* Add initial support for LSP DidChangeWatchedFiles
* Move file event Handler to helix-lsp
* Simplify file event handling
* Refactor file event handling
* Block on future within LSP file event handler
* Fully qualify uses of the file_event::Handler type
* Rename ops field to options
* Revert newline removal from helix-view/Cargo.toml
* Ensure file event Handler is cleaned up when lsp client is shutdown
The LSP spec has this to say about initialize:
> Until the server has responded to the `initialize` request with an
> `InitializeResult`, the client must not send any additional requests
> or notifications to the server.
(https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#initialize)
The spec is not really explicit about how to handle this scenario.
Before a client sends the 'initialize' request we are allowed to send an
'exit' notification, but after 'initialize' we can't send any requests
(like shutdown) or notifications (like exit). So my intepretation is
that we should forcefully close the server in this state.
This matches the behavior of Neovim's built-in LSP client:
5ceb2238d3/runtime/lua/vim/lsp.lua (L1610-L1628)
* Add `helix_lsp::client::Client::supports_feature(&self, LanguageServerFeature)`
* Extend `doc.language_servers_with_feature` to use this method as filter as well
* Add macro `language_server_with_feature!` to reduce boilerplate for non-mergeable language server requests (like goto-definition)
* Refactored most of the `find_map` code to use the either the macro or filter directly via `doc.language_servers_with_feature`
Language Servers are now configured in a separate table in `languages.toml`:
```toml
[langauge-server.mylang-lsp]
command = "mylang-lsp"
args = ["--stdio"]
config = { provideFormatter = true }
[language-server.efm-lsp-prettier]
command = "efm-langserver"
[language-server.efm-lsp-prettier.config]
documentFormatting = true
languages = { typescript = [ { formatCommand ="prettier --stdin-filepath ${INPUT}", formatStdin = true } ] }
```
The language server for a language is configured like this (`typescript-language-server` is configured by default):
```toml
[[language]]
name = "typescript"
language-servers = [ { name = "efm-lsp-prettier", only-features = [ "format" ] }, "typescript-language-server" ]
```
or equivalent:
```toml
[[language]]
name = "typescript"
language-servers = [ { name = "typescript-language-server", except-features = [ "format" ] }, "efm-lsp-prettier" ]
```
Each requested LSP feature is priorized in the order of the `language-servers` array.
For example the first `goto-definition` supported language server (in this case `typescript-language-server`) will be taken for the relevant LSP request (command `goto_definition`).
If no `except-features` or `only-features` is given all features for the language server are enabled, as long as the language server supports these. If it doesn't the next language server which supports the feature is tried.
The list of supported features are:
- `format`
- `goto-definition`
- `goto-declaration`
- `goto-type-definition`
- `goto-reference`
- `goto-implementation`
- `signature-help`
- `hover`
- `document-highlight`
- `completion`
- `code-action`
- `workspace-command`
- `document-symbols`
- `workspace-symbols`
- `diagnostics`
- `rename-symbol`
- `inlay-hints`
Another side-effect/difference that comes with this PR, is that only one language server instance is started if different languages use the same language server.
Utf-8 support has been around for a while as an unstable feature but
utf-32 is fairly new. A bunch of LS (like rust-analyzer) added this in
a pinch, but it's pretty broken right now. The performance overhead is
not very large (still a lot better than utf-16). We can switch back
once the ecosystem has matured.
Snippet text elements can contain escape sequences
that must be treated properly. Furthermore snippets
must always escape certain characters (like `}`
or `\`). The function has been updated to account
for that. `text` is now also included with
`anything` to match the grammar and can also
match empty text. To avoid infinite loops the
`non-empty` combinator has been added which is
automatically used in the `one_or_more` and
`zero_or more` combinator where the problemn would
occur.
Previously any remaining text of the snippet that could not be parsed
was ignored. This commit adds the `parse_all` function which reports
an error if any text was not consumed by the parser
* misc: missing inline, outdated link
* doc: Add new theme keys and config option to book
* fix: don't panic in Tree::try_get(view_id)
Necessary for later, where we could be receiving an LSP response
for a closed window, in which case we don't want to crash while
checking for its existence
* fix: reset idle timer on all mouse events
* refacto: Introduce Overlay::new and InlineAnnotation::new
* refacto: extract make_job_callback from Context::callback
* feat: add LSP display_inlay_hint option to config
* feat: communicate inlay hints support capabilities of helix to LSP server
* feat: Add function to request range of inlay hint from LSP
* feat: Save inlay hints in document, per view
* feat: Update inlay hints on document changes
* feat: Compute inlay hints on idle timeout
* nit: Add todo's about inlay hints for later
* fix: compute text annotations for current view in view.rs, not document.rs
* doc: Improve Document::text_annotations() description
* nit: getters don't use 'get_' in front
* fix: Drop inlay hints annotations on config refresh if necessary
* fix: padding theming for LSP inlay hints
* fix: tracking of outdated inlay hints should not be dependant on document revision (because of undos and such)
* fix: follow LSP spec and don't highlight padding as virtual text
* config: add some LSP inlay hint configs
Multicursor completions may overlap and therefore overlapping
completions must be dropped to avoid crashes. Furthermore, multicursor
edits might simply be out of range if the word before/after the cursor
is shorter. This currently leads to crashes, instead these selections
are now also removed for completions.
This commit also significantly refactors snippet transaction generation
so that tabstops behave correctly with the above rules. Furthermore,
snippet tabstops need to be carefully mapped to ensure their position
is correct and consistent with our selection semantics. Finally,
we now keep a partially updated Rope while creating snippet
transactions so that we can fill information into snippets that
depends on the position in the document.
