Postgres spreadsheet view

A quick experiment.

1. Spin up postgres docker container
2. UI: a. Query editor in topbar -- pushes data into datagrid b. Data grid -- 2d table of input elements, matching data size c. Bottom (add new row) button
3. Edit row wise using primary key as id, push back out to database (transaction processing) a. Need row-table provenance for each cell.
4. Either:
   • Forward errors to frontend in modal + reject action, or
   • Accept edit
5. Propagate changes to other clients
   • Can use pgoutput / wal2json for notification mechanism (via web sockets)

TODO

• 4. Optimistic update to frontend + loading spinner if accept takes a while -- can do this because of 204 status -- easy heuristic

• 4. Error modals

• 5. WAL streaming to other clients.

  • pg_lineage extensions

    • column / table provenance

    • [o] row level provenance

      • rebuild queries with better edithandles (column + pk), and do filtration to present clean edithandles to users
        • fixes missing edithandle problem
  • parse WAL stream and provide edithandles for tuples
    • fixes reactivity because we reconcile on frontend via edithandle
  • switch architecture to "subscribe" to a query, and only push the relevant changes instead of the entire WAL to the client.
  • Store queries

• Add a notimplemented error in case we're editing a join.

• switch from modifying group by clause to group keys, where group keys are computed per group

  • ignore this case for now, group keys / aggregates are a particularly difficult case

1. schema introspection endpoint + ui
2. better navigation / baked (SELECT * FROM table;) / saved (SELECT [...complicated mess...]) queries
3. better collaboration (change notifications, live cursor)
4. time travel + undo (need activities table)
5. csv imports (no cap frfr)