gzipped data is coming through funny

I would expect the following script to have the same output as `curl -H "Accept-Encoding: gzip" https://api.github.com:443/repos/JuliaWeb/GnuTLS.jl/git/refs/heads/master`. But it seems to be quite different. They have a similar length 184 bytes vs 189 but otherwise completely different. If I remove the "Accept-Encoding: gzip" header the results match up.

``` julia
using GnuTLS

sess = GnuTLS.Session()
set_priority_string!(sess)
set_credentials!(sess, GnuTLS.CertificateStore())
associate_stream(sess, connect("api.github.com", 443))
handshake!(sess)
write(sess, """GET /repos/JuliaWeb/GnuTLS.jl/git/refs/heads/master HTTP/1.1\r
User-Agent: Julia/$VERSION\r
Host: api.github.com\r
Accept: */*\r
Accept-Encoding: gzip\r
\r
""")

readuntil(sess, "\r\n\r\n")

while !eof(sess)
write(STDOUT, read(sess, Uint8))
end
```

[](https://www.bountysource.com/issues/7632539-gzipped-data-is-coming-through-funny?utm_source=7632539&utm_medium=shield&utm_campaign=ISSUE_BADGE)

Start by running the provided Julia script and comparing its gzip response with the matching curl request, both with and without the Accept-Encoding header. Trace the GnuTLS session read path involved after the HTTP headers are consumed; done means the gzip-enabled response matches curl rather than producing different bytes.

Written by the indexing model from the issue text.