[Bug] After one slow ASCII width table build, overflow: 'break' wraps every later font as if each character were as wide as 'a'

未关闭 适合新手
#21,812 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
78/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃

调研方向

从 src/contain/text.ts 开始,重点关注 tryCreateASCIIWidthMap 和 measureCharWidth。启用 SLOW_MOMENT 运行链接的 Official Editor 复现,然后验证回退逻辑会测量每个字符,而不是使用 a 的宽度。完成意味着在慢速构建后,后续字体仍能正确换行和截断,并且现有的字体宽度缓存行为保持不变。

由索引模型根据 Issue 内容生成。

描述

en pending
Version

6.1.0 (the latest release, with zrender 6.1.0). zrender master has the same code. 5.6.0 wraps correctly with the same repro.

Link to Minimal Reproduction

Official Editor, ECharts 6.1.0. The same page as a standalone HTML file, with the screenshots below: https://gist.github.com/floke75/f14389932cffd44dc37f5d7ee98f12f8

Steps to Reproduce
  1. Open the editor link. The code gives ECharts a plain canvas measureText through echarts.setPlatformAPI. With SLOW_MOMENT = true, it is slow (0.2 ms a call) only while one throwaway text in another font is wrapped. This stands in for a slow device. The chart itself is laid out at full speed afterwards.
  2. Look at the x-axis labels. Each one wraps at width: 90, and the grey box shows that width.
  3. Set SLOW_MOMENT = false and run it again to compare.

It also reproduces without the hook. Open the standalone repro.html (without ?slow), set DevTools > Performance > CPU to 20x slowdown, and reload. On an Apple M3, it reproduced in every run at 10x and 20x, in some runs at 6x, and in none at 1x to 4x.

Editor code
// true: measureText is slow for one moment, a stand-in for a slow device. false: labels wrap correctly,
// unless the page itself is slow (DevTools CPU throttling at 20x on an Apple M3 reproduces it with false).
const SLOW_MOMENT = true;

const ctx = document.createElement('canvas').getContext('2d');
let slow = SLOW_MOMENT;
echarts.setPlatformAPI({
  measureText(text, font) {
    if (slow) { const t = performance.now(); while (performance.now() - t < 0.2) {} }
    ctx.font = font;
    return ctx.measureText(text);
  }
});

// Some other text on the page wraps first, in another font. When building that font's ASCII width table
// takes more than 16 ms, zrender builds no table for any later font in the page.
const other = echarts.init(document.createElement('div'), null, { width: 100, height: 50 });
other.setOption({ graphic: { type: 'text', style: { text: 'Other text', width: 80, overflow: 'break', font: '12px serif' } } });
slow = false;

// The chart itself is laid out at full speed. Each label wraps at width 90 (the grey box).
option = {
  animation: false,
  xAxis: {
    type: 'category',
    data: ['illi lilla fiil tillit', 'WWW MMM WWW'],
    axisLabel: { interval: 0, fontSize: 14, width: 90, overflow: 'break', backgroundColor: '#ddd' }
  },
  yAxis: { show: false },
  series: [{ type: 'bar', data: [3, 5] }]
};
Current Behavior

After the slow moment, the labels wrap as if every ASCII character were as wide as a (7.8 px here):

  • illi lilla fiil tillit is 77.8 px wide and fits in 90 px. It is counted as 171 px and breaks into two lines.
  • WWW MMM WWW is 122 px wide and should wrap after MMM. It is counted as 85.6 px, stays on one line and runs out of its box.
With the slow moment (?slow) No hook, DevTools CPU 20x
Labels wrapped with the width of a The same result under CPU throttling

This lasts for the rest of the page's life. Every font string that is first wrapped or truncated after the slow moment is affected.

Expected Behavior

The labels wrap as they do without the slow moment, and as 5.6.0 wraps them in both cases:

Labels wrapped correctly

Environment
- OS: macOS 26.6.1, Apple M3
- Browser: Chromium 148.0.7778.96 (Playwright), and the official editor in the same browser
- Framework: none
Any additional comments?

Where it comes from, as far as I can tell (src/contain/text.ts):

  • The first time measureCharWidth sees a font, tryCreateASCIIWidthMap measures its 128 ASCII characters into a table. If one build takes more than 16 ms, or five builds take more than 2 ms, the module-level _getASCIIWidthMapLongCount reaches its maximum. After that, no table is built for any later font in the page, and the counter never resets.
  • Without a table, measureCharWidth returns asciiCharWidth, the width of a, for every ASCII character.
  • wrapText (overflow: 'break' and 'breakAll', for plain and rich text) adds up these widths and nothing corrects them afterwards, so it breaks lines in the wrong places. Truncation uses them only for its first estimate and then measures whole strings. For truncation the fallback is the same a-width estimate 5.x used, so it is no worse than 5.x.
  • The table came in with zrender fabe0ca8 (2025-06-06, "perf(trancate)"), first released in zrender 6.0.0. Before that, wrapText measured each character with getWidth(ch, font), through the per-font string cache.

The comment above tryCreateASCIIWidthMap says "PENDING: is it necessary? Re-examine it if bad case reported.", so this may be such a case.

Suggested fix. When there is no table, measure the character itself through the font's existing string cache, as 5.x did, instead of using the width of a:

--- a/src/contain/text.ts
+++ b/src/contain/text.ts
@@ -89,7 +89,7 @@
     return (0 <= charCode && charCode <= 127)
         ? (fontMeasureInfo.asciiWidthMap != null
             ? fontMeasureInfo.asciiWidthMap[charCode]
-            : fontMeasureInfo.asciiCharWidth
+            : measureWidth(fontMeasureInfo, String.fromCharCode(charCode))
         )
         : fontMeasureInfo.stWideCharWidth;
 }

This costs at most one measureText per distinct character per font, and cache lookups after that, which is what wrapping cost in 5.x. We carry this line as a local patch to 6.1.0 in a newsroom charting tool. With it, the repro wraps correctly with the hook and under 6x and 20x throttling.

Thank you for ECharts!

主要语言
TypeScript
星标
67.7k
派生
19.9k
PR 合并指标
30 天内没有已合并 PR

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

apache/echarts 的其他 Issue

查看 apache/echarts 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。