errors instead of panics

[Johann150]

resolves #189

This is just a first idea of how it could look like. One point right away to consider is when some of these unwraps might fail and if the errors should be more specific then.

Something else is also that std::io::Error has to be wrapped. Maybe there is a nicer solution for that.

[brendanzab]

Nice! Thanks for looking into this. Does this remove all our unwraps?

[Johann150]

There are still unwraps in

• some tests
• the examples
• codespan-lsp, although that crate uses its own error enum

  • codespan/codespan-lsp/src/lib.rs

    Line 166 in 48e863d

    let line_span = files.line_range(file_id, line_index).unwrap();

  • codespan/codespan-lsp/src/lib.rs

    Lines 231 to 232 in 48e863d

    	let line_span = files.line_range(file_id, position.line as usize).unwrap();
    	let line_str = source.get(line_span.clone()).unwrap();

[brendanzab]

Yeah I think it's fine to ignore those files for now - the main point of the issue is to tackle the stuff in codespan-reporting.

[brendanzab]

One thing that gave me pause when attempting to do this in the past was wondering if I needed to let users define their own error types on the Files trait. I'm not sure how much of an issue that is in practice though. And we could always add that later I guess, if the need was there?

[Johann150]

We could implement something similar to Io with a generic type or a trait.

[brendanzab]

Ok, merged this! We might want to add some more descriptive information later, but I think this is good.