rustfix: Autofix try! macro in edition lint to use ? operator

[tesuji]

Now try! macro has been deprecated in #62672. I think it's time to automatically
fix try!(expr) to expr? in edition lint.

[jugglerchris]

I've got code with a load of r#try so this one's bugging me. :-)

I'm interested in working on it if no-one else is. I haven't worked on rustc so there'll be some learning for me to do first.

[taiki-e]

FYI: rustfmt already provides a feature to replace try!(expr) to expr?.: https://github.com/rust-lang/rustfmt/blob/master/Configurations.md#use_try_shorthand

[Centril]

iirc we took a fairly explicit decision to migrate to r#try!(...) because migrating to ? can break code. And we cannot be sure everyone has migrated to 2018. cc @rust-lang/lang

[tesuji]

But this is edition lint for anyone want to migrate from 2015 to 2018 edition.

[jugglerchris]

@Centril under what circumstances can ? break code compared to r#try!(...)?

[Centril]

The lowering of ? has ::from(...) and uses type inference more as compared to try!(...).

[nikomatsakis]

We discussed this in our lang team meeting on 2019-10-17. We concluded that we would prefer not to try to create such a lint. In short, nothing has changed since the Edition handover: it is still hard to get this lint correct and it doesn't seem worth the effort to maintain it (and fix the inevitable bugs). We prefer the rustfmt "normalize try!" option. Hence, I'm going to close the issue for now. Thanks for the suggestion @lzutao.

[joshtriplett]

@nikomatsakis The one downside of using rustfmt for this: it's difficult to disable the rest of rustfmt, if you don't want to combine reformatting your code with substantive changes to your code. Would it be possible to add (and document) a method for converting try! to ? without reformatting code?

[nikomatsakis]

@joshtriplett Personally, I think you should reformat the rest of your code first, and then enable the rustfmt option to convert try! to ? (I don't believe it does that by default). Of course, I'm presuming that you want to reformat your code (which I think everyone should, though I realize others will disagree). But I think that the set of people who don't want to reformat their code is fairly small -- and the set than then further want to remove r#try smaller still. This all argues that supporting a lint in the compiler is pretty overkill.

That said, I think that there used to be some tool for converting try! to ? that we used back when ? was first introduced. Probably it could be dusted off and put to use again. I don't remember much about it though -- I want to say @japaric wrote it?