# cepis

Lua client for the [public CEP API at cep.is](https://cep.is). Official
Correios data (DNE), refreshed every two weeks.

```lua
local cepis = require("cepis")

local address = cepis.lookup("01001-000")
print(address.logradouro, address.localidade, address.uf)
-- Praça da Sé  São Paulo  SP
```

## Installation

```sh
luarocks install cepis
```

Targets **Lua 5.4**. It also runs unmodified on 5.1, 5.2, 5.3 and LuaJIT —
nothing here needs anything newer, so the compatibility was free.

> The rockspec points at `git+https://git.cep.is/lua.git` with tag `v0.1.0`.
> That repository is not published yet, so `luarocks install cepis` will not
> resolve until it is. Until then, `luarocks make cepis-0.1.0-1.rockspec` from
> a clone works.

Three rocks are needed, because Lua's standard library ships neither an HTTP
client nor a JSON parser:

| rock | why |
| --- | --- |
| `luasocket` | the HTTP client, plus `gettime`/`sleep` for backoff |
| `luasec` | TLS, so the default `https://cep.is` works |
| `dkjson` | JSON, and pure Lua, so no C toolchain is needed to install it |

CSV is parsed in-tree — a real reader, not `split(",")`, but too small to
justify a fourth dependency.

`luasec` is loaded lazily, only when `base_url` is `https://`. Pointing at a
local instance over plain HTTP therefore needs just `luasocket` and `dkjson`.

## Usage

### A single lookup

`lookup` accepts the CEP masked or unmasked:

```lua
local address, err = cepis.lookup("01310-100")

address.logradouro        -- "Avenida Paulista"
address.localidade        -- "São Paulo"
address.uf                -- "SP"
address.ibge              -- "3550308"
address.cep               -- "01310100"  (always unmasked)
address:formatted_cep()   -- "01310-100"
```

Fields the database does not have for that CEP come back as empty strings,
never `nil`. `address.raw` holds the decoded response untouched, including a
field the API grew after this release.

Expected failures are returned, never thrown:

```lua
local address, err = cepis.lookup(cep_from_form)
if not address then
  print(err.kind, err.message)
end
```

### Coordinates

About 57% of CEPs have a coordinate. When they do not, `geo` is `nil`:

```lua
if address.geo then
  print(address.geo.lat, address.geo.lon)  -- -23.567202  -46.649659
  print(address.geo.raio_m)                -- 1084
  print(address.geo.precisao)              -- "logradouro"
end
```

**Check `precisao` before computing distance.** There are two kinds of point,
and the difference between them is tens of kilometres:

| `precisao` | typical radius | what the point is |
| --- | --- | --- |
| `logradouro` | 110–140 m | the stretch of street — good for maps and distance |
| `localidade` | 7–40 km | the municipality centroid: the CEP *is* the whole city |

`address.geo:is_precise()` is the shorthand for `precisao == "logradouro"`:

```lua
local cost
if address.geo and address.geo:is_precise() then
  cost = shipping_from(address.geo.lat, address.geo.lon)
else
  cost = shipping_from_municipality(address.ibge)
end
```

### Many CEPs at once

Do not loop over `lookup`. Use `lookup_many`, which resolves up to 500 CEPs per
request:

```lua
local ceps = { "01001000", "99999999", "20040020" }
local result, err = cepis.lookup_many(ceps)

for index, address in ipairs(result) do
  if address then
    print(ceps[index], address.localidade)
  else
    print(ceps[index], "not found")
  end
end
```

The result has **the same length and the same order** as the input.

**A CEP that does not exist, or is malformed, is `false` at its position — not
`nil`.** That is not cosmetic. A `nil` inside a Lua sequence truncates it at
the first hole, so `#result` and `ipairs` would stop at the first missing CEP
and silently drop every address after it. `false` keeps the list dense, and
`if address then` still reads as the natural test.

Lists longer than 500 are chunked automatically.

