nilsandClaude Sonnet 5.5
a36cebc827
router.ddns keeps provider records on the uplink's addresses with ddclient, run by a timer and by a watcher that starts it when an address changes. The Dynamic DNS page edits it; the secret is a private file on the router. The overview shows the running kernel and says when the current system boots a different one. PLAN.md records why the A/B image approach was not pursued. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AeFqV9dZrK6nipHkPfyM4F
62 lines
2.9 KiB
Markdown
62 lines
2.9 KiB
Markdown
# Developing
|
|
|
|
Three tools, from instant to slow. Run them from the repository root; the
|
|
flake only sees files git knows about, so `git add` new files first.
|
|
|
|
## `nix run .#webui-dev`: the web UI on your machine
|
|
|
|
Sample settings, sample DHCP leases and your real interface counters, on
|
|
<https://127.0.0.1:8443> (password `devpassword1`). Templates and CSS are
|
|
re-read on every request: edit, reload the browser. A template with a typo
|
|
keeps the last good version and logs the error. Pings need root and show "no
|
|
answer"; Apply does nothing useful here.
|
|
|
|
## `nix run .#fast`: no VMs
|
|
|
|
Go formatting, vet and tests, then builds the installer, a router built from
|
|
settings (through the release) and the home gateway. That runs shellcheck over
|
|
the scripts and evaluates the modules: most failures of the VM tests
|
|
(typos, type errors, a script that does not build) show up here, in about a
|
|
minute (half of that when little changed).
|
|
|
|
nix run .#fast -- --drivers also builds every VM test's driver: type-checks
|
|
the test scripts, builds the VMs' systems
|
|
nix run .#fast -- --all also the checks that need no VM and `nix flake check`
|
|
|
|
## `nix run .#test -- NAME...`: VM tests
|
|
|
|
One at a time (each needs gigabytes of memory), a line per test with its time,
|
|
and for a failure the step that failed and what it said; the full log is in
|
|
`${TMPDIR:-/tmp}/router-test-logs/NAME.log`.
|
|
|
|
nix run .#test -- update setup
|
|
nix run .#test -- --all
|
|
nix run .#test -- --list
|
|
|
|
`nix run .#test -- -i update` builds the test's interactive driver and gives a
|
|
Python prompt: `start_all()`, `target.succeed("...")`, `test_script()`. Boot
|
|
the VMs once and try a step again instead of running the whole test.
|
|
|
|
### Where the time goes
|
|
|
|
A VM test spends most of its time in the system under test, not in building:
|
|
`router-install` (about a minute), booting (20 s each), and everything that
|
|
runs `router-apply` on the router (building a NixOS system there takes about a
|
|
minute) or waits for a rollback timer. A source change rebuilds the systems
|
|
again on the host. So:
|
|
|
|
Measured (one at a time): `webui` 45 s, `ddns` 43 s, `basic` 107 s, `onestick` 77 s; the
|
|
ones that install and rebuild on the device are the slow ones: `apply` and
|
|
`update` several minutes, `setup` 11 minutes (three installs).
|
|
|
|
- Try a change with `webui-dev` or `fast` first; run a VM test only to confirm.
|
|
- A change to the web UI only needs `webui` (a single node, no install).
|
|
`apply`, `setup` and `update` are for what an installed router does:
|
|
on-device builds, rollback, `router-setup`, `router-update`. Run them
|
|
before a release or when you touched those, not after every UI edit.
|
|
- Nix does not rerun a test whose inputs did not change, so `--all` after a
|
|
small change is cheaper than it looks, except that the web UI is part of
|
|
every installed system: any change to it invalidates `apply`, `setup`
|
|
and `update`.
|
|
- Use `-i` to iterate on a test script.
|