Hacker News Reader

Comment by koito17 | Hacker News Reader

Seems like an attempt to bring "lockfiles" or related concepts to Quicklisp. This is a desperately needed quality-of-life improvement for "modern" Common Lisp development.

However, I wish people in the Common Lisp community weren't weary of using ASDF directly. It already supports versioning and many features that do not exist in Quicklisp -- despite the fact Quicklisp wraps ASDF -- such as a package that depends on a version of a library newer (or older) than what exists in the current dist.

It's kind of impressive that Common Lisp is the only language I use where I can't expect code written today to work next month due to Quicklisp lacking versioning outside of the date attached to dists.

View on HN

faré the guy who briefly took over asdf development before quitting common lisp entirely to do scheme, or whatever he's doing right now, spent significant amount of time turning asdf from a 10kb file into 3000kb file in order to solve problems that only he himself cared about, or possibly only google cared about because faré worked for google at the time (building portability library that ignored and force-superseded existing solutions, and trying to eliminate corners cases of SAT solving). there was a period where he was harrassing (this is a subjective interpretation of the events, but this is my side) implementors to incorporate his recent asdf changes, that were breaking left and right to the point that multiple common lisp maintainers of multiple common lisps were considering pulling asdf from their repos entirely. things have stabilized, because faré quit and there's no longer volatility. we're at a point where there's just no concensus about relying on asdf for anything, and while some people go all in on 30% of available functionality (as opposed to customary 5%), equally many people, the majority, use it as a glorified (load)(load)(load). this is an unfortunate side effect of activist developers in small communities: they can have outsized long term effect on how things are done.

In the Meta-CVS project, I carried a copy of asdf.lisp in which the package was renamed to zxcv, to isolate from the ASDF churn and be able to run my version of ASDF possibly with my fixes even in a Lisp that has ASDF preloaded.

ASDF is the only build system I know which broke on me when I tried to fool it with a link farm (lndir command) into building in a separate directory. It resolved the paths using the truename function which eliminates symbolic links, and then calculated the .fas file names from the canonicalized paths, thereby depositing compiled files into the original directory rather than in the symbolically linked tree.

ASDF also depends on symlinks for external organization. .asf files are often symlinks to the real location, and without these symlinks, things don't work.

i have a patched version of asdf 1.369 which was maintained by gary king, with contributions from luminaries like nikodemus, chrodes, weitz, pvaneynd and kevin rosenburg. i've patched it to support the recent extensions that people seem to like (extra parameters to defsystem, like author). you can drop it into your implementation and quicklisp and it just works. asdf 2, asdf 3 happened since, they solve some kind of problems, i guess whatever google needed to do with its fleet, or maybe it's "easier to maintain", but it's not the kind of problems that surface in like 60% of quicklisp packages.

Could you distribute it? I'm not a huge fan of ASDF complexity and have had many headaches trying to make some custom components work. I would welcome any simple alternative, even though I will probably end up writing my own.

> It's kind of impressive that Common Lisp is the only language I use where I can't expect code written today to work next month due to Quicklisp lacking versioning outside of the date attached to dists.

I think the lack of version pinning is a positive force in stability: reducing the pain of breaking changes increases the likelihood that library authors will break things.

This seems to be the widely reported experience from people who develop in Common Lisp daily or very frequently, and even in my haphazard development I find things pretty stable. And it's not like version pinning is all that helpful in a living project anyway, because you often want to update things, and I notice my JavaScript-using friends constantly suffering from breakages every time they try.

With Lisp, I setup my libraries in jenkins to pull the latest SBCL and quicklisp distro once a month and run tests, it's pretty rare for things to break. Once or twice something with an SBCL release itself, once with a bug involving the interplay of two of Fukamachi's libraries. The SBCL one was noticed by others and fixed pretty quickly, for the library one I was able to use git-bisect to easily find the breaking commit and apply a local fix while I waited for the issue I filed to get addressed. This was during Fukamachi's life-threatening illness period, even, but an alternative maintainer did address the bug on their side. The only annoyance these days is Xach is suffering from who knows what preventing the quicklisp distro from updating since October, and there hasn't been communication with people offering to help, so it's forcing everyone to pull projects in locally (and qlot does help with that) for updates or use an alternative.

The only time I've had CL code break due to lack of versioning was when I was depending on stuff written by the author of Qlot, so I either avoid or locally pin stuff written by him.

Most of my dependencies are really quite stable.

[edit]

I thought of a place where things broke that aren't related to Fukamachi; sometimes foreign libraries have changed to the point where they won't grovel any more, but I don't think that's something that Qlot aims to solve.

Honestly, I just don't think it bothers people much.

When I need a specific version of a library, I clone it to my local projects directory and checkout the version I need. Problem solved. Quicklisp is just downloading a snapshot of the upstream repo anyway.

For me, if I could change anything about QuickLisp, I'd add something like Python's `virutalenv`. I suspect I can hack something up by messing ASDF's source registries, but so far I haven't wanted it bad enough to implement it.

My thoughts as soon as I read the README: what’s wrong with asdf?

ASDF is by far the hardest part of the CL ecosystem to work with. I find it unusably complicated. It's the autotools of CL.

Thank god for quicklisp.

That's interesting because I find it to be one of the most ergonomic build tools I've ever worked with. It sits somewhere between make and autotools in terms of what it does and I find it far easier to use than either.

You've probably learned it more thoroughly than I have, or maybe we have different tastes. I find it just impossible to get into, it all rubs me the wrong way. It is so overcomplicated, and unlispy in the way that FORMAT and LOOP are unlispy, only much worse. I actually like FORMAT and LOOP but they're foreign implants amid the lispiness and that's how ASDF feels as well.

it's fukamachi-ware, guy has a bunch of own solutions in the web development space, i suspect driven by his specific needs. they are, fwiw, not as idiosyncratic as some of the artisanal common lisp one-man solutions can be, but they are distinctly and recognizably their own thing

I use his web stack in some of my projects. His stuff is pretty great, it's just that he basically doesn't write any documentation at all, and the small scraps of documentation he does write assume that you're intimately familiar with the details of his packages. So you need to figure a lot out on your own by digging through source code, which is very irritating when you're trying to be productive.

Reply on news.ycombinator.com