOT 3 CLOSED

Hi there!

Thanks for reaching out to us. While writing diag.rs I certainly wished that there was a nice proc macro error reporting library. So I'm very tempted to replace it with a library.

However, I'm not quite sure about proc-macro-error. Reading over the documentation, it seems a bit too much magic to me. Especially the #[proc_macro_error] attribute and what exactly it does to the main function is not clear to me. But maybe it's indeed a good way to do error reporting in proc macros and I just don't understand yet. I would have to try it or see it in action to really be able to judge.

If you want, you can prepare a PR under the condition that I might still say "no" to it. Of course I'm not the only one deciding things for auto-impl. And we should probably wait for your new release so that auto-impl does not loose functionality (notes). But I can also totally understand if you don't want to create a PR that might get rejected.

from auto_impl.

There's no magic involved, actually. #[proc_macro_error] just helps you to build up right entry_point invocation while keeping indentation level flat. What exactly it does:

• It makes a closure of your whole top-level function, wrapping it with AssertUnwindSafe if needed or requested (for example, prior to Rust 1.37 proc_macro::TokenStream didn't implement UnwindSafe so this wrapper is necessary). Then it just gives this closure over to entry_point, the real work gets done there.
• First, entry_point executes your code, catching all the panics via catch_unwind. Then it clears the thread_local storage, resumes panic if it's not especial AbortNow. Then, if storage is empty - it just returns your closure's result, if not - makes a sequence of compile_error! invocations and returns it instead. Basically, that's all.

As you see, there's no magic in it. Moreover, if you decide to implement such global storage at yourself you'll simply end up reimplementing 80% of my crate and there's no guarantee you won't miss something. For example, global storage MUST be cleared (you already know what might happen if you don't do that 😄), even in case of panic, so you would have to use catch_unwind too, which brings all the headache of AssertUnwingSafe.

If you want, you can prepare a PR under the condition that I might still say "no" to it.

This is just sensible, I would be very surprised otherwise. In fact, I was just checking if it's still maintained :)

from auto_impl.

Closed in #60

from auto_impl.