1. There is a typed conversion between one type of error and another, which can keep context. 2. There are traits which can extend error types. 3. Backtraces can optionally be kept on (at a performance hit during errors of course).
I have a lot of code written that looks like this, using the excellent "failure" crate, which allows optionally to add context to any error:
file.seek(SeekFrom::Start(offset)) .context(format!("failed to seek to {} in file", offset))?;
You get explicit errors, minimal code, optional error context. It's pretty perfect.