prettier-plugin-wolfram 0.7.4 → 0.7.5

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "prettier-plugin-wolfram",
-	"version": "0.7.4",
+	"version": "0.7.5",
 	"description": "Prettier plugin for Wolfram Language using tree-sitter",
 	"license": "MIT",
 	"repository": {

package/src/parser/adapter.js

@@ -9,9 +9,12 @@ const GROUP_CLOSE_LEAF = { "}": "Token`CloseCurly", ")": "Token`CloseParen", "]"
 
 // preprocessedSource is the version passed to tree-sitter (may have ⁢ for InvisibleTimes);
 // source is the original — used only for the unformattable fallback.
-export function adapt(tree, source, preprocessedSource) {
+// map (optional) translates preprocessed char offsets back to original offsets;
+// it is attached to lineIndex so nodeSource can record exact original positions.
+export function adapt(tree, source, preprocessedSource, map) {
 	const ps = preprocessedSource ?? source;
 	const lineIndex = makeLineIndex(ps);
+	lineIndex.map = map;
 	const ctx = { source: ps, lineIndex };
 	const root = tree.rootNode;
 	if (subtreeHasError(root)) {

package/src/parser/index.js

@@ -20,10 +20,24 @@ async function getLanguage() {
 
 // Replace space-based implicit multiplication (a  b) with U+2062 (InvisibleTimes)
 // so the grammar can parse it. Skip content inside strings and nested comments.
-export function preprocessInvisibleTimes(src) {
+//
+// Returns { text, map } where `text` is the preprocessed source and `map` is an
+// index translation table: map[i] is the original-source character offset that
+// corresponds to preprocessed-text offset i (map[text.length] === src.length).
+// Because the only length-changing transform collapses a run of spaces into a
+// single InvisibleTimes char, this map lets callers translate tree-sitter node
+// positions (computed on the preprocessed text) back to exact offsets in the
+// original source, without lossy line/col round-trips.
+export function preprocess(src) {
 	let result = "";
-	let i = 0;
+	const map = [];
 	const n = src.length;
+	// Copy src[start..end) verbatim, recording the source offset of each char.
+	const copyVerbatim = (start, end) => {
+		for (let k = start; k < end; k++) map.push(k);
+		result += src.slice(start, end);
+	};
+	let i = 0;
 	while (i < n) {
 		// Skip quoted string
 		if (src[i] === '"') {
@@ -33,7 +47,7 @@ export function preprocessInvisibleTimes(src) {
 				i++;
 			}
 			if (i < n) i++;
-			result += src.slice(start, i);
+			copyVerbatim(start, i);
 			continue;
 		}
 		// Skip nested WL comment (* ... *)
@@ -46,7 +60,7 @@ export function preprocessInvisibleTimes(src) {
 				else if (src[i] === "*" && src[i + 1] === ")") { depth--; i += 2; }
 				else i++;
 			}
-			result += src.slice(start, i);
+			copyVerbatim(start, i);
 			continue;
 		}
 		// Two or more spaces between word chars on same line → InvisibleTimes
@@ -59,14 +73,22 @@ export function preprocessInvisibleTimes(src) {
 				while (j < n && src[j] === " ") j++;
 				if (j < n && /\w/.test(src[j])) {
 					result += "⁢"; // InvisibleTimes, spaces stripped (they're extras)
+					map.push(i); // the single char maps to the start of the run
 					i = j;
 					continue;
 				}
 			}
 		}
+		map.push(i);
 		result += src[i++];
 	}
-	return result;
+	map.push(n); // sentinel for end offsets (node endIndex === text.length)
+	return { text: result, map };
+}
+
+// Backward-compatible wrapper returning only the preprocessed text.
+export function preprocessInvisibleTimes(src) {
+	return preprocess(src).text;
 }
 
 export class WolframParser {
@@ -74,8 +96,8 @@ export class WolframParser {
 		const lang = await getLanguage();
 		const parser = new Parser();
 		parser.setLanguage(lang);
-		const preprocessed = preprocessInvisibleTimes(sourceText);
+		const { text: preprocessed, map } = preprocess(sourceText);
 		const tree = parser.parse(preprocessed);
-		return adapt(tree, sourceText, preprocessed);
+		return adapt(tree, sourceText, preprocessed, map);
 	}
 }

package/src/parser/position.js

@@ -55,8 +55,26 @@ export function offsetToLineCol(lineIndex, charOffset) {
 }
 
 export function nodeSource(tsNode, lineIndex) {
-	return [
+	const source = [
 		offsetToLineCol(lineIndex, tsNode.startIndex),
 		offsetToLineCol(lineIndex, tsNode.endIndex),
 	];
+	// When a preprocessing offset map is available, record the exact original
+	// character offsets (non-enumerably, so node.source stays a [[l,c],[l,c]]
+	// pair for lint rules). These bypass the lossy WL-byte/visual-column line/col
+	// round-trip in addOffsets, which otherwise mismaps offsets whenever the
+	// preprocessed text differs from the original (collapsed spaces, tabs, or
+	// non-ASCII characters earlier on the line).
+	const map = lineIndex?.map;
+	if (map) {
+		const charStart = map[tsNode.startIndex];
+		const charEnd = map[tsNode.endIndex];
+		if (typeof charStart === "number" && typeof charEnd === "number") {
+			Object.defineProperties(source, {
+				charStart: { value: charStart, enumerable: false },
+				charEnd: { value: charEnd, enumerable: false },
+			});
+		}
+	}
+	return source;
 }

package/src/utils/offsets.js

@@ -57,6 +57,14 @@ export function lineColToOffset(table, line, col) {
 
 function sourceToOffsets(source, table) {
 	if (!Array.isArray(source) || source.length !== 2) return null;
+	// Exact original char offsets recorded by the parser (see nodeSource) are
+	// authoritative when present — they avoid the lossy line/col conversion below.
+	if (
+		typeof source.charStart === "number" &&
+		typeof source.charEnd === "number"
+	) {
+		return { locStart: source.charStart, locEnd: source.charEnd };
+	}
 	const [start, end] = source;
 	if (!Array.isArray(start) || !Array.isArray(end)) return null;
 	const [startLine, startCol] = start;