DNS now exists for the subdomain, so retired the podsteadr.atobitcoin.io/regress/
path-prefix arrangement in favor of a clean root deployment on its own
domain: real Let's Encrypt cert (certbot certonly --webroot, same method
as podsteadr's own cert), independent nginx server blocks, no more
ROUTE_PREFIX/VITE_BASE needed for the live deploy (though still supported
and tested for future use elsewhere). Removed the now-dead /regress/
location from podsteadr's own nginx config.
Regress needs to live at podsteadr.atobitcoin.io/regress/ since / is
already podsteadr. Two coordinated pieces:
- VITE_BASE build arg (frontend asset/script paths, %BASE_URL% in
index.html for the nostr-provider.js script tag)
- ROUTE_PREFIX runtime env (backend routes registered under the prefix
via Fastify's plugin-encapsulation, including /api/health)
The nginx location must forward the prefix unstripped (proxy_pass with
no trailing path) — NIP-98 login signs the exact URL it calls, so a
stripped prefix makes the backend reconstruct a different URL than what
was signed and every login fails. Caught this with a real nginx+docker
integration test locally before it could break the live migration, then
fixed a matching bug in auth.ts (it was signing the unprefixed URL while
api.ts fetched the prefixed one). New routePrefix.test.ts proves the
prefix is actually enforced, including a negative case. 40 tests passing.
Also fixes a latent bug: import.meta.env usage had no vite/client type
reference, so it only ever passed typecheck by accident in earlier local
runs — added the standard vite-env.d.ts.
Dedicated nginx server block on :8543 (not proxied under the dashboard's
80/443) because the dashboard sets Permissions-Policy: geolocation=()
server-wide, which would silently break Regress's core claim/link
mechanic. Also documents a found-but-not-fixed gap: the dashboard's
app iframe has no allow="geolocation", so the in-dashboard launch path
won't get location access either until that shared component is patched.
ssmithx