fix(runtime-tags): treat bogus comments as comments, not text

A processing instruction or empty end tag left a gap in the scan, so the run
after it looked like text and reported as moved out of a table. Both are kept
out of the tree by the parser. An unparseable tag really does become text and
is still reported.

Author: DylanPiercey

packages/runtime-tags/src/__tests__/validate-structure.test.ts

@@ -207,6 +207,22 @@ describe("html structure validation", () => {
         html("<table><tbody><!><tr><td>x</td></tr></tbody></table>"),
       ));
 
+    it("a processing instruction is a comment, not text", () =>
+      assertClean(
+        html("<table><tbody><?pi?><tr><td>x</td></tr></tbody></table>"),
+      ));
+
+    it("an empty end tag is dropped, not text", () =>
+      assertClean(
+        html("<table><tbody></><tr><td>x</td></tr></tbody></table>"),
+      ));
+
+    it("an unparseable tag really is text and is reported", () =>
+      assertReports(
+        [html("<table><tbody><123><tr><td>x</td></tr></tbody></table>")],
+        "Text is moved out of",
+      ));
+
     it("a bogus declaration is not text", () =>
       assertClean(
         html("<table><tbody><!bogus><tr><td>x</td></tr></tbody></table>"),

packages/runtime-tags/src/html/validate-structure.ts

@@ -110,7 +110,7 @@ const closesParagraph = new Set(
 // `<!>` and other bogus declarations are comments, not text, so they must not
 // look like content the parser would move out of a table.
 const tokenSource =
-  /<!--[\s\S]*?-->|<![^>]*>|<(\/?)([a-zA-Z][^\s/>]*)((?:"[^"]*"|'[^']*'|[^>"'])*)>|([^<]+)/
+  /<!--[\s\S]*?-->|<[!?][^>]*>|<\/(?![a-zA-Z])[^>]*>|<(\/?)([a-zA-Z][^\s/>]*)((?:"[^"]*"|'[^']*'|[^>"'])*)>|([^<]+)/
     .source;
 
 export type StructureSegment =
@@ -146,8 +146,10 @@ export function createStructureValidator() {
     tokens.lastIndex = 0;
     let match: RegExpExecArray | null;
     while ((match = tokens.exec(html))) {
-      const [raw, closing, rawName, attrs, text] = match;
-      if (raw[1] === "!") continue;
+      const [, closing, rawName, attrs, text] = match;
+      // A bogus comment or declaration matches no group; the parser keeps it
+      // out of the tree, so it is neither an element nor text.
+      if (rawName === undefined && text === undefined) continue;
       while (
         segmentIndex + 1 < starts.length &&
         starts[segmentIndex + 1] <= match.index