hare-http/net/http/transport.ha

use errors;
use bufio;
use bytes;
use io;
use os;
use strconv;
use strings;
use types;

// Configures the Transport-Encoding behavior.
//
// If set to NONE, no transport decoding or encoding is performed on the message
// body, irrespective of the value of the Transport-Encoding header. The user
// must perform any required encoding or decoding themselves in this mode. If
// set to AUTO, the implementation will examine the Transport-Encoding header
// and encode the message body appropriately.
//
// Most users will want this to be set to auto.
export type transport_mode = enum {
	AUTO = 0,
	NONE,
};

// Configures the Content-Encoding behavior.
//
// If set to NONE, no transport decoding or encoding is performed on the message
// body, irrespective of the value of the Content-Encoding header. The user must
// perform any required encoding or decoding themselves in this mode. If set to
// AUTO, the implementation will examine the Content-Encoding header and encode
// the message body appropriately.
//
// Most users will want this to be set to AUTO.
export type content_mode = enum {
	AUTO = 0,
	NONE,
};

// Describes an HTTP [[client]]'s transport configuration for a given request.
//
// The default value of this type sets all parameters to "auto".
export type transport = struct {
	// Desired Transport-Encoding configuration, see [[transport_mode]] for
	// details.
	request_transport: transport_mode,
	response_transport: transport_mode,
	// Desired Content-Encoding configuration, see [[content_mode]] for
	// details.
	request_content: content_mode,
	response_content: content_mode,
};

fn new_reader(
	conn: io::file,
	resp: *response,
	scan: *bufio::scanner,
) (*io::stream | errors::unsupported | protoerr) = {
	// TODO: Content-Encoding support
	const cl = header_get(&resp.header, "Content-Length");
	const te = header_get(&resp.header, "Transfer-Encoding");

	if (cl != "" || te == "") {
		let length = types::SIZE_MAX;
		if (cl != "") {
			length = match (strconv::stoz(cl)) {
			case let z: size =>
				yield z;
			case =>
				return protoerr;
			};
		};
		return new_identity_reader(conn, scan, length);
	};

	// TODO: Figure out the semantics for closing the stream
	// The caller should probably be required to close it
	// It should close/free any intermediate transport/content decoders
	// And it should not close the actual connection if it's still in the
	// connection pool
	// Unless it isn't in the pool, then it should!
	let stream: io::handle = conn;
	let buffer: []u8 = bufio::scan_buffer(scan);
	const iter = strings::tokenize(te, ",");
	for (const tok => strings::next_token(&iter)) {
		const te = strings::trim(tok);

		// XXX: We could add lzw support if someone added it to
		// hare-compress
		const next = switch (te) {
		case "chunked" =>
			yield new_chunked_reader(stream, buffer);
		case "deflate" =>
			abort(); // TODO
		case "gzip" =>
			abort(); // TODO
		case =>
			return errors::unsupported;
		};
		stream = next;

		buffer = [];
	};

	if (!(stream is *io::stream)) {
		// Empty Transfer-Encoding header
		return protoerr;
	};
	return stream as *io::stream;
};

type identity_reader = struct {
	vtable: io::stream,
	conn: io::file,
	scan: *bufio::scanner,
	src: io::limitstream,
};

const identity_reader_vtable = io::vtable {
	reader = &identity_read,
	closer = &identity_close,
	...
};

// Creates a new reader that reads data until the response's Content-Length is
// reached; i.e. the null Transport-Encoding.
fn new_identity_reader(
	conn: io::file,
	scan: *bufio::scanner,
	content_length: size,
) *io::stream = {
	const scan = alloc(*scan);
	return alloc(identity_reader {
		vtable = &identity_reader_vtable,
		conn = conn,
		scan = scan,
		src = io::limitreader(scan, content_length),
		...
	});
};

fn identity_read(
	s: *io::stream,
	buf: []u8,
) (size | io::EOF | io::error) = {
	let rd = s: *identity_reader;
	assert(rd.vtable == &identity_reader_vtable);
	return io::read(&rd.src, buf)?;
};

fn identity_close(s: *io::stream) (void | io::error) = {
	let rd = s: *identity_reader;
	assert(rd.vtable == &identity_reader_vtable);

