
EmDash 1.5.0 was published on npm on 11 November 2026. On 13 and 14 November we moved seventeen sites from 1.4.0 to 1.5.0: the demos of our seven themes and the production sites built on the same core. None of them had to be rolled back. This is what the upgrade did to the databases, what we had to change in the theme code, and the order of operations we would use again.
Everything about EmDash itself comes from the 1.5.0 entry of the core changelog and from the updating guide. Everything with a number was measured on our own sites.
What 1.5.0 adds
The short list, for a theme author:
- Settings > Sidebars. Each admin user hides and reorders the navigation entries and the sections of the editor sidebar. The drag handles on editor sections are gone.
- Co-editing, off by default. It is switched on with
coEditing: true, and on Cloudflare an existing site also needs a Durable Object binding and the room classes exported from the Worker entry point. - A type check on
emdash()options. A value of the wrong type now stopsastro dev,astro buildandastro sync. - A staging mode that keeps a newly set up site out of search engines until an administrator selects Go live. Existing sites are not affected.
- A built-in SMTP provider.
- A
formatsprop on theImagecomponent, to render a<picture>with a fallback. no-schedulingas a collection support flag. The oldschedulingvalue no longer does anything.$media.filein seed files, to import a local image that sits next to the seed.
None of these is forced on an existing site. We left co-editing, SMTP and staging off on every site.
Pin the versions together
The Cloudflare adapter now declares emdash as a peer dependency at its exact matching version. The two packages move together or the install complains. Our themes pin all three EmDash packages, without a caret:
"@emdash-cms/admin": "1.5.0",
"@emdash-cms/cloudflare": "1.5.0",
"emdash": "1.5.0"
We also pin @cloudflare/kumo to 2.15.0, because the screens we add to the back office are built with the same component library as the 1.5.0 admin.
What the migration does to the database
In the default auto mode, a deployed site applies pending core migrations on its first request. So the migration does not run when you deploy: it runs when someone asks for the first page.
Measured on our sites:
- The first request after deployment took 8 to 21 seconds, depending on the site. That request carries the migration. The next ones are normal.
- The migrations table went from 90 to 96 entries. Six core migrations, 092 to 097, the last one named
097_room_storage. - The row counts of the menus, the settings and the plugin storage were identical before and after. We counted them on each side of the upgrade.
- The
revisionstable gained two columns, with the same rows. That matches the changelog:kindandedited_byare added, and existing versions are not backfilled. - Two empty tables appeared for co-editing rooms.
One consequence from the changelog deserves a line in your runbook: once a site has run the revisions migration, an earlier EmDash version can no longer export it or import into it with site transfer.
Take a real recovery point first
We expected to dump the database before each upgrade. It does not work: wrangler d1 export fails on an EmDash database, because EmDash search uses FTS5 tables. Cloudflare documents the limit: export is not supported for virtual tables, and FTS5 tables are virtual tables.
What works is D1 Time Travel. Before each deployment we noted two things:
npx wrangler d1 time-travel info <database>
npx wrangler deployments list --name <worker>
The first prints the current bookmark of the database, the second the id of the Worker version in service. Time Travel keeps 30 days of history on the paid plan and 7 on the free one, and a restore overwrites the database in place. We also keep a JSON copy of the tables, but that copy is for reading, not for restoring.
The rollback, rehearsed
We rehearsed the rollback on a copy of one demo database before touching production. The Worker went back to its previous version in 5 seconds and the database was restored to its bookmark in 6 seconds.
The order matters:
npx wrangler rollback <version-id> --name <worker> -y
# wait until /_emdash/api/health answers "version":"1.4.0"
npx wrangler d1 time-travel restore <database> --bookmark=<bookmark>
Worker first, database second. If you restore the database while the 1.5.0 Worker is still serving, its next request migrates the database again. Everything written after the bookmark is lost, which is why we picked a window with no one writing and asked editors to save their open entries first.
What a theme has to adapt
A theme that only reads content had nothing to change. Ours go further: they add screens to the back office and adjust a few native ones. That is where the work was.
- French interface. We maintain a French catalogue for the admin. After the upgrade, 387 messages were back in English: the sidebars page, SMTP, staging, co-editing and form validation. 295 were simple strings, 81 were compound messages, and 11 are identical in both languages. A self-check now fails the build if an English string is left.
- Anything hooked into the compiled admin. Our product list filter patches thirteen exact passages of the admin bundle. Twelve were unchanged in 1.5.0 and one had changed. The build stops and names the missing passage when the admin version is not the expected one, so we found it before deploying, not after.
- List thumbnails. In 1.5.0 the title cell of a list also carries a status and date reminder for narrow screens. Our thumbnail style targeted the whole cell and showed the reminder twice on wide screens. It now targets the link only.
- Active entry in the navigation. 1.5 highlights the root of a plugin on each of its sub-pages. Our own highlight was added on top, so two entries were lit. One rule fixed it.
- Menu icons. The component library now spaces the icon of a dropdown item itself. We removed a manual margin in 27 places.
And three things we checked that needed nothing: the new type check on emdash() options raised no error on our configurations, no site used the scheduling value, and no plugin of ours declares unique indexes, so the migration that changes them had nothing to repair. If one of your plugins does declare uniqueIndexes, read the maintenance window section of the updating guide before you deploy: that migration is not meant for a rolling deploy.
What changes for editors
Two things your clients will notice:
- On our sites, the Editor role no longer opens the General, Social links and SEO settings pages. They are for administrators now. If an editor used to change the site title or a social link, either do it for them or change their role. We observed this on the upgraded sites; we did not find it described in the release notes.
- Check the Menus screen. The changelog explains that deleting a menu item used to leave its nested items behind on SQLite and D1. After the upgrade we looked at each menu for leftovers and removed them.
The order we would use again
- In the site: pin the four versions, install, then type check, tests and the engine build. All three green before production is touched.
- Note the Time Travel bookmark and the Worker version id.
- Check that no plugin storage index is unique.
- Pick a window with no one writing.
- Deploy, then poll
/_emdash/api/healthuntil it reports1.5.0, and open the admin. One of the first answers is slow: that is the migration. - Verify: the migrations table ends at
097_room_storage; open the home page, a content page, a product page; in the admin, open Plugins, one plugin settings screen, a content list and Menus. - If a check fails: Worker rollback, then database restore.
Our themes shipped at 6.5.0 with EmDash 1.5.0 pinned, and today’s templates still pin 1.5.0, the latest EmDash release. The list is on the templates page and each release is recorded in the changelog.




