event-loop icon indicating copy to clipboard operation
event-loop copied to clipboard

Common Driver Implementations

Open trowski opened this issue 8 years ago • 12 comments

The Driver interface leaves no room for differing behaviors between implementations. Once a particular vendor has written performant Driver implementations for various backends there is little reason for another vendor to do the same. Changes in behavior between Driver implementations render them non-interoperable. Perhaps then this group should also offer Driver implementations, dissociating Driver implementations from any particular vendor.

amphp/loop contains highly optimized Driver implementations based on the native stream_select(), libevent (ext-event), libev (ext-ev), and libuv (ext-uv). Should this library be migrated to be a part of this project?

trowski avatar Feb 22 '17 06:02 trowski

Huge :+1:, makes using the loop way easier. People can just require async-interop/event-loop and things work. I'd still keep the factory mechanism, so people can still write their own implementations, maybe with custom underlying extensions.

kelunik avatar Feb 22 '17 14:02 kelunik

What if we ended up with two versions of the uv driver which worked with different versions of the uv extension? Perhaps the drivers which rely on extensions should be packaged separately and we should just include a stream_select loop?

Edit: I suppose in that case we could instead have two uv drivers in this package, with the correct one picked based on the extension version. Or the driver could simply be designed to switch behaviour based on the extension version. Maybe this isn't an issue.

joshdifabio avatar Feb 22 '17 14:02 joshdifabio

@joshdifabio This package will take care of it then and add a new driver, just like any other package would. Not including the extension based drivers is a no-go IMO, as it effectively doesn't bring any value then.

kelunik avatar Feb 22 '17 15:02 kelunik

This would certainly simplify usage. It does raise the question of what the difference between this package and, say, amphp/loop would actually be. Clearly replacing vendor-specific loop packages would be the intention here but I wonder if React and others would actually be on board with this.

joshdifabio avatar Feb 22 '17 15:02 joshdifabio

@joshdifabio React adapting the interop loop is highly unlikely. Everything we can do is provide an adapter so every interop library / application can use any React library.

It does raise the question of what the difference between this package and, say, amphp/loop would actually be.

None, they're just merged then, it's the exact proposal.

kelunik avatar Feb 22 '17 15:02 kelunik

Okay. I think I'm on board as well.

joshdifabio avatar Feb 22 '17 15:02 joshdifabio

Ping @bwoebi @jsor @sagebind @WyriHaximus @AndrewCarterUK

trowski avatar Feb 22 '17 16:02 trowski

I'm in favor. Only issue I see is from a community perspective it might seem like a conflict of interest. It might send the wrong idea that implementing the standard is discouraged, since we would point to using our implementation instead.

sagebind avatar Feb 22 '17 18:02 sagebind

@sagebind Implementing the already supported backend extensions again is indeed unnecessary. If there's room for improvement, submit a PR to this repository instead. Custom implementations will be allowed, but not the usual way of doing things.

kelunik avatar Feb 25 '17 17:02 kelunik

Also ping @cboden and @clue.

kelunik avatar Feb 25 '17 17:02 kelunik

:+1: ... particularly if including a react adapter is a feasible option

rdlowrey avatar Mar 02 '17 16:03 rdlowrey

@rdlowrey I wouldn't include it in this package, but having a separate package under async-interop for a React adapter would be a possibility.

trowski avatar Mar 02 '17 16:03 trowski