Show some inner errors (do not hide them with .context)

[Hocuri]

No description provided.

[flub]

Should this not work for all errors that have source? Is a simple way to do that to cast to an anyhow::Error if it not already is and then use it's :# formatter as this already does?

[Hocuri]

@flub I am afraid I don't understand what you mean.

[csb0730]

Should this not work for all errors that have source? Is a simple way to do that to cast to an anyhow::Error if it not already is and then use it's :# formatter as this already does?

Did a similar action.
Updated all occurrences where "err" is printed/logged by info! and warn! macros from {} formatter to {:?}
=> This leads all errors to speak, really ;-)

But personally I don't know the detailed difference between {:#} and {:?}

[flub]

I mean we have a bunch of errors which are not anyhow::Error but which do return something when you call Error::source(). We mostly create those by deriving thiserror:

#[derive(Debug, thiserror::Error)]
enum SomeError {
    #[error("msg")]
    SomeVariant(#[from] OrigError),
    #[error("msg")]
    OtherVarian(#[source] OtherOrigError),
}

Here both of these variants will return something when you call Error::source(). It would be nice to also show those errors with their causes/sources as well, even though they are not anyhow::Error errors. I think the simplest way for that is to first wrap them using anyhow::Error::new() and then log it using {:#} as proposed here. But I don't know how you'd know whether you have an std::error::Error here or an anyhow::Error, there probably is a way.

[Hocuri]

Now I understand.

As we are mostly using anyhow::Result<_>, most errors will be anyhow::Error. std::error::Error is just a trait, not a type/struct. The only other possibility is some special error that we create with thiserror like deltachat::configure::read_url::Error.

In VSCode + Rustanalyzer you can hover your mouse over a variable and then click Go to Error to find out which error this is. It will probaby be the same for the whole file.

Hmmmm... Do you think we should replace all 101 occurences in 24 files of {}", e with {:#}", e? Or check where anyhow::Error is used first? (should be done in 5min)

[link2xt]

I opened an issue in thiserror repository: dtolnay/thiserror#98

I would rather replace only errors which are tested with {:#}, it may not be appropriate in all cases. Causes may be not user-readable, not every library cares about formatting its causes properly.

[flub]

Perhaps for warn!() and info!() it is fine to do a bulk replacement. But error!() it would not be suitable as that's directly shown to users instead of ending up in logs.

It might be possible to write a formatter helper with a signature that accepts Into<anyhow::Error> and does the formatting. Since anyhow::Error impls From<StdError + ...> this could be enough. If that can't work we may need to look at the various downcasting and is() methods of StdError and anyhow::Error