fix(frontend): enhance repairFlankingEmphasis to handle opening delimiters correctly

- Update regex patterns to account for opening delimiters preceded by letters/numbers and followed by punctuation, ensuring proper rendering of emphasis in markdown.
- Add tests to validate new behavior for various edge cases, including punctuation-adjacent emphasis and exponent markers.
- Improve overall markdown rendering consistency in chat components.
This commit is contained in:
wizardchen
2026-06-22 12:30:20 +08:00
committed by lyingbug
parent 483f507104
commit 8ada130ee1
2 changed files with 29 additions and 4 deletions
@@ -128,6 +128,15 @@ test('repairFlankingEmphasis bolds punctuation-adjacent emphasis CommonMark woul
assert.equal(repairFlankingEmphasis('~~删除)~~文字'), '<del>删除)</del>文字')
assert.equal(repairFlankingEmphasis('*斜体)*文字'), '<em>斜体)</em>文字')
// Opening delimiter sits between a letter/number and punctuation -> CommonMark
// refuses to open emphasis even though the closer is valid; we bold it too.
assert.equal(
repairFlankingEmphasis('知识库中**《xxx》学程手册**整理'),
'知识库中<strong>《xxx》学程手册</strong>整理',
)
assert.equal(repairFlankingEmphasis('在**《书名》**'), '在<strong>《书名》</strong>')
assert.equal(repairFlankingEmphasis('看到~~《删》文字~~后面'), '看到<del>《删》文字</del>后面')
// Cases marked already handles, literal markers, and code must be untouched.
assert.equal(repairFlankingEmphasis('**中文**后面'), '**中文**后面')
assert.equal(repairFlankingEmphasis('**中文)** 后面'), '**中文)** 后面')
@@ -135,6 +144,10 @@ test('repairFlankingEmphasis bolds punctuation-adjacent emphasis CommonMark woul
assert.equal(repairFlankingEmphasis('`**a)**b`'), '`**a)**b`')
assert.equal(repairFlankingEmphasis('2 ** 3 ** 4'), '2 ** 3 ** 4')
assert.equal(repairFlankingEmphasis('价格 3 * 5 * 7 元'), '价格 3 * 5 * 7 元')
// Exponent / glob markers stay literal: the opener is followed by a number or
// path content, not an emphasis-opening punctuation run.
assert.equal(repairFlankingEmphasis('x**2 + y**2'), 'x**2 + y**2')
assert.equal(repairFlankingEmphasis('2**3**4'), '2**3**4')
})
test('renderChatMarkdown bolds punctuation-adjacent emphasis both mid-stream and when complete', () => {
+16 -4
View File
@@ -181,10 +181,21 @@ const FLANKING_BOLD = /(?<!\*)\*\*(?=\S)([^*\n]*?\p{P})\*\*(?=[\p{L}\p{N}])/gu
const FLANKING_STRIKE = /(?<!~)~~(?=\S)([^~\n]*?\p{P})~~(?=[\p{L}\p{N}])/gu
const FLANKING_ITALIC = /(?<![*\p{L}\p{N}])\*(?=\S)([^*\n]*?\p{P})\*(?=[\p{L}\p{N}])/gu
// Opening-delimiter variant of the rule above. CommonMark also refuses to *open*
// emphasis when the run is preceded by a letter/number and immediately followed
// by punctuation, so `知识库中**《手册》**整理` stays literal even though the
// closing `**` is fine. The `(?=\p{P})` guard keeps exponent/glob/math markers
// like `x**2`, `2**3**4`, and `**/*.js` untouched (those are followed by a
// number or slash-as-content, not an emphasis-opening punctuation run).
const FLANKING_BOLD_OPEN = /(?<=[\p{L}\p{N}])\*\*(?=\p{P})([^*\n]+?)\*\*/gu
const FLANKING_STRIKE_OPEN = /(?<=[\p{L}\p{N}])~~(?=\p{P})([^~\n]+?)~~/gu
function repairFlankingEmphasisSegment(segment: string): string {
return segment
.replace(FLANKING_BOLD, '<strong>$1</strong>')
.replace(FLANKING_BOLD_OPEN, '<strong>$1</strong>')
.replace(FLANKING_STRIKE, '<del>$1</del>')
.replace(FLANKING_STRIKE_OPEN, '<del>$1</del>')
.replace(FLANKING_ITALIC, '<em>$1</em>')
}
@@ -194,10 +205,11 @@ function repairFlankingEmphasisSegment(segment: string): string {
* CommonMark's flanking rule rejects a closing `**`/`*`/`~~` that is preceded by
* punctuation and immediately followed by a letter/number, so a very common
* model pattern like `**XBRL(…语言)**是一种` (and even ASCII `**a)**b`) renders
* as literal asterisks — both mid-stream and when complete. We convert just that
* blocked pattern (closing delimiter after punctuation, before an alphanumeric)
* into explicit HTML so it bolds reliably. Code spans/fences are skipped so
* their literal markers are untouched.
* as literal asterisks — both mid-stream and when complete. The mirror case is
* an *opening* `**`/`~~` preceded by a letter/number and immediately followed by
* punctuation (`知识库中**《手册》**整理`), which CommonMark also refuses to open.
* We convert just those blocked patterns into explicit HTML so they bold
* reliably. Code spans/fences are skipped so their literal markers are untouched.
*/
export function repairFlankingEmphasis(text: string): string {
if (!text || !/[*~]/.test(text)) return text