	// Flush the remainder of the response in case the caller did not read
	// it out entirely
	io::copy(io::empty, &rd.src)?;

	bufio::finish(rd.scan);
	free(rd.scan);
	io::close(rd.conn)?;
};

type chunk_state = enum {
	HEADER,
	DATA,
	FOOTER,
};

type chunked_reader = struct {
	vtable: io::stream,
	conn: io::handle,
	buffer: [os::BUFSZ]u8,
	state: chunk_state,
	// Amount of read-ahead data in buffer
	pending: size,
	// Length of current chunk
	length: size,
};

fn new_chunked_reader(
	conn: io::handle,
	buffer: []u8,
) *io::stream = {
	let rd = alloc(chunked_reader {
		vtable = &chunked_reader_vtable,
		conn = conn,
		...
	});
	rd.buffer[..len(buffer)] = buffer[..];
	rd.pending = len(buffer);
	return rd;
};

const chunked_reader_vtable = io::vtable {
	reader = &chunked_read,
	...
};

fn chunked_read(
	s: *io::stream,
	buf: []u8,
) (size | io::EOF | io::error) = {
	// XXX: I am not satisfied with this code
	let rd = s: *chunked_reader;
	assert(rd.vtable == &chunked_reader_vtable);

	for (true) switch (rd.state) {
	case chunk_state::HEADER =>
		let crlf = 0z;
		for (true) {
			const n = rd.pending;
			match (bytes::index(rd.buffer[..n], ['\r', '\n'])) {
			case let z: size =>
				crlf = z;
				break;
			case void =>
				yield;
			};
			if (rd.pending >= len(rd.buffer)) {
				// Chunk header exceeds buffer size
				return errors::overflow;
			};

			match (io::read(rd.conn, rd.buffer[rd.pending..])?) {
			case let n: size =>
				rd.pending += n;
			case io::EOF =>
				if (rd.pending > 0) {
					return errors::invalid;
				};
				return io::EOF;
			};
		};

		// XXX: Should we do anything with chunk-ext?
		const header = rd.buffer[..crlf];
		const (ln, _) = bytes::cut(header, ';');
		const ln = match (strings::fromutf8(ln)) {
		case let s: str =>
			yield s;
		case =>
			return errors::invalid;
		};

		match (strconv::stoz(ln, strconv::base::HEX)) {
		case let z: size =>
			rd.length = z;
		case =>
			return errors::invalid;
		};
		if (rd.length == 0) {
			return io::EOF;
		};

		const n = crlf + 2;
		rd.buffer[..rd.pending - n] = rd.buffer[n..rd.pending];
		rd.pending -= n;
		rd.state = chunk_state::DATA;
	case chunk_state::DATA =>
		if (rd.pending == 0) {
			match (io::read(rd.conn, rd.buffer)?) {
			case let n: size =>
				rd.pending += n;
			case io::EOF =>
				return io::EOF;
			};
		};
		let n = len(buf);
		if (n > rd.pending) {
			n = rd.pending;
		};
		if (n > rd.length) {
			n = rd.length;
		};
		buf[..n] = rd.buffer[..n];
		rd.buffer[..rd.pending - n] = rd.buffer[n..rd.pending];
		rd.pending -= n;
		rd.length -= n;
		rd.state = chunk_state::FOOTER;
		return n;
	case chunk_state::FOOTER =>
		for (rd.pending < 2) {
			match (io::read(rd.conn, rd.buffer[rd.pending..])?) {
			case let n: size =>
				rd.pending += n;
			case io::EOF =>
				return io::EOF;
			};
		};
		if (!bytes::equal(rd.buffer[..2], ['\r', '\n'])) {
			return errors::invalid;
		};
		rd.buffer[..rd.pending - 2] = rd.buffer[2..rd.pending];
		rd.pending -= 2;
		rd.state = chunk_state::HEADER;
	};
};