import { describe, expect, it } from 'vitest'
import { resolveParsedNodes, splitParagraphChildren } from '../packages/markstream-angular/src/components/shared/node-helpers'
import { renderMarkdownToHtml } from '../packages/markstream-angular/src/renderMarkdownHtml'
const REAL_WORLD_MULTILINE_INPUT = `$2.897771955 times 10^{-3}text{m·K}$^[1]^
测试[3]。
$x$^[1]^
$x$ ^[1]^
测试^[1]^
$2.897771955 \\times 10^{-3}\\text{m·K}$^[1]^
[1]
测试[12]结束
A[3]B
$x$^[1]^
测试^[1]^
[3]
测试[12]结束`
function paragraphChildren(markdown: string) {
const nodes = resolveParsedNodes({
content: markdown,
final: true,
}) as any[]
expect(nodes).toHaveLength(1)
expect(nodes[0]?.type).toBe('paragraph')
return nodes[0]?.children ?? []
}
describe('markstream-angular issue #386 paragraph segmentation regressions', () => {
it('keeps inline html embedded inside the same paragraph segment', () => {
const children = paragraphChildren('A[3]B')
const segments = splitParagraphChildren(children)
expect(segments).toHaveLength(1)
expect(segments[0]).toMatchObject({ kind: 'inline' })
expect((segments[0] as any).nodes.map((node: any) => node.type)).toEqual(['text', 'html_inline', 'text'])
})
it('keeps adjacent text around inline html in the same paragraph segment for the real issue-386 line', () => {
const children = paragraphChildren('测试[12]结束')
const segments = splitParagraphChildren(children)
expect(segments).toHaveLength(1)
expect(segments[0]).toMatchObject({ kind: 'inline' })
expect((segments[0] as any).nodes.map((node: any) => node.type)).toEqual(['text', 'html_inline', 'text'])
})
it('renders the real multiline input without leaking raw superscript syntax in html output', () => {
const html = renderMarkdownToHtml({
content: REAL_WORLD_MULTILINE_INPUT,
final: false,
})
expect(html).not.toContain('^[1]^')
expect(html).toContain('[3]')
expect(html).toContain('A[3]B')
expect(html).toContain('测试[12]结束')
expect(html).toContain('class="markstream-angular-text-node"')
})
})