4119: Cache proc-macro dlls r=matklad a=edwin0cheng
This PR try to fix a deadlock in proc-macro srv by not unloading dlls.
Currently we load and unload dlls for each request, however rustc TLS is leaky , such that if we do it a lot of times, all TLS index will be consumed and it will be deadlocked inside panic (it is because panic itself is using TLS too).
Co-authored-by: Edwin Cheng <edwin0cheng@gmail.com>
4204: Use specific pattern when translating if-let-else to match r=matklad a=matklad
We *probably* should actually use the same machinery here, as we do
for fill match arms, but just special-casing options and results seems
to be a good first step.
bors r+
🤖
Co-authored-by: Aleksey Kladov <aleksey.kladov@gmail.com>
We *probably* should actually use the same machinery here, as we do
for fill match arms, but just special-casing options and results seems
to be a good first step.
4161: lsp-types 0.74 r=kjeremy a=kjeremy
* Fixes a bunch of param types to take partial progress into account.
* Will allow us to support insert/replace text in completions
Co-authored-by: kjeremy <kjeremy@gmail.com>
4193: Make it impossible to forget to add a semantic token type / modifier r=kjeremy a=matklad
bors r+
🤖
Co-authored-by: Aleksey Kladov <aleksey.kladov@gmail.com>
4173: Use core instead of std for builtin derive macros r=edwin0cheng a=edwin0cheng
Fixed#4087.
We can't use `$crate` here right now because :
1. We have to able to detect `macro` 2.0 in collecting phase for finding `rustc_builtin_macro` attrs.
2. And we have to make hygiene works for builtin derive macro.
r= @flodiebold
Co-authored-by: Edwin Cheng <edwin0cheng@gmail.com>
So e.g. if we have `fn foo<T: SomeTrait<u32>>() -> T::Item`, we want to lower
that to `<T as SomeTrait<u32>>::Item` and not `<T as SomeTrait<_>>::Item`.
4113: Support returning non-hierarchical symbols r=matklad a=kjeremy
If `hierarchicalDocumentSymbolSupport` is not true in the client capabilites
then it does not support the `DocumentSymbol[]` return type from the
`textDocument/documentSymbol` request and we must fall back to `SymbolInformation[]`.
This is one of the few requests that use the client capabilities to
differentiate between return types and could cause problems for clients.
See https://github.com/microsoft/language-server-protocol/pull/538#issuecomment-442510767 for more context.
Found while looking at #144
4136: add support for cfg feature attributes on expression #4063 r=matklad a=bnjjj
close issue #4063
4141: Fix typo r=matklad a=Veetaha
4142: Remove unnecessary async from vscode language client creation r=matklad a=Veetaha
Co-authored-by: kjeremy <kjeremy@gmail.com>
Co-authored-by: Benjamin Coenen <5719034+bnjjj@users.noreply.github.com>
Co-authored-by: veetaha <veetaha2@gmail.com>
4139: Add check for rules that no repetition which could match an empty token r=matklad a=edwin0cheng
Fix#4103
for `/ui/issues/issue-57597.rs`
This is `ParseError` of the macro rules , because it is how rustc handle it :
a58b1ed44f/src/librustc_expand/mbe/macro_rules.rs (L558)
Co-authored-by: Edwin Cheng <edwin0cheng@gmail.com>
4133: main: eagerly prime goto-definition caches r=matklad a=BurntSushi
This commit eagerly primes the caches used by goto-definition by
submitting a "phantom" goto-definition request. This is perhaps a bit
circuitous, but it does actually get the job done. The result of this
change is that once RA is finished its initial loading of a project,
goto-definition requests are instant. There don't appear to be any more
surprise latency spikes.
This _partially_ addresses #1650 in that it front-loads the latency of the
first goto-definition request, which in turn makes it more predictable and
less surprising. In particular, this addresses the use case where one opens
the text editor, starts reading code for a while, and only later issues the
first goto-definition request. Before this PR, that first goto-definition request
is guaranteed to have high latency in any reasonably sized project. But
after this PR, there's a good chance that it will now be instant.
What this _doesn't_ address is that initial loading time. In fact, it makes it
longer by adding a phantom goto-definition request to the initial startup
sequence. However, I observed that while this did make initial loading
slower, it was overall a somewhat small (but not insignificant) fraction
of initial loading time.
-----
At least, the above is what I _want_ to do. The actual change in this PR is just a proof-of-concept. I came up with after an evening of printf-debugging. Once I found the spot where this cache priming should go, I was unsure of how to generate a phantom input. So I just took an input I knew worked from my printf-debugging and hacked it in. Obviously, what I'd like to do is make this more general such that it will always work.
I don't know whether this is the "right" approach or not. My guess is that there is perhaps a cleaner solution that more directly primes whatever cache is being lazily populated rather than fudging the issue with a phantom goto-definition request.
I created this as a draft PR because I'd really like help making this general. I think whether y'all want to accept this patch is perhaps a separate question. IMO, it seems like a good idea, but to be honest, I'm happy to maintain this patch on my own since it's so trivial. But I would like to generalize it so that it will work in any project.
My thinking is that all I really need to do is find a file and a token somewhere in the loaded project, and then use that as input. But I don't quite know how to connect all the data structures to do that. Any help would be appreciated!
cc @matklad since I've been a worm in your ear about this problem. :-)
Co-authored-by: Andrew Gallant <jamslam@gmail.com>
