nrc / error-docs

Documentation of Rust error handling

🦍🦍🦍🦍🦍🦍🦍🦍🦍🦍

• Mastodon: @nrc
• Twitter: @nick_r_cameron
• Blog: ncameron.org/blog
• Website: ncameron.org
• Pronouns: he/him

🦀🦀🦀🦀🦀🦀🦀🦀🦀🦀

error-docs's Issues

Coarse-grained error handling is a refactoring risk

The larger the scope of code handled by an error handler, the more fragile that code is

Add error-stack to ecosystem

I haven't used it myself so I can't really say anything about the viability, but wanted to let it here nevertheless.

https://crates.io/crates/error-stack

Some explanatory comment on reddit:
https://www.reddit.com/r/rust/comments/vpg49v/comment/ieppvjv/?context=3

Case studies

I want to add some case studies to the error design section. Ideas:

• compiler/parser errors
• a prototype library
• a library with network client or something
• a CLI app

Talk about failpoints

Talk about failpoints (and alternatives) in the testing section. Probably should reference the fail crate (https://docs.rs/fail/latest/fail/). I'm not actually sure failpoints are always the best solution, so this should be a discussion rather than a recommendation, but they are sometimes useful and are definitely doing something which is necessary

UnwindSafe is unrelated to `unsafe`/rust safety.

I might be worth to more clear point out that UnwindSafe is misleading, in that it is unrelated to unsafe/rust safety guarantees.

I.e. your type must be rust-safe to reuse after a catch_unwind no matter weather it implements UnwindSafe or not. If it's not it is unsound. Something which is often missed.

UnwindSafe is more like a marker indicating that something plays well with unwinding, instead of it being "rust-safe" wrt. unwinding.

To make things worse it is relatively easy to have code which doesn't play well with unwinding, but does accidentally implement UnwindSafe due *mut T being UnwindSafe if T: RefUnwindSafe... (it is that way due to *mut T often being an owning pointer, but if it's used for a &mut T-like thing you have to explicitly remember to disable it's UnwindSafe implementation).

Collected feedback

I'm using this issue to collect feedback from various places after initial publication