Should miette be eyreish, or should it be thiserrorish

[CAD97]

My take: miette is doing a bit too much by being an eyreish and providing a Report type. It should focus on just providing the Error type (derive).

For use in libraries, you already recommend just using #[derive(Error, Diagnostic)], and not using Report.

For use in applications, you provide Report and IntoDiagnostic. IntoDiagnostic is effectively lossy: it wraps an arbitrary impl Error into effectively a "null object" impl Diagnostic which provides None for all of its diagnostic info. The underlying error is seen through Error::source.

GraphicalReportHandler does render the source chain, at least, and wrap_err maintains the diagnostic information.

But I'm not certain exactly what benefit miette::Report provides over using eyre::Report. It feels to me that what miette wants is an eyre Section for the diagnostic information.

What I'm most concerned about, I guess, is that I make a nice miette diagnostic for my library's error, and then none of it is seen unless the application sticks that error into a miette::Report. If the error is wrapped by another library, the information is lost unless it passes through the diagnostic info.

I honestly don't know the "correct" or "best" solution here. But at least to me it feels like miette is trying to be two things -- a general purpose error reporting library, and a specialized parsing error library -- but not really excelling at either. (Specifically, the requirement of 'static gets in the way of use for parsing libraries which parse &str, since it requires a source clone.)

What I'm doing for the right now in my library is just providing a non-'static Diagnostic and pointing at how to use a ReportHandler to render it, providing a static_diagnostic() -> impl Copy + Diagnostic which strips the source information (since it's a parsing library, the caller has the source and can re-inject it), and a to_owned() -> miette::Report which uses with_source_code on miette!(self.static_diagnostic()) to attach the source info in an owned manner.

At the very least, though, even if Report is kept, a decent amount of eyre's complexity could be trimmed off; the fact that the reporting hook creates a separate report handler for every miette::Report seems excessive; I don't really see a use case for downcasting and providing different report handlers for different concrete diagnostic types, and just registering a global &'static ReportHandler can reduce complexity there.

I'll be completely honest and say that as part of writing this down I'm less certain in my position now, but it still feels to me that miette can and should use eyre rather than being an eyreish. (The point of eyre is to be pluggable, after all, especially color-eyre.)

[CAD97]

To be completely transparent:

The reason this has come up for me is I'm using miette to report parsing errors and building a serde deserializer on top of that. The chance of someone using miette to Report errors from a serde deserialization is approximately zero, but I would still like to provide nice errors. I'm still figuring out how best to do so. Having a get_hook() that can handle non-'static reports would allow me to stick the formatting into Debug and/or Display.

[zkat]

eyreish has always been a temporary measure, until provide-any landed, so you could have Diagnostics without being lossy. I guess I don't actually understand how a Section could be used and still have access to Diagnostic fields?

[CAD97]

I'll be transparent: I thought that (color-)eyre provided a pluggable solution for provide-any, and that's how spantraces are attached. But instead, it's just all added to color-eyre's Handler. (The advantage would be getting #14 implicitly, of course.)

I guess what I think I want is for MyDiagnostic as Display to display the global-hook-ily formatted diagnostic. Then miette diagnostics would be trivially pluggable into an anyhow/eyre error stack.

(And for color-eyre, what I think would be ideal, even if not realistic, would be for the current Display message to be provided in the error chain, and the rest of the diagnostic attatched in a ━━━━━━━━ DIAGNOSTIC ━━━━━━━━ section. IIUC after further study of eyre/color-eyre, though, this would require another layer of hooking to allow miette to attach a .with_section when converting MyDiagnostic into eyre::Report (with color_eyre::Handler), as well as the section callback having some way to access MyDiagnostic, so... complicated.)

TL;DR: There's more issues with doing it (what I think/thought to be) the "proper" way than I initially realized (duh!) and the current solution is a fine one for the time being, until dyno/provide-any are usable.