The runtime is somewhat immature. It locks up oddly sometimes under heavy load. Dealing with latency and queuing issues around gc pauses is much less understo

The runtime is somewhat immature. It locks up oddly sometimes under heavy load. Dealing with latency and queuing issues around gc pauses is much less understood/documented than in the JVM world. The set of best practices in general for doing intense things with the ghc runtime is just still young and sparse. STM can exhibit something that looks a hell of a lot like livelock. Error handling is brutal. Catching all classes of exceptions (at the place you want to catch them!) for recovery is surprisingly tricky. This isn't necessary in theory with things like MaybeT, but in practice, lots of odd libraries use things like partial functions and the error function. Not having tracebacks in production code is painful The library community is thriving but it has a lot of volatility. Things break each other quite frequently. Semantic versioning either isn't enough to save it or hasn't been adhered to strictly enough. Thunk leaks and other consequences of unexpected laziness aren't as common as people worry about, but they're kind of a pain to track down when they occur Strict vs. Lazy bytestrings, String, Text, utf8-string, etc. You may find yourself doing a lot of string/bytestring type conversion There's still wars raging about the right way to do efficient, safe I/O streams. Conduit vs. Enumerator vs. Pipes etc. They're all turning into pretty compelling projects, but the fact that there are N instead of 1 is sometimes a drag when you're dealing with libraries and dependencies. There are not a lot of good open source "application server" type frameworks that really handle thread pooling, resource exhaustion, locking, logging, etc, in robust nice ways. We have one internally, and I'm sure a bunch of other haskell-using shops do too, but the ones on hackage are not nearly sophisticated enough (IMO) and I suspect not very battle tested against the kinds of ugly queuing problems you run into in highly loaded environments. If I think of more, I'll add em... these are off the top of my head.