SBCL 20: Steel Bank Common Lisp's 20th Anniversary Workshop | Hacker News Reader

SBCL 20: Steel Bank Common Lisp's 20th Anniversary Workshop

sbcl.org

157 pointskristianp33 comments

SBCL is an implementation of Common Lisp with a long and rich history (if you include CMUCL as a part of that history). It’s an implementation that sees regular contributions from random travelers, and the team of regular developers (if you can even call them a “team”) is small and loosely knit but surprisingly steady. There’s a trusted set of individuals with commit privileges and many of them hack and improve SBCL just about every day. Some of the regular devs are new with their first commits clocking in less than 6 months ago, and others are long retired, having moved on to other things beyond SBCL.

SBCL is a piece of software that’s not inundated with opinions, politics, drama, and other organizational issues. That is in part due to the fact that the folks who develop it must stay grounded. ANSI Common Lisp is, well, a standardized language and the SBCL aims to follow it to a tee.

SBCL is just a small-town coffee shop compared to the Starbucks and Dunkin Donuts of other language implementations. As a professional and commercial user of their software, the size of the user base and developer team used to worry me. But 20 years of stability—compared to even the most contemporary language ecosystems—has given me solace in my choice to use it.

Interesting. I suppose you use it for servers and you don't deploy SBCL to end-user machines?

It is deployed to end-user machines. [0] It’s also open source. [1,2]

[0] https://www.rigetti.com/forest

[1] https://github.com/rigetti/qvm

[2] https://github.com/rigetti/quilc

Really cool!

Very neat! Any blog post on why you chose Common Lisp? I assume you wanted to build your own DSL with macros, but several other tools can do this (Smalltalk, Clojure, Racket, Chez Scheme, Julia... etc). I like CL and think it is as good a choice as any, but what made you decide to use it over some other options?

Not yet. Hopefully one coming soon!

I look forward to it. Do you have a favorite article on Quantum Computing for Dummies? How they work, what problems they're good at...etc. I'm sure many many exist, but I figure you have one you point people towards.

> I suppose you use it for servers and you don't deploy SBCL to end-user machines?

AFAIK, most Common Lisp applications are deployed in a single binary, bundled with the lisp implementation. SBCL isn’t an exception, so if someone would deploy an SBCL app to end-users, one would build a self-contained application with the `sb-ext:save-lisp-and-die` function [0].

[0]: https://lispcookbook.github.io/cl-cookbook/scripting.html

This feature is a godsend when implementing your own language [0] on top since everything is wired up automatically when the application starts.

Compared to emitting C or whatever where you need to include code to stitch the runtime state together.

Makes compiling standalone executables so easy [1] it feels like cheating.

[0] https://github.com/codr7/lila

[1] https://github.com/codr7/lila/blob/master/main.lisp

I never managed to get SLAD to work with Qt bindings and stuff.

CL was my preference many years back, but practical usage at the fringes required me to be a SBCL expert. Sadly, I am/was not :-( so now I am a Python Pleb.

It is easy to make one file executables that handle command line arguments, offer a web or local UI, etc. And apps start instantly.

Still I think it could be much better with the graphical tooling of the surviving commercial Lisps.

No doubt about that! LispWorks and Franz Inc offer great products in this arena.

I ended up buying LispWorks for that reason, and after putting up the work and reading the documentation I found out I am more productive with their IDE than with Slime.

That being said, I am so dependent on SBCL's treatment of types that I always have a Slime session open in order to parallely develop non graphical code on both. It is truly an amazing implementation.

Here’s to another 20 years and beyond!

I think a lot of people would be interested in reading more about your workflow.

Nothing to write home about. Most of the time I'm developing inside of LispWorks, but because I'm so used to Emacs and I'll be using org-mode and magit anyway, having a Slime session as well to try out stuff and run tests comes at no cost. Simply compiling something with SBCL catches some stuff, and it's always good to make sure your code works in different implementations during actual development and not as an afterthought. Besides, as clunky/spartan as they are, Slime and Sly are a lot of fun so why not? Even good old Hemlock gets some use.

When developing I always detach the UI code so it lives in its own package, where I don't hesitate to make use of implementation goodies such as CAPI (of course) and the mp package. The rest of the stuff aims to be portable CL and uses the de-facto compatibility libraries.

$500 for hobbyist edition. Ouch!