* LSP: Support textDocument/prepareRename
'textDocument/prepareRename' can be used by the client to ask the
server the range of the symbol under the cursor which would be changed
by a subsequent call to 'textDocument/rename' with that position.
We can use this information to fill the prompt with an accurate prefill
which can improve the UX for renaming symbols when the symbol doesn't
align with the "word" textobject. (We currently use the "word"
textobject as a default value for the prompt.)
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* clippy fixes
* rustfmt
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* fix clippy from suggestions
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
---------
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
This refactors the snippet logic to be largely unaware of the rest of
the document. The completion application logic is moved into
generate_transaction_from_snippet which is extended to support
dynamically computing replacement text.
When accepting a snippet completion we automatically delete the
placeholders for now as doing so manual is quite cumbersome. In the
future we should keep these as a mark + virtual text that is
automatically removed once the cursor moves there.
* feat(ui): deprecated completions
Mark deprecated completions using strike-through
(CROSSED_OUT modifier). The deprection information
is taken either from the `deprecated` field of the
completion item or from the completion tags.
The field seems to be the older way of passing
the deprecated information and it was already
marked as deprecated for Symbol. In completion
item the field is still valid but it seems that
the LSP is moving in the general direction of using
tags for this kind of information and as such
relying on tags as well seems reasonable and
future-proof.
So far LSP always required that `PositionEncoding.characters` is an
UTF-16 offset. Now that LSP 3.17 is available in `lsp-types` request
the server to send char offsets (UTF-32) or byte offsets (UTF-8)
instead. For compatability with old servers, UTF-16 remains as the
fallback as required by the standard.
* properly handle LSP position encoding
* add debug assertion to Transaction::change
* Apply suggestions from code review
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
---------
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
The JSONRPC spec says:
> If present, parameters for the rpc call MUST be provided as a
> Structured value
https://www.jsonrpc.org/specification#parameter_structures
(Where a "Structured value" is elsewhere defined as either a map or
array.)
This change skips the serialization of the `params` field for JSONRPC
method calls and notifications if the `params` field is the `None`
variant. This fixes compatibility with LSP servers which adhere closely
to that part of the spec: `ocamllsp` in the wild.
Completion edits - either basic `insert_text` strings or structured
`text_edit`s - are assumed by the LSP spec to apply to the current
cursor (or at least the trigger point). We can use the range (if any)
and text given by the Language Server to create a transaction that
changes all ranges in the current selection though, allowing auto-
complete to affect multiple cursors.
This prevents a freeze while shutting down when using `efm-langserver`.
`efm-langserver` exits immediately after seeing a shutdown request,
without responding to the request. We block awaiting the reply to the
shutdown request which will never come, so we time out.
This change responds to any pending requests with `Err` saying that the
stream has been closed.
Language Servers may signal that they do not support a method in
the initialization result (server capabilities). We can check these
when making LSP requests and hint in the status line when a method
is not supported by the server. This can also prevent crashes in
servers which assume that clients do not send requests for methods
which are disabled in the server capabilities.
There is an existing pattern the LSP client module where a method
returns `Option<impl Future<Output = Result<_>>>` with `None` signaling
no support in the server. This change extends this pattern to the rest
of the client functions. And we log an error to the statusline for
manually triggered LSP calls which return `None`.
This change handles a language server exiting. This was a UX sore-spot:
if a language server crashed, Helix did not recognize the exit and
continued to send requests to it. All requests would timeout since they
would not receive responses. This would also hold-up Helix closing
itself down since it would try to gracefully shutdown the server which
is implemented in the LSP spec as a request.
We could attempt to automatically restart the language server on crash.
I left this for future work since that change will need to be slightly
complicated: it will need to cover the case of a language server
repeatedly crashing.
d7d0d5ffb7 resolves completion items on
the idle-timeout event. The `Completion::resolve_completion_item`
function blocks on the LSP request though, which blocks the compositor
and in turn blocks the event loop. So until the language server returns
the resolved completion item, Helix is unable to respond to keypresses
or other LSP messages.
This is typically ok since the resolution request is fast but for some
language servers this can be problematic, and ideally we shouldn't be
blocking like this anyways.
When receiving a `completionItem/resolve` request, the Volar server
sends a `workspace/configuration` request to Helix and blocks itself
on the response, leading to a deadlock. Eventually the resolve request
times out within Helix but Helix is locked up and unresponsive in that
window.
This change resolves the completion item without blocking the
compositor.
* Split helix_core::find_root and helix_loader::find_local_config_dirs
The documentation of find_root described the following priority for
detecting a project root:
- Top-most folder containing a root marker in current git repository
- Git repository root if no marker detected
- Top-most folder containing a root marker if not git repository detected
- Current working directory as fallback
The commit contained in https://github.com/helix-editor/helix/pull/1249
extracted and changed the implementation of find_root in find_root_impl,
actually reversing its result order (since that is the order that made
sense for the local configuration merge, from innermost to outermost
ancestors).
Since the two uses of find_root_impl have different requirements (and
it's not a matter of reversing the order of results since, e.g., the top
repository dir should be used by find_root only if there's not marker in
other dirs), this PR splits the two implementations in two different
specialized functions.
In doing so, find_root_impl is removed and the implementation is moved
back in find_root, moving it closer to the documented behaviour thus
making it easier to verify it's actually correct
* helix-core: remove Option from find_root return type
It always returns some result, so Option is not needed
