PublicShow -- HTTP Streams

This module realises encoding and decoding filters, implemented as Prolog streams that read/write to an underlying stream. This allows for sequences of streams acting as an in-process pipeline.

The predicate http_chunked_open/3 realises encoding and decoding of the HTTP Chunked encoding. This encoding is an obligatory part of the HTTP 1.1 specification. Messages are split into chunks, each preceeded by the length of the chunk. Chunked encoding allows sending messages over a serial link (typically a TCP/IP stream) for which the reader knows when the message is ended. Unlike standard HTTP though, the sender does not need to know the message length in advance. The protocol allows for sending short chunks. This is supported totally transparent using a flush on the output stream.

The predicate stream_range_open/3 handles the Content-length on an input stream for handlers that are designed to process an entire file. The filtering stream claims end-of-file after reading a specified number of bytes, dispite the fact that the underlying stream may be longer.

See also
- The HTTP 1.1 protocol
Source http_chunked_open(+RawStream, -DataStream, +Options) is det
Create a stream to realise HTTP chunked encoding or decoding. The technique is similar to library(zlib), using a Prolog stream as a filter on another streQam. Options:
If true (default false), the parent stream is closed if DataStream is closed.
Define the maximum size of a chunk. Default is the default buffer size of fully buffered streams (4096). Larger values may improve throughput. It is also allowed to use set_stream(DataStream, buffer(line)) on the data stream to get line-buffered output. See set_stream/2 for details. Switching buffering to false is supported.

Here is example code to write a chunked data to a stream

    http_chunked_open(Out, S, []),
    format(S, 'Hello world~n', []),

If a stream is known to contain chunked data, we can extract this data using

    http_chunked_open(In, S, []),
    read_stream_to_codes(S, Codes),

The chunked protocol allows for two types of out of band data. Each chunk may be associated with additional metadata. That is achieved using http_chunked_flush/2. The last chunk may be followed by HTTP header lines. That can be achieved by calling http_chunked_add_trailer/3 before closing the chunked stream.

After http_chunked_open/3, the encoding of DataStream is the same as the encoding of RawStream, while the encoding of RawStream is octet, the only value allowed for HTTP chunked streams. Closing the DataStream restores the old encoding on RawStream.

- io_error(read, Stream) where the message context provides an indication of the problem. This error is raised if the input is not valid HTTP chunked data.
Source http_is_chunked(+DataStream) is semidet
True if DataStream is created using http_chunked_open/3.
Source http_chunked_flush(+DataStream, +Extensions) is det
Emits the next chunk flush_output/1 on DataStream, but in addition adds extension parameters to the chunk. Extensions is a list of Key=Value terms. For example, to close a chunked stream with an error chunk we something like below. First, we flush the last pending data, next we fill a new chunk and flush it with extensions.
    format("Sorry, something went wrong!\n"),
    http_chunked_flush(current_output, [error=true])
- It turns out that most clients ignore chunked extensions.
Source http_chunked_add_trailer(+DataStream, +Key:atom, +Value:atom) is det
Add a trailer key/value to DataStream. If the stream is closed, each call adds a line Key: Value\r\n to the output. The strings for Key and Value need to be compliant with the HTTP header syntax.
- It turns out that most clients ignore trailer lines. The JavaScript fetch() method should make these available as response.trailer''.
Source stream_range_open(+RawStream, -DataStream, +Options) is det
DataStream is a stream whose size is defined by the option size(ContentLength). Closing DataStream does not close RawStream. Options processed:
Number of bytes represented by the main stream.
Calls call(Closure, RawStream, BytesLeft) when DataStream is closed. BytesLeft is the number of bytes of the range stream that have not been read, i.e., 0 (zero) if all data has been read from the stream when the range is closed. This was introduced for supporting Keep-alive in http_open/3 to reschedule the original stream for a new request if the data of the previous request was processed.
Source multipart_open(+Stream, -DataSttream, +Options) is det
DataStream is a stream that signals end_of_file if the multipart boundary is encountered. The stream can be reset to read the next part using multipart_open_next/1. Options:
Close Stream if DataStream is closed.
Define the boundary string. Text is an atom, string, code or character list.

