Bump to r-lib/tree-sitter-r@cbfa8fb

An empty `()`, or `()` with >1 expressions is now a syntax error, which is good, but does make it harder for us to give a precise error message. When we have a custom R parser that can report the exact problem in the parse result, we can do much better. For now it's a lot of headache and shooting a moving target.

Author: DavisVaughan

Cargo.lock

@@ -47,7 +47,7 @@ stdext = { path = "../stdext" }
 tokio = { version = "1.26.0", features = ["full"] }
 tower-lsp = "0.19.0"
 tree-sitter = "0.23.0"
-tree-sitter-r = { git = "https://github.com/r-lib/tree-sitter-r", rev = "9a63f592f0d8736912b474260a49931766542ae4" }
+tree-sitter-r = { git = "https://github.com/r-lib/tree-sitter-r", rev = "cbfa8fbce269d6e1c46bddc5c4fcc4cdee6c3e9f" }
 uuid = "1.3.0"
 url = "2.4.1"
 walkdir = "2"

crates/ark/Cargo.toml

@@ -690,25 +690,11 @@ fn recurse_parenthesized_expression(
     context: &mut DiagnosticContext,
     diagnostics: &mut Vec<Diagnostic>,
 ) -> Result<()> {
-    let mut n = 0;
-    let mut cursor = node.walk();
-
-    for child in node.children_by_field_name("body", &mut cursor) {
-        recurse(child, context, diagnostics)?;
-        n = n + 1;
-    }
-
-    if n > 1 {
-        // The tree-sitter grammar allows multiple `body` statements, but we warn
-        // the user about this as it is not allowed by the R parser.
-        let range = node.range();
-        let range = convert_tree_sitter_range_to_lsp_range(context.contents, range);
-        let message = format!("Expected at most 1 statement within parentheses, found {n}.");
-        let diagnostic = Diagnostic::new_simple(range, message);
-        diagnostics.push(diagnostic);
-    }
-
-    ().ok()
+    let Some(body) = node.child_by_field_name("body") else {
+        // Would be unexpected, grammar requires exactly 1 `body`
+        return Ok(());
+    };
+    recurse(body, context, diagnostics)
 }
 
 /// Default recursion for arguments of a call-like node

crates/ark/src/lsp/diagnostics.rs

@@ -73,39 +73,17 @@ fn recurse_children(
 }
 
 fn syntax_diagnostic(node: Node, context: &DiagnosticContext) -> anyhow::Result<Diagnostic> {
-    if let Some(diagnostic) = syntax_diagnostic_missing_open(node, context)? {
-        return Ok(diagnostic);
-    }
+    // We used to try and analyze the `ERROR` structure of the tree to provide "precise"
+    // error messages, but this is extremely prone to false positives due to the fact that
+    // tree-sitter reports a "generic" error range, and it's up to us to try and infer the
+    // problem based on the surrounding nodes, which tree-sitter often "recovers"
+    // incorrectly. It's better to wait for us to have a custom R pratt parser, which
+    // would be able to report precise error locations/messages, because it "knows" why a
+    // parse error occurred.
 
     Ok(syntax_diagnostic_default(node, context))
 }
 
-// Use a heuristic that if we see a syntax error and it just contains a `)`, `}`, or `]`,
-// then it is probably a case of missing a matching open token.
-fn syntax_diagnostic_missing_open(
-    node: Node,
-    context: &DiagnosticContext,
-) -> anyhow::Result<Option<Diagnostic>> {
-    let text = context.contents.node_slice(&node)?;
-
-    let open_token = if text == ")" {
-        "("
-    } else if text == "}" {
-        "{"
-    } else if text == "]" {
-        "["
-    } else {
-        // Not an unmatched closing token
-        return Ok(None);
-    };
-
-    let range = node.range();
-
-    Ok(Some(new_missing_open_diagnostic(
-        open_token, range, context,
-    )))
-}
-
 fn syntax_diagnostic_default(node: Node, context: &DiagnosticContext) -> Diagnostic {
     let range = node.range();
     let row_span = range.end_point.row - range.start_point.row;
@@ -309,15 +287,6 @@ fn diagnose_missing_close(
     Ok(())
 }
 
-fn new_missing_open_diagnostic(
-    open_token: &str,
-    range: Range,
-    context: &DiagnosticContext,
-) -> Diagnostic {
-    let message = format!("Unmatched closing delimiter. Missing an opening '{open_token}'.");
-    new_syntax_diagnostic(message, range, context)
-}
-
 fn new_missing_close_diagnostic(
     close_token: &str,
     range: Range,
@@ -436,8 +405,17 @@ identity(1)
         let diagnostic = diagnostics.get(0).unwrap();
         insta::assert_snapshot!(diagnostic.message);
 
+        // tree-sitter grammar knows R doesn't allow empty `()` without a body
         let diagnostics = text_diagnostics("()");
-        assert!(diagnostics.is_empty());
+        assert_eq!(diagnostics.len(), 1);
+        let diagnostic = diagnostics.get(0).unwrap();
+        insta::assert_snapshot!(diagnostic.message);
+
+        // tree-sitter grammar knows R doesn't allow >1 expressions in `()`
+        let diagnostics = text_diagnostics("(1+2\n1+2)");
+        assert_eq!(diagnostics.len(), 1);
+        let diagnostic = diagnostics.get(0).unwrap();
+        insta::assert_snapshot!(diagnostic.message);
 
         let diagnostics = text_diagnostics("( 1 + 2 )");
         assert!(diagnostics.is_empty());

crates/ark/src/lsp/diagnostics_syntax.rs

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '{'.
+Syntax error

crates/ark/src/lsp/snapshots/ark__lsp__diagnostics__tests__mixed_syntax_and_semantic_diagnostics.snap

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics_syntax.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '('.
+Syntax error

crates/ark/src/lsp/snapshots/ark__lsp__diagnostics_syntax__tests__error_precision.snap

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics_syntax.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '{'.
+Syntax error

crates/ark/src/lsp/snapshots/ark__lsp__diagnostics_syntax__tests__unmatched_binary_operator.snap

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics_syntax.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '('.
+Syntax error

crates/ark/src/lsp/snapshots/ark__lsp__diagnostics_syntax__tests__unmatched_closing_token-2.snap

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics_syntax.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '['.
+Syntax error

crates/ark/src/lsp/snapshots/ark__lsp__diagnostics_syntax__tests__unmatched_closing_token-3.snap

@@ -2,4 +2,4 @@
 source: crates/ark/src/lsp/diagnostics_syntax.rs
 expression: diagnostic.message
 ---
-Unmatched closing delimiter. Missing an opening '{'.
+Syntax error