Three parallel jobs:
- test: existing Go tests (unchanged)
- race: Go tests with -race flag (catches data races)
- quality: govulncheck + svelte-check + eslint/prettier (no PG needed)
Race job has its own PG service container so all three can run in parallel.
CI test command: switch from explicit package list to ./... so new
packages are automatically included.
local-dev-2.sh: override POSTGRES_HOST=localhost for the host-side
go run -tags dev ./main.go. The dev-tagged Connect() now reads
POSTGRES_HOST from env (needed for CI where service containers use
Docker DNS). Locally, .env sets POSTGRES_HOST=postgres, but that
name only resolves inside Docker — not from the host where the dev
server runs. Override to localhost so it connects via Docker's port
forwarding.
resetEnv() conditionally set POSTGRES_HOST only when empty. When
TestConnect_InvalidCredentials explicitly set it to 'localhost' and
then called resetEnv(), the value was preserved because it wasn't
empty. This leaked into TestConcurrentQueries, which then tried to
connect to localhost:5432 instead of the workflow-configured postgres
hostname.
Fix: capture the POSTGRES_HOST value at init() time in a package-level
variable (savedPOSTGRESHost) and always restore it in resetEnv(), so
the correct value is always used regardless of which tests ran before.
The dev-tagged Connect() in db_dev.go hardcoded localhost:5432 in the
DSN instead of reading the POSTGRES_HOST env var. Since CI tests run
with -tags "test,dev", db_dev.go is compiled and the POSTGRES_HOST=postgres
env var was silently ignored, causing db package tests to try connecting
to 127.0.0.1:5432 where no PostgreSQL is listening (service container
is only reachable via Docker DNS hostname postgres).
Also remove the -a flag from the workflow now that caching is no longer
suspected of causing issues.
popertots
