Support transparent gzip decompression

[Alexander-Barth]

Some server always respond with a compressed stream, e.g. the stackoverflow API:

import HTTP
r = HTTP.get("https://api.stackexchange.com/2.2/questions"; query = Dict("site" => "stackoverflow"));

It would be nice if I can directly access the decompressed response body.
As a workaround, I can use CodecZlib

using CodecZlib
String(transcode(GzipDecompressor,r.body))

Would it be possible to support transparent decompression, or is it already implemented?

Julia 0.6.3
HTTP.jl master 61e6b05
MbedTLS.jl 0.5.9

[samoconnor]

Yes, this could be supported.
This is the place to hook it in:

HTTP.jl/src/Messages.jl

Lines 395 to 401 in b602c4e

	function decode(m::Message, encoding::String)::Vector{UInt8}
	if encoding == "gzip"
	# Use https://github.com/bicycle1885/TranscodingStreams.jl ?
	end
	@warn "Decoding of HTTP Transfer-Encoding is not implemented yet!"
	return m.body
	end

[EricForgy]

This is cool. You guys ok to add CodecZlib as a dependency? Making the change is easy. Any idea what would be a good test?

[Alexander-Barth]

A test case could be the use of http://httpbin.org/gzip:

julia> r = Requests.get("http://httpbin.org/gzip")
Response(200 OK, 13 headers, 221 bytes in body)

julia> print(readstring(r))
{"gzipped":true,"headers":{"Accept":"text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8","Connection":"close","Host":"httpbin.org","User-Agent":"Requests.jl/0.0.0"},"method":"GET","origin":"139.165.106.124"}

julia> r.headers
Dict{String,String} with 13 entries:
  "Connection"                       => "keep-alive"
  "Via"                              => "1.1 vegur"
  "Access-Control-Allow-Credentials" => "true"
  "Date"                             => "Wed, 20 Jun 2018 08:13:50 GMT"
  "Access-Control-Allow-Origin"      => "*"
  "http_minor"                       => "1"
  "Keep-Alive"                       => "1"
  "status_code"                      => "200"
  "Server"                           => "gunicorn/19.8.1"
  "Content-Length"                   => "191"
  "http_major"                       => "1"
  "Content-Type"                     => "application/json"
  "Content-Encoding"                 => "gzip"

[jakewilliami]

I'm not sure if this is still being worked on, but regarding the method we use to decompress, there seem to be two reasonable options (that I can find). Option 1 uses (as suggested by @samoconnor above) TranscodingStreams and CodecZlib. Option 2 uses Libz, which seems to be a faster Gzip. My test data is a 105-element Vector{UInt8}, but perhaps there is better test data out there.

using BenchmarkTools, TranscodingStreams, CodecZlib, Libz

function decode1(data::Vector{UInt8})
    return String(transcode(GzipDecompressor, data))
end

function decode2(data::Vector{UInt8})
    return String(read(ZlibInflateInputStream(data)))
end

Benchmarking this, we get

julia> @btime decode1(test);
  1.770 μs (7 allocations: 832 bytes);

julia> @btime decode2(test);
  1.517 μs (7 allocations: 9.77 KiB)

Or, using the suggested test data by @Alexander-Barth:

julia> @btime decode1(r.body);
  4.948 μs (7 allocations: 1.31 KiB)

julia> @btime decode2(r.body);
  4.417 μs (7 allocations: 9.89 KiB)

Not sure if this information is useful, but thought I'd share what I've found.

[ashwanirathee]

@jakewilliami this was useful, tysm!!

[quinnj]

Going to put this on the 1.0 milestone. I'll try to work on this soon.

[quinnj]

PR is up: #838

[Alexander-Barth]

Thank you very much for this great package and this new feature 🙂

[Anastasgrek]

Hello everyone! I have a problem about this topic. What about WebSocket connection? I getting packages with compress data from server, but I can't decompress it. I get zlib error: <no message> (code: -5) from CodecZlib. it works if I do this:

using HTTP
using TranscodingStreams, CodecZlib
using JSON
function extract_data(io::IOBuffer)
    stream = TranscodingStream(DeflateDecompressor(), io)
    try
        for line in eachline(stream) end
    catch e
        if e.msg != "zlib error: <no message> (code: -5)"
            throw(e.msg)
        end
    end
    l = stream.state.buffer1.transcoded
    d = stream.state.buffer1.data[1:l]
    return JSON.parse(String(d))
end

HTTP.WebSockets.open(
    "wss://stream.binance.com:9443/stream?streams=btcusdt@depth5@100ms",
    headers = [
        "sec-websocket-extensions" => "permessage-deflate; client_no_context_takeover; server_no_context_takeover, client_max_window_bits",
    ]
    ) do ws

    while true
        data = readavailable(ws) |> IOBuffer |> extract_data
        println(data["stream"])
        # break
    end
end

I understand, in this PR gzip decomression protocol, but in my opinion, it's not much different.
Do you have any ideas, how fix it?