consistency with stdlib io api about fsnotify

[nathany]

Thanks for the suggestion. The current API isn't final for sure, we definitely want to change how error handling works: #2.

[sbakke]

Thank for the response. I would also reconsider the forced goroutine/channel usage as well, in case the library is used without the need for concurrency. This way it would look more like other IO packages in the standard library. Keep in mind the nightmare that is Java's standard library as an example where different generations of packages have different API style, especially within the IO library.

[nathany]

There was an issue opened at howeyc/fsnotify#7 regarding the requirement for a goroutine in the calling code.

#5 is part of looking at that, specifically on Linux.

for BSD I've found that having kqueue block #24 results in less CPU utilization when idle, but doing so may make it harder to avoid having a goroutine in the calling code