The error variant "Could initialize the http client" provides no further detail

[Shnatsel]

Current behavior 😯

Users of cargo audit are reporting that they see an error "Could initialize the http client" with no further detail: rustsec/rustsec#1029
This makes it non-actionable for users and impossible to debug for developers.

It is defined in this code:
https://github.com/Byron/gitoxide/blob/c0e72fbadc0a494f47a110aebb46462d7b9f5664/gix-transport/src/client/blocking_io/http/traits.rs#L6-L10

Expected behavior 🤔

First off, the message should include a "not" 😉

The Display implementation for the error should also call the Display of the underlying error, so that the cause of the failed initialization could be understood and acted upon.

Steps to reproduce 🕹

No response

[Shnatsel]

As a wild guess, perhaps changing

    #[error("Could not initialize the http client")]
    InitHttpClient {
        source: Box<dyn std::error::Error + Send + Sync + 'static>,
    },

to

    #[error("Could not initialize the http client: {source}")]
    InitHttpClient {
        source: Box<dyn std::error::Error + Display + Send + Sync + 'static>,
    },

would do the trick?

But I am not familiar with thiserror so please take this with a grain of salt.

[Byron]

Thanks for letting me know about this missing word.

For good measure I assured thiserror is working as advertised and added this passing test.

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn http_error_has_source() {
        assert!(std::error::Error::source(&Error::InitHttpClient {
            source: Box::new(Error::Detail {
                description: "something".into()
            })
        })
        .is_some());
    }
}

If source() is used to generate errors, these will be displayed. anyhow will do that for instance.
Error messages that concatenate their source in the error messages aren't used in gix, instead it relies on a source() trail which is maintained by thiserror automatically.

I hope that helps.