Better
- Postgres and Supabase columns keep their exact types. Copying between the two now recreates
text[],integer,numeric,jsonband friends as they are, with the original column names. Elsewhere, lists stay lists wherever the destination can hold one. - Empty values stay empty. A null going into a nullable column lands as null, not an empty string.
- The workspace toolbar separates status from actions. Pull progress shows in one status chip, and the buttons next to it keep their normal look while a pull runs.
Under the hood
- Large pulls, publishes and syncs lean far less on the database. On a 110k-record folder, bulk lookups went from several seconds to tens of milliseconds.
- More database CPU and worker memory in production, so one big job can't slow down everyone else.
- A new debug view in
scratch-gitshows which operations are holding or waiting on a repo, for faster diagnosis when a publish looks stuck. - Docker images and desktop builds for Linux and Windows are pinned to x86-64 runners, and heavy CI jobs are tagged so light ones can run on small runners.