BlockMd class abstract

Block component of the legacy regex pipeline.

A subclass is reachable only through components, which selects the legacy pipeline: the document is rendered as one text tree, with no incremental segment cache, no span-level streaming reveal and no lazy sliver, so a streaming reply re-parses and re-lays-out the whole message on every append. Register the block with MarkdownBlockComponent instead, using FencedBlockSyntax or a MarkdownBlockSyntax subclass for the syntax; the parsed node is cached, so a rebuild does not run the syntax again.

Before:

class CalloutMd extends BlockMd {
  @override
  String get expString => r':::(\w+)\n([\s\S]*?)\n:::';

  @override
  Widget build(
    BuildContext context,
    String text,
    GptMarkdownConfig config,
  ) {
    final match = exp.firstMatch(text);
    return Row(
      children: [
        Icon(
          match?.group(1) == 'warning' ? Icons.warning : Icons.info,
        ),
        Flexible(child: Text(match?.group(2) ?? '')),
      ],
    );
  }
}

GptMarkdown(
  text,
  components: [CalloutMd(), ...MarkdownComponent.globalComponents],
)

After:

GptMarkdown(
  text,
  blockComponents: [
    MarkdownBlockComponent(
      syntax: const FencedBlockSyntax(
        type: 'callout',
        opening: ':::warning',
      ),
      builder: (context, node, config) => Row(
        children: [
          const Icon(Icons.warning),
          Flexible(child: Text(node.body)),
        ],
      ),
    ),
  ],
)