[gumby]

MACLISP had state of the art type checking and compiler optimization, for the 1970s state of the art, and produced better math code than I could write by hand in assembler. These capabilities were used extensively for MACSYMA.

The point of the article wasn't to say that all common lisps support type declaration but to show how easy it is to add support for such decoration (and optional support to boot).

Arbitrary amounts of inference could be added in such declarations before feeding the result to the compiler; the same work could be easily extended for other compilers -- and that other pre-analysis would work there too.

[reikonomusha]

The article described a simpler way for writing DECLARE’s—a feature built-in to Lisp (not a feature the user/author is adding)—and discussed the benefits you could get from such declarations from SBCL (and CMUCL). The same code would not provide much benefit in other implementations of Common Lisp, and the passerby of HN might not gather this nuanced but important detail.

No amount of inference can extend Lisp’s type system to support parametric polymorphism. It’s just not an idea Lisp’s type system can express.

[gumby]

> No amount of inference can extend Lisp’s type system to support parametric polymorphism. It’s just not an idea Lisp’s type system can express.

Well of course it has runtime polymorphism. I assume you mean the metasyntactic polymorphism of, say, C++'s std::vector<foobar>. Yeah, you could sort of get there with some macrology (not even extreme macrology) but you'd probably need to add some call-site sugar which wouldn't be very transparent. Clearly the generic type system could handle it, but there's no inherent representation, which I presume is what you mean.