tinymist_query/code_action.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
use crate::{analysis::CodeActionWorker, prelude::*, SemanticRequest};
/// The [`textDocument/codeAction`] request is sent from the client to the
/// server to compute commands for a given text document and range. These
/// commands are typically code fixes to either fix problems or to
/// beautify/refactor code.
///
/// The result of a [`textDocument/codeAction`] request is an array of `Command`
/// literals which are typically presented in the user interface.
///
/// [`textDocument/codeAction`]: https://microsoft.github.io/language-server-protocol/specification#textDocument_codeAction
///
/// To ensure that a server is useful in many clients, the commands specified in
/// a code actions should be handled by the server and not by the client (see
/// [`workspace/executeCommand`] and
/// `ServerCapabilities::execute_command_provider`). If the client supports
/// providing edits with a code action, then the mode should be used.
///
/// When the command is selected the server should be contacted again (via the
/// [`workspace/executeCommand`] request) to execute the command.
///
/// [`workspace/executeCommand`]: https://microsoft.github.io/language-server-protocol/specification#workspace_executeCommand
///
/// # Compatibility
///
/// ## Since version 3.16.0
///
/// A client can offer a server to delay the computation of code action
/// properties during a `textDocument/codeAction` request. This is useful for
/// cases where it is expensive to compute the value of a property (for example,
/// the `edit` property).
///
/// Clients signal this through the `code_action.resolve_support` client
/// capability which lists all properties a client can resolve lazily. The
/// server capability `code_action_provider.resolve_provider` signals that a
/// server will offer a `codeAction/resolve` route.
///
/// To help servers uniquely identify a code action in the resolve request, a
/// code action literal may optionally carry a `data` property. This is also
/// guarded by an additional client capability `code_action.data_support`. In
/// general, a client should offer data support if it offers resolve support.
///
/// It should also be noted that servers shouldn’t alter existing attributes of
/// a code action in a `codeAction/resolve` request.
///
/// ## Since version 3.8.0
///
/// Support for [`CodeAction`] literals to enable the following scenarios:
///
/// * The ability to directly return a workspace edit from the code action
/// request. This avoids having another server roundtrip to execute an actual
/// code action. However server providers should be aware that if the code
/// action is expensive to compute or the edits are huge it might still be
/// beneficial if the result is simply a command and the actual edit is only
/// computed when needed.
///
/// * The ability to group code actions using a kind. Clients are allowed to
/// ignore that information. However it allows them to better group code
/// action, for example, into corresponding menus (e.g. all refactor code
/// actions into a refactor menu).
#[derive(Debug, Clone)]
pub struct CodeActionRequest {
/// The path of the document to request for.
pub path: PathBuf,
/// The range of the document to get code actions for.
pub range: LspRange,
}
impl SemanticRequest for CodeActionRequest {
type Response = Vec<CodeActionOrCommand>;
fn request(self, ctx: &mut LocalContext) -> Option<Self::Response> {
let source = ctx.source_by_path(&self.path).ok()?;
let range = ctx.to_typst_range(self.range, &source)?;
let root = LinkedNode::new(source.root());
let mut worker = CodeActionWorker::new(ctx, source.clone());
worker.work(root, range);
(!worker.actions.is_empty()).then_some(worker.actions)
}
}