feat: Virtual macro files #19130
feat: Virtual macro files #19130
Conversation
rustbot
added
the
S-waiting-on-review
|
This is extremely cool! Although without formatting the files are pretty unreadable, you can still see IDE features work in them. BTW, with this feature does search search in macro expansions as well? |
Yes, unfortunately we can't use rustfmt, but I believe we can at least fix up some whitespace and record those whitespace insertions to map things, that should make it a bit nicer to look at.
In theory we could do that, in practice it would make everything crawl to a halt. (we'd need to collect all relevant macro calls and well then search through all of them) |
Why not? I thought about maintaining some kind of mapping, token range before formatting to range after formatting, and using it in all features. Creating the mapping should be easy enough given that the output is just the input plus some whitespaces. |
Is it? I am not sure it is. The other issue is that we preserve |
Veykril
force-pushed
the
push-pzkwlqyxvnvx
branch
from
d7f120b to
3dbf138
Compare
| pub(crate) transform: PositionTransform, | ||
| } | ||
|
|
||
| impl LineIndex { | ||
| pub(crate) fn line_col(&self, mut offset: TextSize) -> ide::LineCol { | ||
| if !self.transform.insertions.is_empty() { | ||
| offset += TextSize::new( | ||
| self.transform | ||
| .insertions | ||
| .iter() | ||
| .copied() | ||
| .take_while(|&(off, _)| off < offset) | ||
| .map(|(_, len)| ws_kind_width(len)) | ||
| .sum::<u32>(), | ||
| ); | ||
| } | ||
| self.index.line_col(offset) | ||
| } | ||
|
|
||
| pub(crate) fn line(&self, line: u32) -> Option<ide::TextRange> { | ||
| let mut range = self.index.line(line)?; | ||
| if !self.transform.insertions.is_empty() { | ||
| let mut iter = self.transform.insertions.iter(); | ||
| let mut nl_seen = 0; | ||
| let overall_sub = TextSize::new( | ||
| iter.peeking_take_while(|&&(_, ws)| { | ||
| let m = nl_seen != line; | ||
| if ws == PrettifyWsKind::Newline { | ||
| nl_seen += 1; | ||
| } | ||
| m | ||
| }) | ||
| .copied() | ||
| .map(|(_, len)| ws_kind_width(len)) | ||
| .sum::<u32>(), | ||
| ); | ||
| let end_sub = TextSize::new( | ||
| iter.copied() | ||
| .take_while(|&(_, ws)| ws != PrettifyWsKind::Newline) | ||
| .map(|(_, len)| ws_kind_width(len)) | ||
| .sum::<u32>(), | ||
| ); | ||
| range = | ||
| TextRange::new(range.start() - overall_sub, range.end() - overall_sub - end_sub); | ||
| } | ||
| Some(range) | ||
| } | ||
| } | ||
|
|
||
| #[derive(Default)] | ||
| pub struct PositionTransform { | ||
| pub insertions: Vec<(TextSize, PrettifyWsKind)>, | ||
| } | ||
|
|
||
| fn ws_kind_width(ws: PrettifyWsKind) -> u32 { | ||
| match ws { | ||
| PrettifyWsKind::Space => 1, | ||
| PrettifyWsKind::Indent(indent) => 4 * (indent as u32), | ||
| PrettifyWsKind::Newline => 1, | ||
| } |
There was a problem hiding this comment.
I messed up my commits so the whitespace fixup mapping stuff didnt get its own commit unfortunately but this is the relevant part (super in efficient as one can tell as I hacked it together real quick). There is also an off by one error here somewhere as some offsets seem to not map cleanly causing things to break
Veykril
force-pushed
the
push-pzkwlqyxvnvx
branch
from
3dbf138 to
a95bc49
Compare
Very WIP
This allows a client to opt-into functionality to treat macro expansions as virtual files (something r-a already does, but not yet exposes). Notably this means go to def onto a reference to a definition defined within a macro expansion will now jump into the macro expansion (when the feature is enabled).
This means upmapping has to be handled by the caller (the one who invokes the IDE features now), making that property lazy in a sense. The one annoyance is that this exposes
MacroFileIds which are not necessarily stable.Some missing things:
NavigationTargetwrt to upmapping behaviorFollow up feature to this to be implemented, new command to enter into a macro expansion
(see https://bsky.app/profile/lukaswirth.bsky.social/post/3lhqqelx6hs2h for a first prototype)