Reel adapter/tests seam to be broken

[Asmod4n]

@tarcieri if you got time, might you try to take a look at why all reel specs are failing?

[bethesque]

👍

[tarcieri]

No idea, looks like it might be related to upstream Celluloid changes /cc @digitalextremist

[jimvm]

This is the culprit:

@server = ::Reel::Server::HTTP.supervise(@options[:host], @options[:port], &method(:process))

See here.

Running the specs with the latest Celluloid version gives you a deprecation warning:

INFO -- : Celluloid 0.17.1.1 is running in BACKPORTED mode. [ http://git.io/vJf3J ]

See the link. The supervise method call needs to be updated. The host and port options need to be put in an array in an :args options hash but it's not clear to me where the callback method goes.

On the other hand, using ::Reel::Server::HTTP.new instead of .supervise gets the specs green again. It can handle the host, port, and callback arguments. Maybe we can go with that?

I assume that the HTTPS path needs to be updated as well. But when I got the HTTP path passing again, I did not get any failing specs for the HTTPS path.

Also, there are spec files in the lib directory. Do they belong there or can they be moved?

[digitalextremist]

@jimvm is right that the API is different, requiring args: [] but there is also an issue reported about preserving blocks passed in, as questioned above: https://github.com/celluloid/celluloid-supervision/issues/24 ... Will resolve the celluloid-supervision block clobbering piece and recut that gem.

[Asmod4n]

@seancribbs was there a reason the adapter_lint isn't in the spec folder? I found it a bit confusing too.

[samwgoldman]

@Asmod4n The idea is that those spec files can be used by 3rd party adapters, to verify conformance. At some point in the distant past, we wanted to extract all the adapters into separate gems.

[Asmod4n]

Yes, right. Thats actually a very good idea.

[jimvm]

Would it be difficult to extract them into separate gems?

[rgarner]

It'd be great to extract them into separate gems – it's busy looking like fixing the build for webmachine-ruby tends to get stalled because it's difficult to make five adapters based on differing concurrency models run consistently across half a dozen rubies.

I've had a stab at addressing some of this without extracting gems for now in #233. Just requireing celluloid/autostart before reel has the effect of making the reel adapter specs pass again (although it's still reporting running in backported mode for a reason I've not been able to ascertain).

There are a few other fixes in there, notably one for Socket#close not blocking until the socket's really gone. Does anyone feel inclined to review?

[bethesque]

Sounds great. On mat leave with twins, sorry, don't have time to review code changes!

[rgarner]

Woah, congrats and good heavens, @bethesque. I'm surprised you have time to breathe, let alone reply!

[bethesque]

One handed typing while they sleep!