ElyMark: tables compare by alignment and cell source, in quotes and lists too; rewritten cells stay raw (R2.13)

apps/ely-mark/web/src/preserve.js

@@ -1,5 +1,6 @@
 // Blocks Milkdown cannot write back stay raw text (§14.3).
-import { editorViewCtx, parserCtx, remarkCtx, schemaCtx, serializerCtx } from '@milkdown/kit/core'
+import { editorViewCtx, remarkCtx, schemaCtx, serializerCtx } from '@milkdown/kit/core'
+import { canon } from './tables.js'
 import { Fragment } from '@milkdown/kit/prose/model'
 import { Selection } from '@milkdown/kit/prose/state'
 import { $nodeSchema } from '@milkdown/kit/utils'
@@ -51,10 +52,11 @@ export function faithful(ctx, original) {
   const schema = ctx.get(schemaCtx)
   const serializer = ctx.get(serializerCtx)
   const written = (nodes) => serializer(schema.topNodeType.create(null, nodes)).replace(/\n+$/, '')
-  const parser = ctx.get(parserCtx)
-  // A table's spacing is free; its cells must match.
-  const kept = (node, text) =>
-    written([node]) === text || (node.type.name === 'table' && parser(text).firstChild?.eq(node))
+  // Table spacing is free; every cell's source must match.
+  const kept = (node, text) => {
+    const own = written([node])
+    return own === text || canon(own) === canon(text)
+  }
   const rawOf = ([from, to]) => {
     const text = original.slice(from, to)
     return schema.nodes.ely_raw.create(null, text ? schema.text(text) : null)

apps/ely-mark/web/src/style.js

@@ -36,7 +36,8 @@ export function style(text) {
         if (node.depth <= 2) vote('setext', head[0] !== '#')
         break
       case 'table': {
-        const rows = text.slice(at, node.position.end.offset).split('\n').map((row) => row.trim())
+        // Row nodes leave out quote and list prefixes.
+        const rows = node.children.map((row) => text.slice(row.position.start.offset, row.position.end.offset).trim())
         const padded = /\| /.test(rows[0])
         const cells = rows.flatMap((row) => row.replace(/^\||\|$/g, '').split('|'))
         const inner = (cell) => (padded ? cell.replace(/^ | $/g, '') : cell)

apps/ely-mark/web/src/tables.js

@@ -0,0 +1,30 @@
+// Tables compare by alignment and cell source, not spacing.
+import { unified } from 'unified'
+import remarkParse from 'remark-parse'
+import remarkGfm from 'remark-gfm'
+import { SKIP, visit } from 'unist-util-visit'
+
+const reader = unified().use(remarkParse).use(remarkGfm)
+
+/** A cell's own source, without its padding. */
+function cell(text, node) {
+  const kids = node.children
+  return kids.length ? text.slice(kids[0].position.start.offset, kids.at(-1).position.end.offset) : ''
+}
+
+/** `text` with each table's spacing made canonical. */
+export function canon(text) {
+  const tables = []
+  visit(reader.parse(text), 'table', (node) => {
+    tables.push(node)
+    return SKIP
+  })
+  let out = text
+  // Later tables first, so earlier offsets hold.
+  for (const table of tables.reverse()) {
+    const rows = table.children.map((row) => row.children.map((node) => cell(text, node)).join('|'))
+    const shape = [table.align.join(','), ...rows].join('\n')
+    out = `${out.slice(0, table.position.start.offset)}\u0000${shape}\u0000${out.slice(table.position.end.offset)}`
+  }
+  return out
+}