Show HN: A Common Lisp implementation in development, supports ASDF(savannah.nongnu.org)
savannah.nongnu.org
Show HN: A Common Lisp implementation in development, supports ASDF
https://savannah.nongnu.org/p/alisp
65 comments
Hey I'm curious as why you chose nongnu to host your project instead of github/gitlab! I don't know much about it, hence my curiosity ;)
I don't like sites with heavy Javascript, especially if it's non-free. (Though recently I started using Github for a different project.)
Savannah is very basic, perhaps too much, but it's okay for my project.
Savannah is very basic, perhaps too much, but it's okay for my project.
I hosted the TXR git on nongnu first, starting at around late 2009 or early 2010 maybe?
I abandoned that when I discovered there's no control. I seem to recall having to wait like over a week for someone to enable non-fast-forward pushes. Overly strict and understaffed. I opted for self hosting.
I kept the project web page there, though.
I abandoned that when I discovered there's no control. I seem to recall having to wait like over a week for someone to enable non-fast-forward pushes. Overly strict and understaffed. I opted for self hosting.
I kept the project web page there, though.
Tbh, this is the first time I see nongnu.org used for something other than Emacs packages (I know that’s on me), so much so that I even thought this was a solution to substitute Emacs Lisp with Common Lisp. :O
At what point does a CL implementation need to be before it can start hoovering up the available library code from other implementations (license permitting).
How many LOOP macros does the community need, particularly when bootstrapping an implementation, as an example.
Similarly with, arguably 70-80% of the runtime. The CL spec is dominated by the large library which, ideally, should be mostly portable CL, at least I would think.
How many LOOP macros does the community need, particularly when bootstrapping an implementation, as an example.
Similarly with, arguably 70-80% of the runtime. The CL spec is dominated by the large library which, ideally, should be mostly portable CL, at least I would think.
You're not the first one to think so: https://github.com/robert-strandh/SICL
I'm unsure how complete it is, but it seems to cover much of the standard.
I'm unsure how complete it is, but it seems to cover much of the standard.
Here's a recently-written summary of the different libraries in SICL (including each library's purpose and status) http://metamodular.com/SICL-related-libraries/sicl-related-l...
a hypothetical portable layer exists, but it starts diverging once deployed, because of cleanups, refactoring, or implementation specific hacks.
LOOP is a great example, because all loop is just MIT LOOP version 829, originally cleaned up by burke. but nobody can resist deploying their personal architectural touch, so while the basic framework of loop remains identical across impelementations, there's superficial refactoring done by pretty much everyone. if you take SBCL and franz lisp as state of the art in free software and commercial respectivaly, they have equally solid improvements on original loop, that actually produce incompatible behavior in underspecified corners of spec. respective developer communities are very defensive about their incompatible behavior being the correct behavior of course. beach's SICL from sibling comment is the xkcd joke about standards "20 standards? we need a new standard the unifies them all! -- now we have 21 standards"
LOOP in this case is a very simple example, but for example CLOS was originally implemented on top of PCL, Portable CommonLoops, an interlisp system, that was massaged into being compliant CLOS over years. for example sbcl uses a ship of theseus PCL, but franz lisp did from scratch rewrite. the hypothetical portability of that layer is significantly trickier than LOOP since clos is is deeply tied to the type system, and the boundary between some hypothetical base system common lisp and its clos layer becomes complicated during system bootstrapping. but that's not all! of course clos has to be deeply tied to the compiler, the type system, all kinds of things, to provide optimizations. discovering the appropriate slicing boundary is difficult to say the least.
LOOP is a great example, because all loop is just MIT LOOP version 829, originally cleaned up by burke. but nobody can resist deploying their personal architectural touch, so while the basic framework of loop remains identical across impelementations, there's superficial refactoring done by pretty much everyone. if you take SBCL and franz lisp as state of the art in free software and commercial respectivaly, they have equally solid improvements on original loop, that actually produce incompatible behavior in underspecified corners of spec. respective developer communities are very defensive about their incompatible behavior being the correct behavior of course. beach's SICL from sibling comment is the xkcd joke about standards "20 standards? we need a new standard the unifies them all! -- now we have 21 standards"
LOOP in this case is a very simple example, but for example CLOS was originally implemented on top of PCL, Portable CommonLoops, an interlisp system, that was massaged into being compliant CLOS over years. for example sbcl uses a ship of theseus PCL, but franz lisp did from scratch rewrite. the hypothetical portability of that layer is significantly trickier than LOOP since clos is is deeply tied to the type system, and the boundary between some hypothetical base system common lisp and its clos layer becomes complicated during system bootstrapping. but that's not all! of course clos has to be deeply tied to the compiler, the type system, all kinds of things, to provide optimizations. discovering the appropriate slicing boundary is difficult to say the least.
