Issue with init_file using anyhow instead of thiserror

[0xpr03]

Currently init_file does use anyhow which does not implement std::error::Error.
Thus it is impossible to directly convert from this type or use ? on it when using anything else than anyhow.

In my case this is stable_eyre for backtraces.
the trait std::error::Erroris not implemented foranyhow::Error``

I'd suggest using thiserror which as stated per docs is more suited to libs versus applications.

See also eyre-rs/eyre#31
and dtolnay/anyhow#28

[estk]

@0xpr03 I agree that in general thiserror is more suitable for libs. In our case, we actually cannot enumerate all of our errors since we have a plugin interface that allows people to extend log4rs (such as using syslog). So, unfortunately, it's not possible to use thiserror.

[0xpr03]

@0xpr03 I agree that in general thiserror is more suitable for libs. In our case, we actually cannot enumerate all of our errors since we have a plugin interface that allows people to extend log4rs (such as using syslog). So, unfortunately, it's not possible to use thiserror.

You could also settle for anyhow, it's the non-specific version with the same constraints.

[estk]

@0xpr03 maybe I don't understand your question but we are using anyhow: https://docs.rs/log4rs/1.0.0/log4rs/append/trait.Append.html#tymethod.append

[0xpr03]

Ah wait, I just realized my error, I mixed up another issue. I guess we'd have to somehow add the StdError trait ? I'll re-investigate this tomorrow hopefully and try to come up with a better solution. Meanwhile I'll close this prevent further noise.

[0xpr03]

What I mean was to use an std:error, fixed size approach like thiserror

[OvermindDL1]

thiserror supports having an error case of an unknown error type, can just box an stderror or take a reference to it if its possible in that situation (rarely), which is what anyhow does internally.