All parts of a multipart input can be read using the following skeleton:

process_multipart(Stream) :-
      multipart_open(Stream, DataStream, [boundary(...)]),

process_parts(DataStream) :-
      (   multipart_open_next(DataStream)
      ->  process_parts(DataStream)
      ;   close(DataStream)
- The multipart parser contains code licensed under the MIT license, based on node-formidable by Felix Geisendoerfer and Igor Afonov.
Source multipart_open_next(+DataStream) is semidet
Prepare DataStream to read the next part from the multipart input data. Succeeds if a next part exists and fails if the last part was processed. Note that it is mandatory to read each part up to the end_of_file.
Source cgi_open(+OutStream, -CGIStream, :Hook, +Options) is det
Process CGI output. OutStream is normally the socket returning data to the HTTP client. CGIStream is the stream the (Prolog) code writes to. The CGIStream provides the following functions:
  • At the end of the header, it calls Hook using call(Hook, header, Stream), where Stream is a stream holding the buffered header.
  • If the stream is closed, it calls Hook using call(Hook, data, Stream), where Stream holds the buffered data.

The stream calls Hook, adding the event and CGIStream to the closure. Defined events are:

Called if the header is complete. Typically it uses cgi_property/2 to extract the collected header and combines these with the request and policies to decide on encoding, transfer-encoding, connection parameters and the complete header (as a Prolog term). Typically it uses cgi_set/2 to associate these with the stream.
Called if the HTTP header must be sent. This is immediately after setting the transfer encoding to chunked or when the CGI stream is closed. Typically it requests the current header, optionally the content-length and sends the header to the original (client) stream.
Called from close/1 on the CGI stream after everything is complete.

The predicates cgi_property/2 and cgi_set/2 can be used to control the stream and store status info. Terms are stored as Prolog records and can thus be transferred between threads.

Source cgi_property(+CGIStream, ?Property) is det
Inquire the status of the CGI stream. Defined properties are:
The original request
Term is the header term as registered using cgi_set/2
Stream is the original output stream used to create this stream.
ThreadID is the identifier of the `owning thread'
One of chunked or none.
One of Keep-Alive or close
Total byte-size of the content. Available in the close handler if the transfer_encoding is none.
Codes represents the header collected. Available in the header handler.
One of header, data or discarded
Request sequence number. This number is guaranteed to be unique.
Source cgi_set(+CGIStream, ?Property) is det
Change one of the properies. Supported properties are:
Associate a request to the stream.
Register a reply header. This header is normally retrieved from the send_header hook to send the reply header to the client.
One of Keep-Alive or close.
One of chunked or none. Initially set to none. When switching to chunked from the header hook, it calls the send_header hook and if there is data queed this is send as first chunk. Each subsequent write to the CGI stream emits a chunk. The implementation does not use the chunked stream filter defined by http_chunked_open/3. It shares most of the implementation though and CGI streams do support http_is_chunked/1, http_chunked_flush/2 and http_chunked_add_trailer/3.
Source cgi_discard(+CGIStream) is det
Discard content produced sofar. It sets the state property to discarded, causing close to omit the writing the data. This must be used for an alternate output (e.g. an error page) if the page generator fails.
Source is_cgi_stream(+Stream) is semidet
True if Stream is a CGI stream created using cgi_open/4.
Source http:encoding_filter(+Encoding, +In0, -In) is semidet[multifile]
Install a filter to deal with chunked encoded messages. Used by library(http_open).
Source http:current_transfer_encoding(?Encoding) is semidet[multifile]
True if Encoding is supported. Used by library(http_open).
Source cgi_statistics(?Term)
Return statistics on the CGI stream subsystem. Currently defined statistics are:
Total number of requests processed
Total number of bytes sent.