> a debugger with stepping, a feature that most free CL implementations lack.
I think most free CL implementations have a stepper. Which ones do not?
I think most free CL implementations have a stepper. Which ones do not?
I tried stepping in various free implementations, but I couldn't really follow the source forms and execute them one by one. Also, I couldn't find much information online. Maybe your experience is different?
I haven’t used CL recently so I can’t speak from experience. But it looks like:
CMU CL, SBCL, and LispWorks have steppers.
Clozure does not. (Edit: an answer on https://stackoverflow.com/questions/37754935/what-are-effici... suggests it does...)
As I understand it, those are the big 4.
Clisp, ABCL, and Allegro also appear to have steppers.
Always cool to see a new implementation, though!
CMU CL, SBCL, and LispWorks have steppers.
Clozure does not. (Edit: an answer on https://stackoverflow.com/questions/37754935/what-are-effici... suggests it does...)
As I understand it, those are the big 4.
Clisp, ABCL, and Allegro also appear to have steppers.
Always cool to see a new implementation, though!
In most of those implementations (certainly in SBCL) it's either you break or step; you can't start stepping from a breakpoint. SBCL got some support for that this year, see https://news.ycombinator.com/item?id=43791709. It, however, doesn't allow stepping into any functions called after the break.
Also, the compilers are allowed to make the code unsteppable in some cases, depending on optimization declaration: generally, debug needs to be >=2 and > speed/compilation-speed/space. In some circumstances, you land in decompiled/macroexpanded code, which is also quite unhelpful.
Anyway, it's not that source-level stepping isn't there at all, it's just quirky and somewhat inconvenient. A fresh implementation that does comparatively little optimization and is byte-code based can probably support debuggers better. I hope such support won't go away later when the native code compiler is implemented.
Also, the compilers are allowed to make the code unsteppable in some cases, depending on optimization declaration: generally, debug needs to be >=2 and > speed/compilation-speed/space. In some circumstances, you land in decompiled/macroexpanded code, which is also quite unhelpful.
Anyway, it's not that source-level stepping isn't there at all, it's just quirky and somewhat inconvenient. A fresh implementation that does comparatively little optimization and is byte-code based can probably support debuggers better. I hope such support won't go away later when the native code compiler is implemented.
Thanks!
If I recall correctly, there are macros to control the level of code optimization? And some implementations can turn it off entirely for interactive use?
Or am I off-base?
If I recall correctly, there are macros to control the level of code optimization? And some implementations can turn it off entirely for interactive use?
Or am I off-base?
> If I recall correctly, there are macros to control the level of code optimization?
Yup, you can either `(proclaim (optimize (debug 3) (speed 1)))` somewhere, which will take effect globally, or you can `(declare (optimize ...))` inside a particular function. It sounds great in theory - and it is great, in some respects - but this granularity makes it harder to ensure all interesting code is steppable when you need it.
Yup, you can either `(proclaim (optimize (debug 3) (speed 1)))` somewhere, which will take effect globally, or you can `(declare (optimize ...))` inside a particular function. It sounds great in theory - and it is great, in some respects - but this granularity makes it harder to ensure all interesting code is steppable when you need it.
Congratulations! Always good to see another Lisp in the world.
Have you thought about writing up your experience?
Have you thought about writing up your experience?
Also, my Patreon page (https://www.patreon.com/andreamonaco) has behind-the-scenes posts, some even in the free tier
Thanks! Maybe I could do that, if I see that people are interested
Out of curiosity, what influenced your decision to prioritize stepping and watchpoints early? Was it driven more by debugging your own compiler code, or anticipating end-user needs
I'd say the latter, since I felt the need for debugging tools similar to other languages like C. Watchpoints for example are entirely non-existent in current implementations, as far as I know
Does alisp plan to eventually support full compilation to native code, or will it mainly stay an interpreter with limited compilation?
Yeah, the goal is first bytecode compilation and then full
Do you have a goal in mind for this project?
Ideally I'd reach ANSI compliance, first with a bytecode compiler and then with a full one
Is there some important shortcoming of all the existing Common Lisp implementations that you would like to correct?
Awaiting answers. Seems stepping is one.
Btw, I stick to sbcl as I used vim and so far the script here works for me. Might try this when back to do lisp.
https://susam.net/lisp-in-vim.html
Btw, I stick to sbcl as I used vim and so far the script here works for me. Might try this when back to do lisp.
https://susam.net/lisp-in-vim.html
Yeah, advanced debugging features like watchpoints are very important to me
What is ASDF?
