# `FeedReader.Downloader`

Streams remote response bodies into owned temporary files while computing the
decoded MD5 checksum. Gzip and raw deflate are decoded incrementally; stacked
and unknown content encodings are rejected. See `download/2` for its options
and resource budgets.

# `close`

```elixir
@spec close(%{path: Path.t()}) :: :ok
```

Removes a downloaded temporary file; safe to call more than once.

# `download`

```elixir
@spec download(String.t(), keyword()) ::
  {:ok, %{path: Path.t(), fetch: FeedReader.Fetch.t()}}
  | {:ok, :not_changed | :not_modified}
  | {:error, term()}
```

Downloads, decodes, checksums, and owns a temporary response file.

## Options

- `:http_client` — module implementing `FeedReader.Transport`; defaults to
  `FeedReader.ReqTransport`.
- `:http_options` — Req request options. The default transport streams raw
  response chunks, disables automatic decompression, and limits redirects to
  its configured limit. A supplied `:headers` list replaces its default
  headers.
- `:previous_etag` — binary ETag sent verbatim as `If-None-Match`. A 304
  response returns `{:ok, :not_modified}` without creating a temporary file.
- `:http_retry_attempts` — non-negative retry count after the initial request;
  defaults to the `FeedReader.ReqTransport` setting. Invalid values use the
  default.
- `:retry_delay` — non-negative delay in milliseconds or an arity-one retry
  delay function; defaults to the `FeedReader.ReqTransport` setting. Invalid
  values use the default.
- `:previous_checksum` — checksum of the previous decoded body. An exact match
  returns `{:ok, :not_changed}` and removes the temporary file.
- `:max_compressed_bytes` — maximum cumulative raw response size; defaults to
  `52428800` bytes.
- `:max_decoded_bytes` — maximum cumulative decompressed body size; defaults to
  `52428800` bytes.
- `:max_temp_bytes` — maximum bytes written to the owned temporary file;
  defaults to `52428800` bytes.
- `:max_expansion_ratio` — maximum decoded-to-compressed byte ratio; defaults
  to `100`.

Invalid resource-limit values use their defaults. Limits are checked as body
chunks are processed. Successful downloads return the owned path and a
`FeedReader.Fetch`; callers must pass the download result to `close/1` when
finished. Errors remove any temporary file created for the request.

---

*Consult [api-reference.md](api-reference.md) for complete listing*