Local dev environment for steakweb

[sparkyfen]

Runs steakweb locally against a throwaway Postgres, with no prod DB or real SAML IdP. I used it to test the activation and directory changes.

Builds on #1 — its test checklist assumes those fixes are in, so merge #1 first. Until it lands, this PR also lists #1's commits; they drop off once #1 is merged, leaving just the dev-env commit.

Everything's under dev/:

• docker-compose.yml — Postgres on 127.0.0.1:5433, loads the schema and seed on first boot.
• schema.sql — reconstructed registered_extensions, since there's no schema in the repo. publish/provisioned are char(1) holding 't'/'f', because that's what the code binds — a real bool column would make asyncpg reject the string.
• seed.sql — a handful of rows to test against: published and unpublished, a name with HTML in it, an out-of-range number, and one that triggers the "already taken" path.
• config.dev.json — points at the local DB. config.json stays gitignored.
• README.md — setup and what to click.

The only thing touched outside dev/ is a STEAKWEB_DEV-gated block in steakweb.py: a /dev/login that fakes a session, a TCP bind, and skipping the SAML metadata fetch. It's all behind os.environ.get('STEAKWEB_DEV'), so prod is unchanged.

The schema's a guess from reading the code, not the real one — swap it in if you have it.

[supersat]

Can you rebase on what is in main now?

I can also provide a definitive schema:

CREATE TYPE public.switch_address_type AS ENUM (
    'AAR',
    'ENP',
    'sipv4',
    'sipv6',
    'other'
);

CREATE TYPE public.switch_type AS ENUM (
    'tdm',
    'isr',
    'sip'
);

CREATE TABLE public.registered_extensions (
    extn integer NOT NULL,
    name text,
    userid integer,
    switch integer,
    port text,
    auth_code text,
    reservation_time timestamp without time zone DEFAULT CURRENT_TIMESTAMP,
    last_updated timestamp without time zone,
    expires_at timestamp without time zone,
    publish boolean DEFAULT true NOT NULL,
    provisioned boolean DEFAULT true NOT NULL
);

CREATE TABLE public.switches (
    id integer NOT NULL,
    type public.switch_type NOT NULL,
    address text,
    address_type public.switch_address_type NOT NULL,
    name text,
    description text,
    sip_gateway text
);

CREATE SEQUENCE public.switches_id_seq
    AS integer
    START WITH 1
    INCREMENT BY 1
    NO MINVALUE
    NO MAXVALUE
    CACHE 1;

ALTER TABLE ONLY public.switches ALTER COLUMN id SET DEFAULT nextval('public.switches_id_seq'::regclass);

[sparkyfen]

Can you rebase on what is in main now?

I can also provide a definitive schema:

CREATE TYPE public.switch_address_type AS ENUM (
    'AAR',
    'ENP',
    'sipv4',
    'sipv6',
    'other'
);

CREATE TYPE public.switch_type AS ENUM (
    'tdm',
    'isr',
    'sip'
);

CREATE TABLE public.registered_extensions (
    extn integer NOT NULL,
    name text,
    userid integer,
    switch integer,
    port text,
    auth_code text,
    reservation_time timestamp without time zone DEFAULT CURRENT_TIMESTAMP,
    last_updated timestamp without time zone,
    expires_at timestamp without time zone,
    publish boolean DEFAULT true NOT NULL,
    provisioned boolean DEFAULT true NOT NULL
);

CREATE TABLE public.switches (
    id integer NOT NULL,
    type public.switch_type NOT NULL,
    address text,
    address_type public.switch_address_type NOT NULL,
    name text,
    description text,
    sip_gateway text
);

CREATE SEQUENCE public.switches_id_seq
    AS integer
    START WITH 1
    INCREMENT BY 1
    NO MINVALUE
    NO MAXVALUE
    CACHE 1;

ALTER TABLE ONLY public.switches ALTER COLUMN id SET DEFAULT nextval('public.switches_id_seq'::regclass);

Rebased on main and swapped your schema in, thanks. I kept one addition in the dev copy: a primary key on extn, since create_extn detects "already taken" by catching the unique violation. If prod doesn't have one either, duplicates insert silently, probably worth checking.

The real schema also caught a bug: publish is boolean but create_extn binds the string 't'/'f', which asyncpg won't accept, so creating an extension 500s. Fixed in #3, together with two mis-indented lines that snuck into the #1 merge (main doesn't parse right now, so #3 should land before anything deploys) and an unpublish fix. This branch sits on top of #3; the extra commits drop off once it merges.

Re-ran the checklist against the new schema and it all works.