Prelude & Edition 2015 import resolution #816
Conversation
| // TODO there must be a nicer way to write this condition | ||
| PathKind::Plain | PathKind::Abs | ||
| if self.edition == Edition::Edition2015 | ||
| && (path.kind == PathKind::Abs || mode == ResolveMode::Import) => |
There was a problem hiding this comment.
An alternative would be to handle this in the path desugaring layer: when converting ast::Path to hir::Path, check if the current path is use from 2015 crate and make it Abs.
On the one hand, this approach seems cleaner: we concentrate special cases in the "easy" de sugaring step, making the main logic simpler.
On the other hand, I believe we construct paths pretty frequently in the ide_api layer, and figuring edition there seems like some work.
Hm, having typed this, I now think that this PR actually does not work 100% correctly for the IDE case, because IDE uses ::Other mode even for paths in use items.
Do you think it makes sense to switch to de sugaring then?
There was a problem hiding this comment.
I think we can't completely handle the edition during desugaring, because Abs paths are handled differently between the editions. So it seems better to me to keep all the logic for this in one place instead of distributing it between the two places...
|
Ok, I made |
|
bors r+ |
816: Prelude & Edition 2015 import resolution r=matklad a=flodiebold I implemented the prelude import, but it turned out to be useless without being able to resolve any of the imports in the prelude 😅 So I had to add some edition handling and handle 2015-style imports (at least the simplified scheme proposed in rust-lang/rust#57745). So now finally `Option` resolves 😄 One remaining problem is that we don't actually know the edition for sysroot crates. They're currently hardcoded to 2015, but there's already a bunch of PRs upgrading the editions of various rustc crates, so we'll have to detect the edition somehow, or just change the hardcoding to 2018 later, I guess... ~Also currently missing is completion for prelude names, though that shouldn't be hard to add. And `Vec` still doesn't resolve, so I need to look into that.~ Co-authored-by: Florian Diebold <flodiebold@gmail.com>
(see the removed comment.)
Canceled |
|
bors r=matklad |
816: Prelude & Edition 2015 import resolution r=matklad a=flodiebold I implemented the prelude import, but it turned out to be useless without being able to resolve any of the imports in the prelude 😅 So I had to add some edition handling and handle 2015-style imports (at least the simplified scheme proposed in rust-lang/rust#57745). So now finally `Option` resolves 😄 One remaining problem is that we don't actually know the edition for sysroot crates. They're currently hardcoded to 2015, but there's already a bunch of PRs upgrading the editions of various rustc crates, so we'll have to detect the edition somehow, or just change the hardcoding to 2018 later, I guess... ~Also currently missing is completion for prelude names, though that shouldn't be hard to add. And `Vec` still doesn't resolve, so I need to look into that.~ Co-authored-by: Florian Diebold <flodiebold@gmail.com>
Build succeeded |
I implemented the prelude import, but it turned out to be useless without being able to resolve any of the imports in the prelude 😅 So I had to add some edition handling and handle 2015-style imports (at least the simplified scheme proposed in rust-lang/rust#57745). So now finally
Optionresolves 😄One remaining problem is that we don't actually know the edition for sysroot crates. They're currently hardcoded to 2015, but there's already a bunch of PRs upgrading the editions of various rustc crates, so we'll have to detect the edition somehow, or just change the hardcoding to 2018 later, I guess...
Also currently missing is completion for prelude names, though that shouldn't be hard to add. AndVecstill doesn't resolve, so I need to look into that.