Let a code span carry the emphasis marks, so bold code opens at all

src/model/schema.test.ts

@@ -240,10 +240,13 @@ describe("inline", () => {
     expect(link.toJSON()).toEqual({ type: "link", attrs: { href: "./a.md", title: "A" } });
   });
 
-  it("lets a code span sit inside a link but not inside emphasis", () => {
+  it("lets a code span sit inside a link and inside emphasis", () => {
     const code = m.code.create();
     expect(code.isInSet(m.link.create({ href: "./a.md" }).addToSet([code]))).toBeTruthy();
-    expect(m.strong.create().addToSet([code])).toEqual([code]);
+    for (const outer of [m.strong, m.em, m.strikethrough]) {
+      const set = outer.create().addToSet([code]);
+      expect([outer.name, set.map((mark) => mark.type.name)]).toEqual([outer.name, [outer.name, "code"]]);
+    }
   });
 
   it("treats math and images as inline atoms", () => {

src/model/schema.ts

@@ -433,10 +433,16 @@ export const marks: { [name in MarkName]: MarkSpec } = {
     toDOM: () => ["s", 0],
   },
 
-  // A code span is literal, so it excludes the emphasis marks, but not link: [`x`](y) is valid.
+  // A code span's own content is literal, so nothing can be emphasised inside one, but a code span
+  // can itself be emphasised or linked: `**\`x\`**` and `[\`x\`](y)` are both ordinary markdown and
+  // both render on GitHub. Marks are a flat set here, so those two readings are the same set and
+  // the direction is decided once, by the serializer: src/markdown/serialize.ts puts code innermost
+  // and writes the emphasis around it. Excluding the formatting group instead, which this mark did
+  // until a document holding `**\`x\`**` could not be opened at all, is not a narrower rule but a
+  // wrong one: the bridge kept building the set the file described and every such document failed
+  // `doc.check()` on the way into the editor.
   code: {
     code: true,
-    excludes: "formatting code",
     parseDOM: [{ tag: "code" }],
     toDOM: () => ["code", 0],
   },