Info
-----
* As discussed in #6452, the `Error('busy')` workaround for semantic token flickering was removed because the underlying issue was believed to be fixed in VS Code.
* It turns out that the fix isn't yet complete, so this caused flickering of the semantic highlighting when making rapid edits (e.g. typing quickly).
* This PR restores that workaround and makes it slightly more robust, covering all areas of semantic token middleware.
Changes
-----
* Added middleware functions for `provideDocumentSemanticTokens`, `provideDocumentSemanticTokensEdits`, and `provideDocumentRangeSemanticTokens` to match the 3 possible middleware hooks defined in https://github.com/microsoft/vscode-languageserver-node/blob/master/client/src/common/semanticTokens.ts#L33
* Each intercepts a `null` or `undefined` return and throws an error with the message `busy` instead, which prevents the tokens from being removed and re-added (causing the flickering behavior)
Tested
-----
* Tested locally that the flickering behavior is gone.
* There don't appear to be any significant tests of the VS Code plugin side of things, other than that it loads. Is there somewhere I can / should add tests to cover this behavior?
Co-authored-by: Charles Pierce <cpierce.grad@gmail.com>
bors[bot] [Wed, 4 Nov 2020 12:37:29 +0000 (12:37 +0000)]
Merge #6456
6456: Support record variants in extract_struct_from_enum_variant r=matklad a=Veykril
As requested :)
This also prevents the assist from being disabled if a definition in the value namespace exists with the same name as our new struct since that won't cause a collision
bors[bot] [Wed, 4 Nov 2020 12:30:09 +0000 (12:30 +0000)]
Merge #6458
6458: Qualify trait impl created by add_custom_impl assist r=matklad a=Veykril
When we find at least one trait with the same name as the derive accessible from the current module we now generate a qualified path to that trait in the generated impl.
If we don't find any we just do what was done before and emit the trait name in the generated impl.
Aleksey Kladov [Tue, 3 Nov 2020 18:49:15 +0000 (19:49 +0100)]
Fix overflow panic in convert_interger_literal assist
This also seizes the opportunity to move integer literal parsing to
the syntax crate, were it logically belongs.
Note though that this is still done in an ad hoc manner -- we probably
should split kitchen sink ast::Literal into a separate APIs for
strings, ints, etc
bors[bot] [Tue, 3 Nov 2020 17:34:59 +0000 (17:34 +0000)]
Merge #6287
6287: Don't replace entire module and file nodes when inserting imports r=matklad a=Veykril
This change minifies the resulting diff of import insertions by inserting or replacing the produced use tree directly through an `action` return value instead replacing the entire container node. This action has to be applied by the caller now. This unfortunately pulls the `AssistBuilder` into scope of `insert_use` back again but I tried to at least keep it away from the `insert_use` fn itself.
bors[bot] [Tue, 3 Nov 2020 16:43:26 +0000 (16:43 +0000)]
Merge #6401
6401: Only show `self` ident when showing parameter self hints r=matklad a=Veykril
This just hints all self parameters with the `self` token, this is therefor equal to how all other parameters are displayed, but given the self param special in how its defined in a function signature it might make sense to keep the `&`/`&mut` parts as well as emitting those tokens for explict `Self` types that are taken by ref like `self: &Rc<Self>`?
bors[bot] [Tue, 3 Nov 2020 16:36:34 +0000 (16:36 +0000)]
Merge #6435
6435: Test Fixture ExplicitRoot + ModuleResolutionRelativePathOutsideRoot. r=matklad a=rickvanprim
Updates `module_resolution_relative_path_outside_root` test to check valid paths outside of the root, by moving the root to a subpath so that paths outside of it are possible. If this would be more appropriate as a new test, or if the original check for an invalid path should be left, I'm happy to update.
