Ask HN: Best Lisp for software development?

Well, it’s solid. I spent some years as a professional Python programmer, and it was annoying how things which were easy to do in Lisp were difficult or impossible to do in Python (e.g. change a file, reload it and have existing objects updated with the new definitions). Common Lisp offers all the hooks you might want to be able to mutate a system at runtime, but if you don’t want to, then you can just ignore them. I find this very graceful: standardise the possibility, but don’t mandate that one think about stuff one doesn’t need to (yet).

Another aspect of its solidity is how well the standard (ca. 1994) has stood the test of time. There are numerous implementations out there, and they all support the same language (modulo some small bugs, no doubt). Common Lisp code tends to be very portable, and tends to be easy to customise for different implementations and platforms when necessary (the way that FEATURES interacts with conditional reading of source is great). Libraries from twenty years ago still just work. That’s almost unheard of in any other language I’ve used.

Yet another aspect of its solidity is its error-handling capabilities, unmatched in any other language I’ve used. Its condition system offers code the ability to signal exceptional and non-exceptional conditions (specifically, errors, warnings and signals), to catch such conditions, to offer recovery strategies for such conditions and finally to choose between recovery strategies for such conditions — to include prompting for human input if desired. It is an amazingly powerful faculty. But like the ability to handle runtime redefinitions, you don’t need to use it at all if you don’t want to.

It’s high-performance, too, getting within an order of magnitude of the performance of C.

This all means that Common Lisp can be used for quick hacks, simple scripts and all the way up to long-running production systems.

To get started with Common Lisp you’ll want to take a look at Quicklisp, which is the community-standard compilation of free software libraries. It hooks into all the common implementations and offers a plethora of packages.

If you’re willing to use emacs (which I also suggest), then SLIME is very good, but the commercial implementations also have their own excellent IDEs which you might find you prefer.

I’ll be the first to admit that Common Lisp has some flaws, too — I would love to see a modern standard which specifies case-preservation as the default, standardises threads (or something like Go’s channels and goroutines …) and updates the pathname portion of the spec; I’d love to see something which works out a few of the historical bits (like weird argument orders in a few cases). But that stuff is mostly just nitpicking, like complaining that the breaks on a Lamborghini are the wrong colour.

Some eggs just don't install and the error messages are really bad at that point. Trying to install awful to develop a web api I had to give up, couldn't find a way of getting it installed on windows (no problem on OSX or Ubuntu 18.04)

Installed Portacle and (ql:quickload "hunchentoot") and was up and running in a few minutes.

A lot of lisp-languages could do with an equivalent to Portacle, an opinionated dev environment that you can install and just start coding, an area that Racket is excellent in.

Option 1: MS has a Linux subsystem now. Problem solved, but maybe not customer-friendly.

Option 2: Cygwin probably works, I've seen John Cowan mention it in IRC.

Option 3: This somewhat messy process: https://wiki.call-cc.org/compiling-chicken-on-windows-xp-wit...

As someone else here mentioned, Python is where most of the deep learning action is. In the end, I settled in using Python for deep learning and use a thin REST service to use my models from Common Lisp or Haskell.

Another approach I used was to convert saved Keras models to something I could import into Racket.

But back to the topic: Hylang is a very cool project and does let you mix Lisp and Python.

But I don't consider that to be a real native option, it is some kind of weird halfway house that sort of installs linux binaries but they are also kind of windows ones polluting your windows path.

Currently SBCL 'just works' on native windows, which puts it ahead for me if you are happy with both Scheme and CL.