I find there's a lot you have to do manually to handle them correctly and it's easy to miss details. I think they're best cordoned off behind isolated, encapsulated parts of a channel-free public interface.
To walk back a tiny bit from my earlier "80%" claim, I reckon channels are probably more like 50% complete. The existence of "context.Context", itself very much half-baked, is a symptom of that.
You have to think long and hard about cancellation and blocking every time you interact with them (even if you decide you don't ultimately need to), and the results, once you've (hopefully) handled all the corner cases can degenerate into a tangled mess that's hard to reason about, harder to test, and impossible to change. I like to call this kind of code "channel soup".
I'd be interested to see proposals for what the 80% version of channels would look like. What we have now is a great start, but it's not great yet.
Whether or not the new way will be better is far from assured.
I'm outing myself as a bit of a "get off my lawn" greybeard here, but as long as such a large swathe of the industry remains trapped in a permanent state of convulsion where everyone competes to build the next monument to their own cleverness, the unprecedented consistency, clarity and productivity that vanilla Go 1 provides will remain under-appreciated.
There's an awful lot of sound and fury over generics and error handling. I sincerely hope that if and when something is done about it, it isn't done in a way that gives anyone a license to be clever.
> "use channels, share memory by communicating"
They walked back a bit from this, thankfully, and I'm seeing less and less nasty "channels for the sake of it" code these days as a result. The party line seems to be closer to "channels are good and bad, mutexes are good and bad, favour whichever is goodest" nowadays, and the larger projects I've worked on have typically been about 50/50 [1]
Whenever I think something is missing, I find it helpful to remind myself that Go is kind of the ultimate "80%" experience. Absolutely everything is 80% complete, from the language, to the tools, to the stdlib, it's performance relative to vanilla C... everywhere. 80%. I also think this is the secret of its success.