Hacker News Reader

Comment by aidenn0 | Hacker News Reader

Somewhat OT, but I'm curious what you think is awkward about CLOS?

Simple member access like

    object.foo.bar

is apt to read more like this:

    (class2-bar-of (class1-foo-of object))

Why not use simpler names?

    (bar (foo object))

I mean, I understand that "a.foo" (class A) does not represent the same name as "b.foo" (class B) due to scoping rules, and so logically you would have to use (a-foo a) and (b-foo b) if you wanted to replicate the same in CLOS. But on the other hand, FOO can be generic and applied to different types of objects. You might even have a mixin providing the FOO slot, which classes A and B inherit from.

I find the notation fine, it looks like regular function application (some people prefer threading macros, I personally don't care).

Note also that CLOS allows you to add code "between the dots" by specializing accessors on qualifiers (:before, :after, :around), which can be useful to add orthogonal things like logging, persistency, ...

That could work within a single, well coordinated project. Beyond that, the Common Lisp package system doesn't cooperate the way you'd like it to, so you're back at

    (libraryA:bar (libraryB:foo object))

Since the package system is not hierarchical and doesn't allow local aliases, if you're unlucky you get fully qualified Java-convention package names and end up writing:

    (com.myfavoritecorp.dept1a.project7.subprojectX:bar
       (com.biggercorp.dept32.project2.subprojectE:foo
          object)

> Beyond that, [it] doesn't cooperate the way you'd like it to

I know what I would like it to do. There are some additional features that people would like to use (https://github.com/fare/package-renaming), but even though the package system could be improved, I don't think it is so bad right know. It can be frustrating to try to use it in a way it wasn't designed.

I seldom use qualified symbols, so it is generally "(bar (foo x))" anyway. The case when fully qualifiable symbols might be the best solution is when you need symbols with the same name from different packages at the same time (without shadowing), which is quite rare IMO. Also, defining short packages as a facade for other packages is possible. When you develop an application, not a library, you can also choose to mess things up by renaming packages and adding nicknames.

By using unqualified packages, I can change the meaning of the code by editing only the package definitions (and recompiling): you code says `(foo x)` and originally meant `(a:foo x)`, but then you change a package and now it means `(b:foo x)`. For example `b:foo` could be a wrapper around `a:foo`.

Packages prefix help a little when trying to understand where a symbol comes from. For example, "json:decode" is a little easier to read than "decode". However, with the right editor (or just the REPL) you point to the symbol and have the information.

> Beyond that, the Common Lisp package system doesn't cooperate the way you'd like it to

I'm pretty sure that it is, and that generally in my package I would import bar from libraryA & foo from libraryB and not have to qualify the names.

It's only in the case that both packages export a symbol with the same name that one is forced to disambiguate.

I'd say that Lisp's package system is very close to perfect — it really does work well.

The CL package system allows local aliases. How do you think you can be in the default cl-user package yet refer to cl:cons as just cons?

Firstly, the exported symbols in one package can be made visible in another package through package use. Secondly, a symbol from package A can actually be made present in package B via import: http://clhs.lisp.se/Body/f_import.htm

Regarding packages, sometimes I wonder whether package names could have been symbols: that would allow for gensymed packages, as well as hierarchical packages.

I implemented a package system influenced by CL's. I thought of the idea of package names themselves being symbols (themselves interned in some root package that has no name, or perhaps has one of those symbols as its name). I couldn't come up with a good set of requirements that would make things better, so I didn't bother with it. It just seemed like self-refererence for the sake of self-reference, like some new appendix to D. Hofstadter's GEB. :)

I prototyped a reader-macro for CL that does this, specifically because it gives you package-local nicknames:

https://github.com/jasom/spm-reader

I once made a kind of "package Swiss Army Knife" reader macro for Common Lisp which lets you do cool things. It's reasonably well documented in a header comment.

It might be up your alley, since you're working on a package-related reader macro:

http://www.kylheku.com/cgit/lisp-snippets/tree/pkg.lisp

The context was: "FOO can be generic and applied to different types of objects." That only works if everybody uses the same symbol FOO. Library:FOO is a different symbol and won't be combined automatically, so you must package-qualify.

(Regarding local aliases, I was talking about local package name aliases, which could make it somewhat less awkward.)

Of course. Why would LOTTERY:DRAW and GRAPHICS:DRAW be smashed together?

but now you're back to (lottery:draw x) which isn't any shorter than (lottery-draw x), so the point about names not being namespaced to the object they are applied to remains.

If I have inherited both, I cannot use the symbol DRAW to refer to both; at most one of them can be just DRAW.

Simpler names is certainly less idiomatic (sibling comments have already mentioned why), so it's a fair complaint.

Note that the lisp equivalent of foo.bar(x).baz(y) would be something like:

(send (send foo :bar x) :baz y)) which can be made even more terse with a macro, doing something like (@ foo (:bar x) (:baz y)). This was how most pre-CLOS object systems in lisp worked, and CLOS supplanted this.

Implementing SEND in CLOS is trivial.

There are many people who experienced SEND style object systems and CLOS but I haven't found any of them to be detractors of CLOS, which is what prompted my original question; lisp had something much closer to smalltalk before switching to CLOS and if there is a group of lispers who want to go back to that, they are in a small minority.

One thing one could do is thinking of

    object.foo.bar

similar as

    1+2+3

The later would be

     (+ 1 2 3)

in Lisp.

Thus slot access could be written as:

     (? object foo bar)

All we need would be a macro ?, which is easy.

Alternatively we can write a reader macro such that we can write:

    #?object.foo.bar

The problem with short names and possible package prefixes remains:

    (? w1 ws:panes ws:background-color color:name)

vs.

    #?w1.ws:panes.ws:background-color.color:name

vs.

    (color:name (ws:back-ground-color (ws:panes w1)))

See my other comment; keywords allow for short names and there is no concern about GF collisions if methods are namespaced to the class anyways.

Yes, there is a concern about collisions when methods are namespaced to classes using non-packaged identifiers.

Collisions occur under inheritance. Example: you would like to add a member foo to a base class. The class is widely derived in lots of code, not all of which you even have access to, which might be injecting its own foo. C++ partially deals with this with hacks, like derived identifiers shadowing base class ones (which is totally non-OOP and breaks for virtual functions, which cannot be non-OOP by their nature). C++ partialy deals with the virtual function issue by treating the type signature as part of the name, so that foo(int) and foo(widget&) are different virtuals.

Collisions could be a problem even if there is just one class, but its contents are controlled by several parties. Every group has to watch out that they aren't reusing any other group's symbols in that class.

Any sort of automatic mechanism which can inject symbols into a class (e.g. in support of aspect-oriented programming or whatever) will run into a clash, unless it uses gensyms, which is inconvenient if the intent is to inject names that are to be publicly known and referenced.

Reply on news.ycombinator.com