Libsyntax

[topecongiro]

Closes #1542, closes #1541, closes #1538 and closes #1537.

[nrc]

Thank you! Could you add some tests for catch blocks please?

[nrc]

I think the keywords are do catch, rather than just catch and you should explicitly emit these rather than rely on the missed span.

[topecongiro]

Thanks!

[topecongiro]

I have few questions:

1. Should do catch { ... } stay in a single line as unsafe when it is short?
2. I found an example without do (let x = match catch { ... }). How should I treat these differently?
   I just found the above example is depricated. Sorry!

[nrc]

Yeah, I think do catch should follow unsafe. I don't think we should ever line-break between do and catch

[topecongiro]

Thanks!
It would be nice if rustfmt could handle single line case inside impl Rewrite for ast::Block along with unsafe (here). However, as far as I know, ast::Block does not carry information whether it is catch block or not. Hence we need to handle single line case for catch before calling block.rewritehere. Is this correct?

[nrc]

Yeah, that is correct (and unfortunate). I'm not sure how to best address it, to be honest. Perhaps pull the code out of the Rewrite impl and into a free function with an extra flag?

[nrc]

How does this fix the comments issue? (It looks like it does fix it from the test, but I don't understand how).

[topecongiro]

The comments are inside macro_rules!, so the issue is fixed via using snippet from the span, rather than using the fallback path.

[topecongiro]

EDIT: rebased with the lastet libsyntax branch.

The latest commit implements catch, though it causes regression (please see tests/target/closure.rs).
Unfortunately, this is caused by fixing a small bug. Currently, when writing unsafe in a single line fail, rustfmt does not try multiline because it returns earlier via try_opt! (here). The commit fixes this bug by try_opt!ing in a different function.

[nrc]

Looks good, thank you!