Roam: Convert a table to a markdown table

[cristip73]

Roam stores a table as a {{[[table]]}} marker whose children are the rows, each row's columns being a chain of first children. Nothing read that shape, so a table arrived as bullets: the marker on one line and the cells nested under it, tabular structure gone. Partially addresses #180.

The marker's children are now read as cells and written as a pipe table. The first row becomes the header, as Roam shows it; a short row is padded; a pipe or a newline in a cell is escaped so it cannot end the cell or the row; and cells go through the same markup rewrites as any other block.

Rebased and revised

This has been rewritten against src/formats/roam/convert.ts and now carries its recordings, so the change is visible in the diff rather than described.

Each point from the review:

• Unbalanced marker. The brackets are one alternation, /^\{\{(\[\[table\]\]|table)\}\}$/i, so {{[[table}} no longer matches. Covered by a test.
• Cells with more than one child. Only the first child continues the row, so anything alongside it cannot be shown. It is now a console.warn naming the cell and the count, rather than a silent drop.
• children: [] vs undefined. The detection no longer requires children. Both go through the same path and both leave nothing behind, marker included.
• Building the table. The separator is spliced in after the header and every row is formatted the same way.
• The table rendered indented under its bullet. This is the one that changed the output shape. The table is now written at the left margin with a blank line either side, whatever depth the marker sat at. Obsidian does not render a pipe table inside a list item, so keeping the outline's indentation would keep the bullets tidy and leave the table as rows of text.

One thing worth flagging: timestamp accumulation moved into its own method. Cells are walked rather than recursed into, so without it a page whose most recent edit lives inside a table would come out dated before its own content.

Recordings

tests/roam/expected/help-graph-excerpt/Table.md and tests/roam/expected/small-test-graph/Theme Tester.md are the only two that move. Nothing else in either graph changes.

Eight named tests cover the shapes the recordings do not: the bare {{table}} spelling, the unbalanced marker, a short row, a pipe in a cell, a multi-line cell, a marker with no rows, and markup inside a cell.

npm test 300 passing, npm run typecheck and npm run lint:check clean, and lint:review reports the same single pre-existing error on this file as master does.

#506 has been closed. The code-block fix that was tangled up in it will come back on its own once this lands.

[tgrosinger]

Thank you for submitting this improvement! Overall it looks great, just a few suggestions.

I also noticed when running the tests/roam/small-test-graph.json, the table is still indented under the bullet point, causing it to not render correctly. Can that be fixed here too?

[tgrosinger]

The current regex has two independent optional groups:

/^\{\{(\[\[)?table(\]\])?\}\}$/i
       ^^^^        ^^^^
       group 1     group 2

Each ? makes its group independently optional, producing 4 possible matches:

Group 1	Group 2	Matches	Valid Roam?
absent	absent	{{table}}	yes
present	present	{{[[table]]}}	yes
present	absent	{{[[table}}	no
absent	present	{{table]]}}	no

The proposed fix ties the brackets together as a single alternation:

/^\{\{(\[\[table\]\]|table)\}\}$/i

This only matches {{[[table]]}} or {{table}} -- brackets are always balanced.

Suggested change

	const roamTableRe = /^\{\{(\[\[)?table(\]\])?\}\}$/i;
	const roamTableRe = /^\{\{(\[\[table\]\]|table)\}\}$/i;

[tgrosinger]

If a Roam table cell has multiple children (e.g., user added sub-items), all but the first are silently discarded with no warning. This could cause data loss without any indication to the user. Consider logging a warning when current.children.length > 1.

[tgrosinger]

• children: [] (truthy) enters convertRoamTable, returns '' -- table marker vanishes silently
• children: undefined falls to the else branch and renders raw {{[[table]]}} in output

Consider having the table-detection guard not require json.children, and letting convertRoamTable handle all cases uniformly.

[tgrosinger]

The formatting of the various table sections could be unified and simplified by inserting the separator row after the header:

const separator = tableData[0].map(() => '---');
tableData.splice(1, 0, separator);

const lines = tableData.map(row => indent + '| ' + row.join(' | ') + ' |');

[cristip73]

Sorry for the long silence on this - thank you for the review, it was more useful than the delay suggests.

Rebased onto master and rewritten against src/formats/roam/convert.ts. All four suggestions are in, and the indentation you spotted is fixed: the table is written at the left margin now, since Obsidian will not render one inside a list item.

The test harness that arrived in the meantime made this much easier to show. Table.md and Theme Tester.md are the only recordings that move, so the before and after is in the diff rather than in a screenshot.

I have closed #506. The code-block fix that was mixed into it will come back on its own branch once this lands - and you were right about the language tags, the real problem there was Roam gluing the fences to the content, not the tags themselves.