Remove failure

[datanel]

Should we consider removing failure and having custom errors implementing https://doc.rust-lang.org/std/error/trait.Error.html ?

failure wraps errors implementing https://docs.rs/failure/0.1.6/failure/trait.Fail.html not https://doc.rust-lang.org/std/error/trait.Error.html

what we might do

• create erros with struct/enum
• implement https://doc.rust-lang.org/std/error/trait.Error.html, we might use https://github.com/dtolnay/thiserror
• use https://github.com/dtolnay/anyhow

[woshilapin]

Sounds like a good idea. It'd be nice to experiment before taking a decision but I can see some limitations with the solution with have today (failure). Therefore, I'm curious about what you propose. Let's see if the situation would be better (I believe it would from what I read).

[woshilapin]

From this survey on error-handlers in Rust, it looks like anyhow is a serious solution to replace failure. One missing feature of anyhow is backtraces (but this should be soon solved since backtraces are part of Rust nightly and anyhow uses them). And thiserror seems like a very good candidate for structured errors.

These are great articles on the topic.

I was planning on experimenting anyhow with thiserror in the near future, for example, converting one of our API fully to this new pattern (I was thinking of transit_model::hellogo_fares::enrich_with_hellogo_fares). Any objection?

[datanel]

failure was replaced by anyhow #829