Co-authored-by: James Leitch <rickvanprim@gmail.com>
bors[bot] [Tue, 3 Nov 2020 07:54:45 +0000 (07:54 +0000)]
Merge #6430
6430: Move completions rendering into a separate module r=popzxc a=popzxc
This PR extracts rendering-related things from `Completions` structure to the new `render` module.
`render` module declares a `Render` structure (which is a generic renderer interface), `RenderContext` (interface for data/methods not required for completions generating, but required for rendering), and a bunch of smaller `*Render` structures which encapsulate logic behind rendering a certain item.
This is just a step in full separation direction, since the following this are still to be done:
- Move some data from `CompletionContext` to the `RenderContext`;
- Forbid any kind of rendering outside of `render` module;
- Extract score computing into a separate module.
This PR is already pretty big, so not to make it even harder to review I decided to split this process into several subsequent PRs.
the scope `meta.function.definition.rust` will continue past the end of the block looking for `{` after `fn`. This PR allows `meta.function.definition.rust` to terminate with `;` in these cases.
bors[bot] [Tue, 3 Nov 2020 07:36:49 +0000 (07:36 +0000)]
Merge #6421
6421: Check for allow(..) attributes in case check diagnostic r=popzxc a=popzxc
Resolves #6348
This is not a full-fledged solution, as it doesn't looks up for parent elements (e.g. function -> module -> parent module -> crate root), but it does at least checks attributes of item being checked.
I played a bit with code, and it seems that implementing a proper solution (which will also check for `deny` / `warn` attributes overriding values for `allow`s from above).
So, this solution should fix all the macros which intentionally do "weird" naming and wrap it with `allow`, such as `lazy_static`.
bors[bot] [Mon, 2 Nov 2020 17:47:08 +0000 (17:47 +0000)]
Merge #6365
6365: Do insertion lookahead in algo::diff r=matklad a=Veykril
This is the last blocker for #6287 after this I can update that PR to properly fix things through using `SyntaxRewriter`.
This PR also shuffles tests around a bit and adds some more.
Ideally this is just a hack until we implement a "proper" diff algorithm that approximates a minimal diff. Maybe something like [gumtree](https://github.com/GumTreeDiff/gumtree)?
6399: Keep generic annotations when qualifying things r=matklad a=Veykril
The `qualify_path` assists currently eats up already annotated generics in all but one cases which can be annoying if one already pre-fills generics of a type before it's been qualified.
Co-authored-by: Matthew Sanetra <matthewsanetra@gmail.com> Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
This reduces the number of progress bar updates we send to the client by collapsing subsequent updates into one. This doesn't work as well as I'd hoped (which is that we end up sending *no* updates, or only `start` and `end`, when the cache is already fresh), but it does reduce the number considerably: instead of ~720 updates on the rust-analyzer codebase, we now only send ~60.
It uses the same approach that is already in use for coalescing VFS events.
Hopefully this is enough to fix https://github.com/rust-analyzer/rust-analyzer/issues/6413.
Co-authored-by: Jonas Schievink <jonas.schievink@ferrous-systems.com>
Aleksey Kladov [Fri, 30 Oct 2020 18:38:29 +0000 (19:38 +0100)]
Respond with JSON-RPC error if we failed to deserialize request
Historically, we intentinally violated JSON-RPC spec here by hard
crashing. The idea was to poke both the clients and servers to fix
stuff.
However, this is confusing for server implementors, and falls down in
one important place -- protocol extension are not always backwards
compatible, which causes crashes simply due to version mismatch. We
had once such case with our own extension, and one for semantic
tokens.
So let's be less adventerous and just err on the err side!
bors[bot] [Tue, 27 Oct 2020 09:51:49 +0000 (09:51 +0000)]
Merge #6376
6376: Avoid impls_fnonce to return true when the trait solving is ambiguous r=flodiebold a=GrayJack
This PR should fix #6375
This adds a variation of `method_resolution::implements_trait` called `method_resolution::implements_trait_unique`, that only returns true when the trait solving is unique, and also change `impls_fnonce` to use the later instead.