block append --file: markdown parser drops tables and crashes on common language aliases

[4ier]

block append --file: markdown parser drops tables and crashes on common language aliases

Summary

notion block append <page> --file <foo.md> uses a very primitive "split on blank lines, one block per section" parser that:

1. Completely loses markdown tables — table rows become raw-text paragraph blocks (| a | b | appears literally on the page).
2. Fails the whole append when a fenced code block uses a common alias like ```ts or ```sh — the Notion API rejects ts, and the one bad block aborts the entire file import.

This makes --file unsuitable for real markdown documents; I had to write a custom parser + raw /v1/blocks/{id}/children calls to upload a design doc that contained 15 tables.

Environment

$ notion --version
notion version dev

$ notion block append --help | grep -i file
--file string   Read content from a file (each double-newline-separated section becomes a block)

Installed binary: ~/.local/bin/notion (10 MB, built as dev). Built from source via go install.

Repro 1 — tables are flattened

Input /tmp/t.md:

# heading

A paragraph.

| col1 | col2 |
|------|------|
| a    | b    |
| c    | d    |

- item one
- item two

Command:

notion block append <page-id> --file /tmp/t.md

Expected: 1 heading + 1 paragraph + 1 table block (with 3 rows) + 2 bulleted_list_item = 5 blocks.

Actual (inspected via /v1/blocks/{id}/children):

total: 8
types: Counter({'paragraph': 5, 'bulleted_list_item': 2, 'heading_1': 1})
  heading_1: 'heading'
  paragraph: 'A paragraph.'
  paragraph: '| col1 | col2 |'
  paragraph: '|------|------|'
  paragraph: '| a    | b    |'
  paragraph: '| c    | d    |'
  bulleted_list_item: 'item one'
  bulleted_list_item: 'item two'

Every table row ends up as a paragraph showing raw pipe characters. No table / table_row blocks are created.

Repro 2 — code language alias crashes the whole append

Input /tmp/t2.md:

# heading

Some prose.

```ts
const x = 1;

• item

**Command**:

notion block append --file /tmp/t2.md

**Actual**:

Error: append block: validation_error: body failed validation: body.children[6].code.language should be "abap", "abc", ..., "typescript", ..., instead was "ts".
→ Check your input format. Use --debug for request details

**Nothing** gets appended (whole request is rejected atomically). Common aliases users write constantly in markdown (`ts`, `js`, `py`, `sh`, `yml`, `rb`, `rs`) all hit this.

The help text and skill docs suggest normalization already happens ("'ts' / 'sh' / 'yml' etc. normalized"), but in this build it does not.

## Impact

- `--file` is essentially unusable for typical technical markdown (design docs, READMEs) because tables and fenced code with language aliases are both extremely common.
- Users have to fall back to `notion api PATCH /v1/blocks/<id>/children` and implement their own markdown parser — which defeats the purpose of the flag.

## Suggested fix

1. Use a real markdown AST parser (e.g. `goldmark` with the table extension) instead of the "blank-line split" heuristic.
2. Map CommonMark `table` nodes to Notion `table` + `table_row` blocks (set `table_width`, `has_column_header: true`).
3. Normalize language aliases to Notion's accepted set before POST:
   - `ts` → `typescript`, `js` → `javascript`, `py` → `python`, `sh`/`bash` → `shell`, `yml` → `yaml`, `rb` → `ruby`, `rs` → `rust`, `md` → `markdown`, empty → `plain text`, unknown → `plain text` (with a warning).
4. Auto-split code blocks that exceed the 2000-char rich-text limit (keep same language, consecutive blocks).
5. For consistency with `set-markdown` (mentioned in docs), consider sharing the same parser between `block append --file` and a future `page set-markdown`.

## Workaround I used

Wrote a ~200-line Python parser that:
- detects fenced code, tables, headings, lists, blockquotes, hrules in source order
- emits a Notion `children` array with proper `table`/`table_row`/`code`/heading/list/quote blocks
- splits code blocks and inline text > 2000 chars
- normalizes language aliases
- uses `notion api PATCH /v1/blocks/{page_id}/children` in batches of 80

Happy to contribute the markdown→blocks mapping logic as a Go port if useful.

## Additional notes

- `page set-markdown`, `page markdown`, `page property`, `page archive`, `comment update/delete`, `file get` (all mentioned in the v0.7 skill/README) are **not present** in my `dev` build — see companion issue on docs vs implementation sync.