fix(editor): avoid parsing JSON and large text paste (#2301)

This commit is contained in:
Qiang Zhang
2026-05-09 16:15:18 +08:00
committed by GitHub
parent c57546159d
commit 1d7aaf582c
2 changed files with 130 additions and 21 deletions

View File

@@ -1,4 +1,4 @@
import { describe, it, expect, afterEach } from "vitest";
import { describe, it, expect, afterEach, vi } from "vitest";
import { Editor } from "@tiptap/core";
import StarterKit from "@tiptap/starter-kit";
import { Markdown } from "@tiptap/markdown";
@@ -60,10 +60,23 @@ function nodeText(node: JsonNode): string {
return (node.content ?? []).map(nodeText).join("");
}
function expectLiteralPaste(editor: Editor, text: string) {
editor.commands.setTextSelection(1);
const parseSpy = vi.spyOn(editor.markdown!, "parse");
const handled = paste(editor, text);
expect(handled).toBe(true);
expect(parseSpy).not.toHaveBeenCalled();
expect(editor.getText()).toBe(text);
expect(editor.getMarkdown()).toBe(text);
}
describe("markdownPaste — code block context", () => {
let editor: Editor | null = null;
afterEach(() => {
vi.restoreAllMocks();
editor?.destroy();
editor = null;
document.body.innerHTML = "";
@@ -127,4 +140,55 @@ describe("markdownPaste — code block context", () => {
// Markdown parsing produced a heading at the top.
expect(types).toContain("heading");
});
it("inserts JSON clipboard text without running the Markdown parser", () => {
editor = makeEditor({
type: "doc",
content: [{ type: "paragraph" }],
});
const json = JSON.stringify(
{
type: "issue.comment",
payload: {
title: "Paste JSON into a reply",
nested: { ok: true, count: 3 },
items: ["alpha", "beta", "gamma"],
},
},
null,
2,
);
expectLiteralPaste(editor, json);
});
it("inserts very large plain text without running the Markdown parser", () => {
editor = makeEditor({
type: "doc",
content: [{ type: "paragraph" }],
});
const text = Array.from(
{ length: 1600 },
(_, index) => `log ${index}: ${"payload".repeat(6)}`,
).join("\n");
expect(text.length).toBeGreaterThan(50_000);
expectLiteralPaste(editor, text);
});
it("does not parse oversized bracketed plain text as JSON", () => {
editor = makeEditor({
type: "doc",
content: [{ type: "paragraph" }],
});
const parseJsonSpy = vi.spyOn(JSON, "parse");
const text = `{${"not-json".repeat(7_000)}}`;
expect(text.length).toBeGreaterThan(50_000);
expectLiteralPaste(editor, text);
expect(parseJsonSpy).not.toHaveBeenCalled();
});
});

View File

@@ -12,20 +12,71 @@
* `data-pm-slice` in the HTML — this attribute is added by ProseMirror's
* own clipboard serializer. If present, the source is another ProseMirror
* editor and its HTML is structurally correct — let ProseMirror handle it.
* Otherwise, ignore the HTML and parse text/plain as Markdown.
* Otherwise, classify text/plain into one of three paths:
* - native: let ProseMirror or another extension handle it
* - literal: insert exact text without Markdown parsing
* - markdown: parse text/plain as Markdown
*
* Why not clipboardTextParser? It only runs when there's NO text/html on
* the clipboard (ProseMirror source: `let asText = !!text && !html`).
*
* Why not heuristic detection (looksLikeMarkdown / hasRichHtml)? Unreliable.
* VS Code's HTML contains <code> tags that fool rich-content detectors.
* Markdown pattern matching has too many edge cases. The data-pm-slice
* check is deterministic — no false positives.
* Markdown pattern matching has too many edge cases. Instead, the classifier
* only keeps narrow deterministic exits for editor-owned slices, code block
* context, structured plain text, and large payloads.
*/
import { Extension } from "@tiptap/core";
import { Plugin, PluginKey } from "@tiptap/pm/state";
import { Slice } from "@tiptap/pm/model";
const LARGE_PASTE_TEXT_THRESHOLD = 50_000;
type PasteMode = "native" | "literal" | "markdown";
interface PasteClassificationInput {
text: string;
html: string;
hasFiles: boolean;
isInsideCodeBlock: boolean;
}
function isJsonDocumentText(text: string): boolean {
const trimmed = text.trim();
if (!trimmed) return false;
const startsLikeJson =
(trimmed.startsWith("{") && trimmed.endsWith("}")) ||
(trimmed.startsWith("[") && trimmed.endsWith("]"));
if (!startsLikeJson) return false;
try {
JSON.parse(trimmed);
return true;
} catch {
return false;
}
}
function isStructuredPlainText(text: string): boolean {
return isJsonDocumentText(text);
}
function classifyPaste({
text,
html,
hasFiles,
isInsideCodeBlock,
}: PasteClassificationInput): PasteMode {
if (hasFiles) return "native";
if (!text) return "native";
if (isInsideCodeBlock) return "literal";
if (html && html.includes("data-pm-slice")) return "native";
if (text.length > LARGE_PASTE_TEXT_THRESHOLD) return "literal";
if (isStructuredPlainText(text)) return "literal";
return "markdown";
}
export function createMarkdownPasteExtension() {
return Extension.create({
name: "markdownPaste",
@@ -40,29 +91,23 @@ export function createMarkdownPasteExtension() {
const clipboard = event.clipboardData;
if (!clipboard) return false;
// If clipboard has files, defer to the fileUpload extension.
if (clipboard.files?.length) return false;
const text = clipboard.getData("text/plain");
if (!text) return false;
// If the caret is inside a code block, insert the text as-is.
// Code blocks must keep newlines literal; running Markdown
// parsing here would split a blank line (\n\n) into two
// paragraphs and tear the code block open. (#1982)
const html = clipboard.getData("text/html");
const { $from } = view.state.selection;
if ($from.parent.type.name === "codeBlock") {
const mode = classifyPaste({
text,
html,
hasFiles: Boolean(clipboard.files?.length),
isInsideCodeBlock: $from.parent.type.name === "codeBlock",
});
if (mode === "native") return false;
if (mode === "literal") {
view.dispatch(view.state.tr.insertText(text));
return true;
}
const html = clipboard.getData("text/html");
// If HTML contains data-pm-slice, the source is another
// ProseMirror editor — let ProseMirror use its native HTML
// clipboard path to preserve exact node structure.
if (html && html.includes("data-pm-slice")) return false;
// Everything else (VS Code, text editors, .md files, terminals,
// web pages): parse text/plain as Markdown.
const json = editor.markdown.parse(text);