missing tests for parallel caching feature

[jjb]

Looks like in a83b39c, the if env[:parallel_manager] condition and the cache_on_complete method it uses were not tested.

The feature itself seems like a bit of a smell — why would the concurrent case not work with threadsafe code?

[mislav]

The parallel case is a separate implementation since it's async and needs to rely on callbacks. The sync implementation of caching middleware is simpler because the Response it gets from app.call(env) is guaranteed to already be finalized. This is an unfortunate product of Faraday's bad support for parallelism and hackish way for Faraday::Response objects to be restored from storage as if they've been constructed from a real response. I'm not sure how to properly test it, but once we have good tests for async cases, we could try consolidating the implementation so we don't have separate execution branches for these modes of operation.

[jjb]

Are we sure that users are using this feature?

Perhaps we should at least document it, or point to an article or something describing its use.

Still seems smelly to me, like this coordination should be the responsibility of a layer higher up.

[mislav]

On Tue, Oct 7, 2014 at 5:19 PM, John Bachir notifications@github.com
wrote:

Are we sure that users are using this feature?

Adapters like EventMachine and Typhoeus support parallelism, so I think
some Faraday users are doing that, yeah.