fix: highlight the case and prql keywords and the brackets in the lezer grammar

grammars/prql-lezer/src/highlight.js

@@ -1,7 +1,14 @@
 import { styleTags, tags as t } from "@lezer/highlight";
 
-export const prqlHighlight = styleTags({
+// The selectors, separate from the `styleTags` call below so that
+// `test/test-highlight.js` can check each one resolves to a term in the
+// grammar: `styleTags` drops a selector that matches nothing without
+// reporting it.
+export const prqlHighlightSpec = {
   "CallExpression/Identifier": t.function(t.variableName),
+  // Keywords are named terms only because the grammar declares them through
+  // `kw<>`; see the note on the literal tokens in `prql.grammar`.
+  prql: t.keyword,
   module: t.moduleKeyword,
   let: t.definitionKeyword,
   case: t.controlKeyword,
@@ -33,4 +40,6 @@ export const prqlHighlight = styleTags({
   "[ ]": t.squareBracket,
   "{ }": t.brace,
   "| ,": t.separator,
-});
+};
+
+export const prqlHighlight = styleTags(prqlHighlightSpec);

grammars/prql-lezer/src/prql.grammar

@@ -19,7 +19,7 @@
 
 statements { newline* QueryDefinition? Module? Annotation? VariableDeclaration* pipelineStatement? end? }
 
-QueryDefinition { @specialize<identPart, "prql"> NamedArg+ newline+ }
+QueryDefinition { kw<"prql"> NamedArg+ newline+ }
 
 Module { kw<"module"> Identifier "{" statements "}" }
 
@@ -46,7 +46,7 @@ Declaration { DeclarationItem { identPart } "=" expression }
 DeclarationTuple { DeclarationItem { identPart } "=" expression }
 CaseBranch { expression "=>" expression }
 // Possibly we could only accept case branches inside the TupleExpression?
-CaseExpression { @specialize<identPart, "case"> TupleExpression }
+CaseExpression { kw<"case"> TupleExpression }
 
 NestedPipeline { "(" newline* Pipeline ~ambigNewline newline? ")" }
 
@@ -239,6 +239,11 @@ LambdaParam { identPart TypeDefinition? (":" expression)? }
 
   "="[@name=Equals]
 
+  // Declaring these literals here names each one after itself, which is what
+  // makes them addressable from `highlight.js` — an undeclared literal is an
+  // anonymous term, so a style tag naming it silently matches nothing.
+  "(" ")" "[" "]" "{" "}" "," "|"
+
   FString { interpolatedString<'f'> }
   RString { "r" (rawStringDouble | rawStringSingle) }
   SString { interpolatedString<'s'> }

grammars/prql-lezer/test/misc.txt

@@ -197,3 +197,24 @@ derive {
 ==>
 
 Query(Pipeline(CallExpression(Identifier,ArgList(TupleExpression(DeclarationTuple(DeclarationItem,Equals,NestedPipeline(Pipeline(Identifier,Identifier))))))))
+
+# Query definition
+
+prql target:sql.duckdb
+
+from foo
+
+==>
+
+Query(QueryDefinition(prql,NamedArg(ArgumentName,Identifier)),Pipeline(CallExpression(Identifier,ArgList(Identifier))))
+
+# Case expression
+
+derive x = case {
+  a => 1,
+  true => 2
+}
+
+==>
+
+Query(Pipeline(CallExpression(Identifier,ArgList(Declaration(DeclarationItem,Equals,CaseExpression(case,TupleExpression(CaseBranch(Identifier,Integer),CaseBranch(Boolean,Integer))))))))

grammars/prql-lezer/test/test-highlight.js

@@ -0,0 +1,95 @@
+// A style tag in `highlight.js` whose selector matches no term in the grammar
+// is silently dropped rather than reported, so these tests pin each token to
+// the tag it should carry — a selector that stops matching turns up here as a
+// failure instead of as unstyled output in a downstream editor.
+import { parser } from "../dist/index.js";
+import { prqlHighlightSpec } from "../src/highlight.js";
+import { highlightTree, tagHighlighter, tags as t } from "@lezer/highlight";
+import * as assert from "assert";
+
+// One class per tag, rather than `classHighlighter`'s groupings, so that a
+// token styled with the wrong tag of a family is still a failure.
+const highlighter = tagHighlighter(
+  [
+    t.keyword,
+    t.moduleKeyword,
+    t.definitionKeyword,
+    t.controlKeyword,
+    t.operatorKeyword,
+    t.paren,
+    t.squareBracket,
+    t.brace,
+    t.separator,
+  ].map((tag) => ({ tag, class: tag.toString() })),
+);
+
+// The tag each styled token carries, as `[text, tag]` pairs. Unstyled tokens
+// are absent, which is what an unmatched style tag produces.
+function tagged(source) {
+  const styled = [];
+  highlightTree(parser.parse(source), highlighter, (from, to, classes) =>
+    styled.push([source.slice(from, to), classes]),
+  );
+  return styled;
+}
+
+describe("highlight", () => {
+  it("tags the keywords the grammar names through `kw<>`", () => {
+    assert.deepStrictEqual(tagged("prql target:sql.duckdb\nfrom x\n"), [
+      ["prql", t.keyword.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("module m {\nfrom x\n}\n"), [
+      ["module", t.moduleKeyword.toString()],
+      ["{", t.brace.toString()],
+      ["}", t.brace.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("let a = (from x)\n"), [
+      ["let", t.definitionKeyword.toString()],
+      ["(", t.paren.toString()],
+      [")", t.paren.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("from x\nderive b = case {a => 1}\n"), [
+      ["case", t.controlKeyword.toString()],
+      ["{", t.brace.toString()],
+      ["}", t.brace.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("from x\nfilter a in b\n"), [
+      ["in", t.operatorKeyword.toString()],
+    ]);
+  });
+
+  it("tags brackets, braces and separators", () => {
+    assert.deepStrictEqual(tagged("from x\nselect {a, b}\n"), [
+      ["{", t.brace.toString()],
+      [",", t.separator.toString()],
+      ["}", t.brace.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("from x\nderive c = [1]\n"), [
+      ["[", t.squareBracket.toString()],
+      ["]", t.squareBracket.toString()],
+    ]);
+    assert.deepStrictEqual(tagged("from x | select a\n"), [
+      ["|", t.separator.toString()],
+    ]);
+  });
+
+  // The snippet cases above pin the tag each token carries, but only for the
+  // tokens they contain; this one covers every selector, including any added
+  // later for a token no snippet happens to use.
+  it("has no selector that matches no term in the grammar", () => {
+    const terms = new Set(parser.nodeSet.types.map((type) => type.name));
+    // A key holds space-separated selectors, each a `/`-separated path, with
+    // `...` standing for any intervening nodes and a trailing `!` or `*`
+    // limiting how far down the tag applies.
+    const selectorTerms = Object.keys(prqlHighlightSpec)
+      .flatMap((key) => key.split(" "))
+      .flatMap((selector) => selector.split("/"))
+      .map((part) => part.replace(/[!*]$/, ""))
+      .filter((part) => part !== "" && part !== "...");
+
+    assert.deepStrictEqual(
+      selectorTerms.filter((term) => !terms.has(term)),
+      [],
+    );
+  });
+});