Fix infinite loop in parser when streaming with unclosed formatting markers

[Copilot]

• Identify the root cause of browser freeze during streaming rendering
• Analyze the infinite loop in parseTextWithFormatting function
• Fix the infinite loop by ensuring currentIndex advances when unclosed formatting markers are found
• Add tests for streaming with incomplete/unclosed formatting markers
• Verify the fix with the example code from the issue
• Run existing tests to ensure no regression
• Get code review and run security checks
• Change streaming example chunk size from 50 to 10 characters
• Add GitHub Action for Surge preview deployment
• Fix additional infinite loop case with markers at string boundaries

Summary

Problem: When streaming T8 syntax rendering (incrementally adding text), the browser would freeze due to infinite loops in the parseTextWithFormatting function.

Root Causes Found:

1. Initial issue: Unclosed formatting markers (**, __, *) without closing pairs didn't advance currentIndex
2. Boundary condition issue: When boundary checks (currentIndex + 2 < textLength and currentIndex + 1 < textLength) were too restrictive, markers at positions near the end weren't checked
3. No-formatting fallback loop: When a marker existed at currentIndex but boundary conditions prevented processing it, the fallback code found the marker at the same position and didn't advance, causing infinite loop

Solutions Applied:

1. Changed boundary checks from < textLength to allow checking markers at the last positions
2. Added explicit handling in the "no formatting found" branch: when a marker is at currentIndex but can't be processed, treat it as plain text and advance past it
3. Added test case for markers at exact string boundaries

Testing:

• ✓ All 29 parser tests pass
• ✓ Streaming with 10-character chunks completes in 5ms (74 iterations)
• ✓ No infinite loops detected

Impact: This fix enables safe streaming rendering for real-time data visualization and AI-generated content scenarios.

Original prompt

当流失输出渲染的时候，会卡死浏览器，估计是 syntax 语法不算的时候有死循环。代码如下：

---

const syntax = `

2024 Smartphone Market Analysis

Market Overview

Global smartphone shipments reached [1.2 billion units](metric_value, origin=1200000000) in 2024, showing a [modest decline of 2.1%](ratio_value, origin=-0.021, assessment="negative") year-over-year.

The premium segment (devices over $800) showed remarkable [resilience](trend_desc, assessment="positive"), growing by [5.8%](ratio_value, origin=0.058, assessment="positive"). Average selling price was [$420](metric_value, origin=420, unit="USD").

Key Findings

1. Asia-Pacific remains the largest market
2. Premium devices showed strong growth
3. Budget segment faced headwinds

Asia-Pacific

Asia-Pacific remains the largest market with [680 million units](metric_value, origin=680000000) shipped, though this represents a [decline of 180 million units](delta_value, origin=-180000000, assessment="negative") from the previous year.

Key markets:

• China: [320M units](metric_value, origin=320000000) - down [8.5%](ratio_value, origin=-0.085, assessment="negative"), [ranked 1st](rank, detail=[320, 180, 90, 65, 45]) globally, accounting for [47%](contribute_ratio, origin=0.47, assessment="positive") of regional sales
• India: [180M units](metric_value, origin=180000000) - up [12.3%](ratio_value, origin=0.123, assessment="positive"), [ranked 2nd](rank, detail=[320, 180, 90, 65, 45]), representing [3 out of 4](proportion, origin=0.45) of the budget segment
• Southeast Asia: [180M units](metric_value, origin=180000000) - [stable](trend_desc, assessment="equal")

The [gap of 140M units](difference, detail=[200, 180, 160, 140]) between China and India is [narrowing](trend_desc, assessment="neutral").

Market Dynamics

Sales showed [strong correlation](association, detail=[{"x":100,"y":105},{"x":120,"y":128},{"x":150,"y":155}]) with economic indicators. The [distribution](distribution, detail=[15, 25, 35, 15, 10]) was [uneven](anomaly, detail=[15, 18, 20, 65, 22]), with [unexpected concentration](anomaly, detail=[15, 18, 20, 65, 22]) in urban areas.

We observed [clear seasonality](seasonality, detail={"data":[80, 90, 95, 135], "range":[0, 150]}) with [Q4 peaks](seasonality, detail={"data":[80, 90, 95, 135]}) driven by holiday shopping.

For detailed methodology, visit our research page.
`;

// Streaming Render
const app1 = document.getElementById('app1');
const text1 = new Text(app1!);
(async () => {
let chunk = '';
for (let i = 0; i < syntax.length; i += 50) {
chunk += syntax.slice(i, i + 50);
text1.render(chunk);
await new Promise((resolve) => setTimeout(resolve, 160));
}
})();

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[hustcc]

@copilot 将 example 中的流式渲染案例，每次增量输出 50 字符改成 10，然后添加一个 GitHub action 将 example 构建出来部署到 sugar.sh 上，便于我访问链接测试。

参考：https://github.com/antvis/GPT-Vis/blob/main/.github/workflows/preview.yml

[Copilot]

@copilot 将 example 中的流式渲染案例，每次增量输出 50 字符改成 10，然后添加一个 GitHub action 将 example 构建出来部署到 sugar.sh 上，便于我访问链接测试。...

已完成以下修改 (commit 4f3c7b1):

1. ✅ 将 example/main.ts 中的流式渲染从每次 50 字符改为 10 字符
2. ✅ 添加 GitHub Action workflow (.github/workflows/preview.yml) 用于自动部署 PR 预览到 Surge
3. ✅ 创建 vite.config.example.mjs 用于构建 example

该 workflow 会在 PR 修改 example/** 或 src/** 时自动触发，构建 example 并部署到 surge.sh，方便您直接访问链接测试。需要在仓库设置中配置 SURGE_TOKEN secret。

[github-actions]

🎊 PR Preview 69b588a has been successfully built and deployed to https://antvis-T8-preview-pr-154.surge.sh

🕐 Build time: 0.012s

_{🤖 By surge-preview}

[codecov-commenter]

Codecov Report

❌ Patch coverage is 69.23077% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.77%. Comparing base (fe8e46d) to head (69b588a).

Files with missing lines	Patch %	Lines
src/parser/syntax-parser.ts	69.23%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #154      +/-   ##
==========================================
- Coverage   79.95%   79.77%   -0.19%     
==========================================
  Files          78       78              
  Lines        1626     1656      +30     
  Branches      270      272       +2     
==========================================
+ Hits         1300     1321      +21     
- Misses        326      335       +9     

Flag	Coverage Δ
t8	79.77% <69.23%> (-0.19%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/parser/syntax-parser.ts	92.14% <69.23%> (-1.96%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hustcc]

@copilot 运行代码来看，浏览器还是卡住的。

[Copilot]

@copilot 运行代码来看，浏览器还是卡住的。

已修复！问题在于边界条件检查太严格，导致当格式标记符（如 * 或 **）出现在字符串末尾附近时无法被处理，进入了死循环。

修复内容 (commit 69b588a):

1. 调整了边界检查条件，允许检查字符串末尾的标记符
2. 在"未找到格式"分支中添加了特殊处理：当标记符在 currentIndex 但无法处理时，将其作为纯文本并前进

测试结果：

• ✅ 所有 29 个解析器测试通过
• ✅ 使用 10 字符块进行流式渲染测试（74 次迭代）在 5ms 内完成
• ✅ 未检测到无限循环

现在浏览器不会再卡住了。