- the stdlib is good for data structures and not much else, which in turn gave rise to at least three different stdlib extensions
- it does not have typeclasses, so implementing a generic "to_string" for your data type is not a thing
- it has some really unfortunate ideas like "universal compare"
- in general, there is little focus on creature comforts (this is 2017 and despite often relying on exceptions for error handling, there is no "finally" clause in the language)
- in the same spirit, strings are simply bytes, it's up to the developer to ensure they have the right encoding
- the language separates the code itself from the type declaration, a feature which, combined with its type inference, can make reading OCaml code similar to reading Python (in the "what type is this 'e' variable" sense)
- its most widely used build system is pretty terrible
That said, it has some solid features as well, despite the small community size. The packaging system is top-notch, it has good text editor integration via Merlin, and while not always numerous, the libraries are often high-quality, and it is an exceedingly pleasant language to develop with. It strikes a really good balance between solidity and ease-of-use.
let to_string stringer : string = stringer#to_string
val to_string : < to_string : string; .. > -> string = <fun>
I think there's also plenty of people who feel just as strongly about Reason syntax in the other direction.As for classes, well... They're not really used outside of specific situations (eg, UI widgets), so it's not really a terribly convenient solution (especially since, for a really useful to_string implementation, you would want the member fields of your class to also implement to_string, so that would mean objects everywhere).
- jbuilder (https://github.com/janestreet/jbuilder) has markedly improved the build system story. Importantly, the community understands that this is an issue and is actively working towards making the situation better.
- Type declaration at the code level is optional. You could always annotate your code with types, and the type checker will ensure that the annotated types are compatible with the inferred ones. If you would like to know "what type is this 'e' variable", there is merlin (https://github.com/ocaml/merlin), which integrates with popular editors.
- Reg. typeclasses, there has been enormous amount of work put into modular implicits (http://ocamllabs.io/doc/implicits.html). It is likely to be the next major features in OCaml along with multicore.
Well, that's good to know.
> Type declaration at the code level is optional. You could always annotate your code with types, and the type checker will ensure that the annotated types are compatible with the inferred ones. If you would like to know "what type is this 'e' variable", there is merlin (https://github.com/ocaml/merlin), which integrates with popular editors.
I am aware of that. I shouldn't have to use Merlin to read third-party code from github, and god forbid that you have to do code review on a "let's all rely on type inference" codebase in a web browser. I think Rust strikes the right balance here, by enforcing the use of type signatures at function boundaries (which also lets it offer world-class error messages).
> Reg. typeclasses, there has been enormous amount of work put into modular implicits (http://ocamllabs.io/doc/implicits.html). It is likely to be the next major features in OCaml along with multicore.
I know about implicits. I also know that multicore has been "just around the corner" for a few years now, this does not bode well for implicits (which have also been "in progress" for a long time). This also touches another issue with the OCaml maintainers: little in the way of information seems to filter down to the unwashed masses, and let us not speak of the word "roadmap"...