I think they are missing an opportunity there. The commercial prices are in line with other professional software offerings. But as a hobbyist I never could quite convice myself to buy a license. As nice as their environment is, there is SBCL. Consequently, at work I would use SBCL too and not argue for a Lispworks license. Of course, while lacking the IDE, SBCL is likely the better Lisp compiler.

Well there's more to it than just the compiler, but if you develop for the server and either SBCL or CCL are behaving like champs for your needs, there aren't many reasons to change. The IDE is really nice but it isn't a game changer. LispWorks shines when you have to ship products that customers will run themselves.

As for the pricing of the Hobbyist edition, I think they are hurting their business this way in the long term. Times have changed and thanks to Quicklisp the ecosystem is better and more important than ever, but a lot of libraries go untested for LW because of such paywall. Not to mention the Personal Edition is barely usable and leaves a bad impression.

But then again, these are just my two cents. I am not going to tell one of the very few commercial vendors alive today how to run their business.

I have to agree. I am using SBCL for all of my Common Lisp work and highly appreciate it. I am able to fulfil my graphical tooling needs with LTk, but I think SBCL would have gained by officially adopting one of the available UI libraries and consequently offering more GUI tools out of the box.

But being as it is, it is easy to load a GUI library of your choice and as a Lisp implementation SBCL is just great.

SBCL purposefully dropped many "extra" libraries that were part of CMU (including GUI library) in order to focus on the lisp implementation itself. Given that the "missing" libraries are recouped by often better external code these days, I find it a pretty good option.

Unfortunately probably nobody has time to work on LW-level GUI library. Significant effort involved from people who mostly have no need for one.

You are probably talking about clx. When talking about GUI libraries, I was more thinking about high-level libraries, which for example bind to Gtk, Qt or Tk. There are plenty of those libraries, and that is good. But I would have liked, if the SBCL maintainers had worked together with one of the maintainers of those libraries with the goal of eventually distributing it with SBCL. So there would be a default library to target when developing tools like e.g. a graphical debugger. Today, quicklist solves a lot of dependency problems, but having an "official" GUI library could have stimulated development of more UI tools around SBCL.

CMUCL also has bindings to Motif, tools like the debugger and inspector use this.

SBCL is amazing. If you walk the development history back trough forks, organizations and names the starting point is Spice Lisp[1] in 1980, then CMUCL in 1985 and finally SBCL in 1999.

[1]: See page 58 in Performance and evaluation of Lisp systems by Richard Gabriel http://rpgpoet.com/Files/Timrep.pdf

I've used SBCL regularly since the initial announcement, truly an excellent piece of software that only keeps getting better.

Being able to sit back and watch bug fixes and performance improvements trickle in without having to deal with the usual torrent of new/deprecated features is a nice change from other languages.

Using implementation-specific features is mostly frowned upon in the CL community, but SBCL is so rock-solid and readily available that I feel less and less motivated to deal with cross-implementation issues. Much like GHC in Haskell land.

Keep it up!

It is truly amazing that SBCL is so stable and fast without much (if any) Big Co. support!

Also, Portacle [1] is a SBCL + (Emacs based) IDE distro that's super easy to setup and use.

[1] https://portacle.github.io

Is this still actively mantained?

I still use Windows and sometimes the machines I'm working on don't have dev environment to build SBCL.

Portacle is actively worked on, though Shinmera (author) afaik doesn't use Windows much, though he attempts to support it equally.

SBCL is great. I do think it would be a significant improvement to include a fuller featured terminal REPL out of the box (with history, s-exp editing keystrokes, completion), and that this would make a better first impression on new users experimenting with the language, before they've given in and installed SLIME - and probably beneficial for the crowd that wants to use their favorite editor rather than Emacs+SLIME, insomuch as they only need to solve the problem of getting definitions reloaded from the editor, without having to also reinvent slime-repl, inspector, debugger, etc.

Fantastic implementation - rock solid. It does feel like the OS X version trails the rest of the platforms, is that true ? It hasn't mattered much for the work I've done with it but curious.

I always compile from scratch on OS X to get any feature I need. Quite easy following the README.

Compiling from source (sh make.sh --fancy; sh install.sh, it’s really that simple) gives you something up to date and stable on OS X.

I have found Portacle [1] to be a great starting point!

[1] https://portacle.github.io

Reply on news.ycombinator.com