> **Batches carry no coordinate.** The endpoint behind them is CSV, and the CSV
> has no geo column. Every address from `lookup_many` has `geo == nil`,
> including for CEPs that do have one. Need `geo`? Use `lookup`, one at a time.

### API health

Useful for two things: checking the API answers, and knowing **which version of
the database** is live:

```lua
local health, err = cepis.health()

health:ok()             -- true
health.dne              -- "26071"   identifier of the fortnightly database
health.ceps             -- 1596168
health.geo              -- 894489
health.base_gerada_em   -- "2026-07-24T03:48:28Z"
```

An API reporting itself degraded is not a failure — that arrives in
`health.status`.

## Configuration

For anything beyond the defaults, build a client and reuse it. It keeps no
state between calls:

```lua
local client = cepis.new({
  base_url = "https://cep.is",   -- point at another instance
  timeout = 10,                  -- seconds, per socket operation
  max_attempts = 3,              -- counting the first; 1 disables retrying
  backoff_base = 0.5,
  backoff_max = 8,
  total_timeout = 15,            -- cap on the whole call
  user_agent = "myapp/1.0",
})

local address, err = client:lookup("01001000")
local result = client:lookup_many({ "01001000", "20040020" })
local health = client:health()
```

`timeout` bounds a single socket operation — the connect, and each read — not
the whole call. A call may make several attempts with waits between them; the
cap on all of that is `total_timeout`.

Retries cover 429, 5xx and network failures, with exponential backoff and full
jitter; 400 and 404 are never retried.

The module-level functions (`cepis.lookup`, etc.) share a default client,
reachable as `cepis.default()`. It applies process-wide, other libraries
included. Need a different configuration? Build your own.

### Pointing at a local instance

`base_url` is the only setting that usually changes between environments. The
API repo brings a local instance up with `just up`:

```lua
local client = cepis.new({ base_url = "http://localhost:8080" })
```

## Errors

Every expected failure is returned as `nil, err`. `error()` is raised only for
programmer mistakes — a bad option table, a wrong argument type.

An error is a table carrying `kind`, `message`, and whichever of `cep`,
`status` and `retry_after` apply. `tostring(err)` yields the message. The kinds
split by **the decision you have to make**, not by the HTTP status that caused
it:

| `err.kind` | when | does retrying help? |
| --- | --- | --- |
| `cepis.INVALID_CEP` | not 8 digits | no |
| `cepis.NOT_FOUND` | well-formed, not in the database | no |
| `cepis.RATE_LIMITED` | 429; see `err.retry_after` | yes, later |
| `cepis.SERVER_ERROR` | 5xx; see `err.status` | yes, later |
| `cepis.NETWORK_ERROR` | DNS, connection, timeout, TLS | yes, later |
| `cepis.INVALID_RESPONSE` | unintelligible response | probably not |

```lua
local address, err = cepis.lookup(cep_from_form)
if not address then
  if err.kind == cepis.NOT_FOUND then
    -- ask the user for another CEP
  elseif err:is_retriable() then
    -- transient failure: reschedule
    retry_in(err.retry_after or 60)
  end
end
```

The kinds are plain strings, so `err.kind == "not_found"` works too.
`cepis.is_error(value)` tells one of this library's errors apart from anything
else a `nil, err` boundary might hand you.

`INVALID_RESPONSE` almost always means something between you and the API is
answering instead of it — a captive Wi-Fi portal, a corporate proxy, an HTML
error page. Useful as a diagnosis: the problem is not in your code.

## Development

Everything runs through `make`:

```sh
make            # list the targets
make test       # run the suite
make lint       # luacheck when installed, else luac -p; touches nothing
make format     # stylua, when installed (writes)
make build      # byte-compile every module into build/
make check      # lint, then test, then build
```

The suite stands up a real HTTP server on loopback and points the client at it
through `base_url`; nothing is stubbed and nothing touches the network. Run it
on another interpreter with `make test LUA=luajit`.

## Licence

ISC. See [LICENSE](LICENSE).