lose --network_engine

[chadwhitacre]

Why do we have to support multiple network engines? Because the Python network lib situation is a hoary mess. Maybe PEP 3156 will clean things up. Any help from that angle is years down the line, though.

[chadwhitacre]

IRC

[chadwhitacre]

Python (unlike Node) has network library religion. Right now our policy is to support them all, but this makes implementing Socket.IO pretty complicated. I'd like to make a push to officially release Socket.IO support (#201) , and network_engine is on the chopping block as part of that.

[chadwhitacre]

Sprinting at @pjz's place, talking about this. Big one. The thinking here is that we've got two pucks to skate ahead of if we can: HTTP 2.0 and Tulip.

[chadwhitacre]

I don't really have a good understanding of either of those. Suggestion on the table is to write some code in a branch to spike this out, understand the issues better.

[chadwhitacre]

We're debating whether to actually lose the network_engine endpoint or just drop to only supporting one network engine plugin, to reduce our maintenance burden. We'd still want to write it against an API though, as a best practice.

[chadwhitacre]

Sounds like we're not going to lose this. We may decide to prune some plugins that we don't bring along with #200, though.

[chadwhitacre]

We may very lose this, if #200 makes it too complicated to keep.