ASDF - another system definition facility - is the de facto standard build system for common lisp.
https://asdf.common-lisp.dev/
In common lisp, you don't need a build system at all; you can `(load "file.lisp")` everything and it should generally just work. But of course, build systems are useful tools, so nonetheless ASDF exists and it's nice enough to the degree that nobody has built a better and more widespread common lisp build system.
Some good trivial examples are in the lisp cookbook:
https://lispcookbook.github.io/cl-cookbook/systems.html
https://asdf.common-lisp.dev/
In common lisp, you don't need a build system at all; you can `(load "file.lisp")` everything and it should generally just work. But of course, build systems are useful tools, so nonetheless ASDF exists and it's nice enough to the degree that nobody has built a better and more widespread common lisp build system.
Some good trivial examples are in the lisp cookbook:
https://lispcookbook.github.io/cl-cookbook/systems.html
No idea why you're being downvoted for asking a simple question about an acronym. From Wikipedia [1]:
> ASDF (Another System Definition Facility) is a package format and a build tool for Common Lisp libraries. It is analogous to tools such as Make and Ant.
Contemporary developers using more mainstream languages are likely more familiar with asdf [2], the "Multiple Runtime Version Manager".
[1] https://en.wikipedia.org/wiki/Another_System_Definition_Faci...
[2] https://asdf-vm.com/
> ASDF (Another System Definition Facility) is a package format and a build tool for Common Lisp libraries. It is analogous to tools such as Make and Ant.
Contemporary developers using more mainstream languages are likely more familiar with asdf [2], the "Multiple Runtime Version Manager".
[1] https://en.wikipedia.org/wiki/Another_System_Definition_Faci...
[2] https://asdf-vm.com/
kazinator(1)
I would be excited to use but since it’s using GPLv3 I can’t actually use it for a lot of projects I’d want to make ;-;
Is it possible to relicense to LGPL or MPL instead?
In general, people who license something as GPLv3 probably consider that a feature, not a bug.
I mentioned here recently that I released a personal project under the GPLv3. The very first issue someone filed in GitHub was to ask me to relicense it as something more business friendly. I don't think I've been so offended by an issue before. If I'm writing something for fun, I could not possibly be less interested in helping else someone monetize my work. They can play by Free Software rules, or they can write their own version for themselves and license it however they want. I don't owe them the freedom to make it un-Free.
The fact that this is hosted on a FSF-managed service indicates the author likely sees it similarly.
I mentioned here recently that I released a personal project under the GPLv3. The very first issue someone filed in GitHub was to ask me to relicense it as something more business friendly. I don't think I've been so offended by an issue before. If I'm writing something for fun, I could not possibly be less interested in helping else someone monetize my work. They can play by Free Software rules, or they can write their own version for themselves and license it however they want. I don't owe them the freedom to make it un-Free.
The fact that this is hosted on a FSF-managed service indicates the author likely sees it similarly.
I generally agree but it's worth noting that languages are a bit different. Obviously there are GPL'd compilers but those often make an explicit carveout for things like the runtime and standard library. Meanwhile in the Lisp world my impression is that most (but certainly not all) implementations are permissively licensed in part due to concerns that shipping an image file is essentially shipping the entire language implementation verbatim.
They can always reward the author, which mostly certainly will make a specific business friendly license for them.
Thanks for pointing that option out! Yes, I am a simple man: you can buy any software I've ever publicly released for the right price. I don't know what those prices are in advance because I've never thought of it, but if you want to give me $10M for some tool I wrote so that I can provide generational wealth to my family, drop me a line.
Of course, no one has expressed interest in doing that yet, so this is purely hypothetical.
Of course, no one has expressed interest in doing that yet, so this is purely hypothetical.
That totally makes sense and I do appreciate why that would be a problem for some users.
And yet, this is a single-user labor of love by one person hosting it on FSF’s servers. I don't know them, and this is pure conjecture, but I suspect they probably couldn't care less if that made it challenging for commercial users. There are plenty of other Lisps for them to choose from.
And yet, this is a single-user labor of love by one person hosting it on FSF’s servers. I don't know them, and this is pure conjecture, but I suspect they probably couldn't care less if that made it challenging for commercial users. There are plenty of other Lisps for them to choose from.
Yeah that makes sense, if it’s it’s just for fun
Hard to believe this comment could be serious, but nonetheless, for the impartial observers, there is a healthy ecosystem of Common Lisp implementations, from "permissive" open source all the way to (expensive) commercial, proprietary ones.
https://common-lisp.net/implementations
I think a full-featured GPLv3 implementation would be very cool, personally.
https://common-lisp.net/implementations
I think a full-featured GPLv3 implementation would be very cool, personally.
Now it ships with ASDF and is capable of loading systems!
Let me know if you like it. Support on Patreon or Liberapay is much appreciated