fixes #36, switch to HTTP/1.0 and drop Connection: keep-alive from httpclient

[jangko]

No description provided.

[coffeepots]

This LGTM but @cheatfate can you have a look too. One thing that makes me pause is I was expecting chunked transfer to be placed in nim-http-utils as a more generic process, however perhaps it's better (aka less complex) to have it in place in RPC.

[coffeepots]

Also I'm wondering how easy it would be to add some simple tests for the chunking. @jangko would this be easy for you to set up?

[jangko]

would this be easy for you to set up?

honestly, I have no idea where to start, that's why I left it blank.
Does it require me to modify the server too?
any clues perhaps?

[cheatfate]

@jangko chunked transfer encoding is not supported by HTTP/1.0, and because geth properly supports HTTP specification you can use HTTP/1.0 requests and avoid chunked answers.

nim-json-rpc is also not the right place for chunked decoder.

[jangko]

luckily, parity also support HTTP/1.0.
does it mean we also drop Connection: keep-alive field too when connecting with HTTP/1.0?
is that acceptable?

[cheatfate]

Connection: keep-alive is also from HTTP/1.1.

[jangko]

switching to HTTP/1.0 and removing Connection: keep-alive prove to be more problematic.
I think I will go through the hard way by putting it into httputils and implement chunked transfer decoding

[cheatfate]

More then 100k requests per second is not enough for you @jangko, you need more speed?
I do not like your approach on chunked processing inside of nim-http-utils, because it looks not as solid improvement, but like temporary fix, it will not allow to easily extend functionality by adding gzip, deflate encoding.

[jangko]

I didn't say about speed, I've located the problem inside asyncdispatch2, perhaps we can pending that chunked transfer problem later and for now we can switch to HTTP/1.0 and drop keep-alive connection.

[coffeepots]

@jangko We've been discussing the chunking work and have come to the conclusion that we should stick with HTTP 1.0 for now (which you've already done here) and later have fully genericised chunking code further up in the tool chain.

[jangko]

done

[jangko]

closed in favor of 6a0b0ff