> It turns out there is very little code that would break if Emacs strings became immutable
Which means that very little Clojure code would break if they just went ahead and used Emacs strings.
> Common Lisp fans claim that lisp-2s have first-class functions, but the way they are kept separate from other first-class values in their own namespace ghetto brings to mind claims of "separate but equal"—at best it is Jim Crow functional programming.
Jim Crow functional programming?? The virtue signalling is strong in this one. He's wrong on the facts, too.
A Lisp-2 doesn't keep the function inside the ghetto, it keeps everyone else out. Functions can be everywhere, but a non-function can't be the value of a function binding. Function's aren't second class in this system, they're royal class.
He means that Emacs Lisp code wouldn't break much if strings were made to be immutable; you don't generally mutate strings. (Buffers, on the other hand, are mutated all the time.) I would suspect that the converse is also true though; Clojure code probably doesn't often rely on strings sharing storage, so making them mutable might not have a huge effect. I've less experience with Clojure than with Elisp, though, so I might well be wrong.
As for the latter, it does occasionally feel like functions aren't truly first-class in a lisp-2 because you have to jump through hoops to pass them along as an argument (funcall, #'foo syntax, etc). In most lisps passing functions as arguments is just as common as "calling" non-function values is in Clojure. Anyway, I suspect that the "Jim Crow" dig was hyperbole, not intentional virtue signalling. :P
> does occasionally feel like functions aren't truly first-class in a lisp-2
The definition of 'first class functions' is not a feeling, but three capabilities:
CL-USER 9 > (let ((sin-fn (function sin)))
(list
; 1) passing functions as arguments to other functions
(mapcar sin-fn '(1 2 3 4))
; 2) returning them as the values from other functions
((lambda () sin-fn))
; 3) and assigning them to variables or storing them in data structures
(let ((v #(1 2 3)))
(setf (aref v 1) sin-fn)
v)))
((0.84147096 0.9092974 0.14112 -0.7568025) ; 1)
#<Function SIN 410007BC14> ; 2)
#(1 #<Function SIN 410007BC14> 3)) ; 3)
As you can see, functions are first-class values in Common Lisp.
> In most lisps passing functions as arguments is just as common as "calling" non-function values is in Clojure
Including Lisp-2s. Passing functions is completely normal and widely used in Common Lisp. The Common Lisp standard has literally zillions of functions where you pass one or more functions in.
You've made my point for me. The very fact that you must write (function sin), or #'sin, or use funcall is a way that it treats functions differently than normal values. Since it requires some amount of extra though and/or typing, this often seen as a bad thing. Compare Scheme, Clojure, or other lisp-1s where if you want to pass the sin function to someone, you just say sin. That whole outer let vanishes in those languages, and the result is much simpler.
And just as we both agree that in Common Lisp passing functions in as arguments is very common, so in Clojure is "calling" a vector on a number to index into it, or "calling" a symbol to use it as an index into a alist or hash table, etc. Not only do you not have to type anything extra, you can treat all values as callable.
> The very fact that you must write (function sin), or #'sin, or use funcall is a way that it treats functions differently than normal values. Since it requires some amount of extra though and/or typing, this often seen as a bad thing.
What a silly thing to say.
The reason you need to put "#'" in Common Lisp to indicate that you are referring to a function, is because in CL functions have their own namespace. So "#'" tells Lisp that you are referring to a symbol that is inside the function namespace. Otherwise, no "#'" and Lisp will look at it on the regular namespace.
Conversely, if you prefix a symbol with ":", such as in ":this :that :foo :bar", the ":" means that this symbol resides on the keyword namespace.
Thus, in Common Lisp you can have a function named "foo", a variable named "foo", a keyword named "foo" as well with no conflict.
It has NOTHING to do with functions being or not being "first class citizens".
3) assigning them to variables or storing them in data structures
None of that requires anything special for functions in Common Lisp.
What you need to do in Common Lisp, is to use FUNCTION to get a function from a lexical name and you need to call such a function via FUNCALL, but that has nothing to do with the requirements for a 'first class datatype', which is defined by the three requirements above (passing, returning, assigning).
> What you need to do in Common Lisp, is to use FUNCTION to get a function from a
> lexical name and you need to call such a function via FUNCALL, but that has nothing
> to do with the requirements for a 'first class datatype', which is defined by the
> three requirements above (passing, returning, assigning).
And I'm saying that in practice, regardless of any technical definitions you'll find in a textbook, this need to treat functions specially is exactly what makes people who learn Common Lisp wonder why people say that functions are first class values. You don't have to look up the value of a variable by writing (variable foo), but you do have to use (function foo).
Clojure goes one step further and lets you treat non-functions as if they were functions, by using them in the function positions of expressions.
> regardless of any technical definitions you'll find in a textbook
It's the definition of first-class values. You may have your own, but then you need to tell us what you mean first. Communication by redefining established terms is getting painful...
> why people say that functions are first class values
Because you can pass functions, return functions and store functions. That's what "first-class datatype" means.
> You don't have to look up the value of a variable by writing (variable foo), but you do have to use (function foo)
Only if you want to get the function from the function namespace. If the function is already referenced in the value namespace, then you don't have to do it.
> Clojure goes one step further and lets you treat non-functions as if they were functions, by using them in the function positions of expressions.
Which existed long before Clojure in Lisp. For example Lisp Machine Lisp had callable arrays 4 decades ago.
Features like these make it a good write-only language.
> Communication by redefining established terms is getting painful...
And here I thought I was telling you that not everybody uses "first-class" (or even "first-class values") in precisely the same way. I agree that the strictest definition only deals with how you use the various types of values you might have in your program, and not with the types of names you can give them. I hope you can also agree that not everyone limits themselves to just the values; they see the differences in how functions get stored in the environment, and recalled from it, and see this as treating the functions differently than the variables.
This does mean that many people are operating with a broader definition of what "first-class" means. It would be perfectly clear to call it "first-class naming", for example, but polysemy is not unique to software engineering; we must allow for it in all fields of endeavour.
My only point all along was merely that when someone hyperbolically calls Common Lisp a "Jim Crow" language, they are not wrong merely because they are talking about names instead of values.
As an aside, I didn't mean to imply that Clojure invented callable arrays, merely that Clojure programs use them frequently; probably just as frequently as they use higher-order functions.
I take no position on whether any language is "write-only" or not; that would be nearly as productive as talking about "virtue signalling". ;)
> And here I thought I was telling you that not everybody uses "first-class" (or even "first-class values") in precisely the same way.
I know of no other definition of 'first-class value', which is an established concept since at least Scheme was invented in the mid 70s, which made functions true first class values. The idea that procedures in some programming languages were not first-class seem to come from C. Stratchey from the mid 60s, where functions in Algol can't be the result of compound expressions, can't be stored in variables and data structures.
> It would be perfectly clear to call it "first-class naming"
If someone would define it and give it some useful semantics, it might be clear.
> I didn't mean to imply that Clojure invented callable arrays, merely that Clojure programs use them frequently; probably just as frequently as they use higher-order functions.
This podcast speaks exactly how an error introduced by callable data influenced the developer leave Clojure for Haskell: https://building.fireside.fm/37
> The very fact that you must write (function sin), or #'sin, or use funcall is a way that it treats functions differently than normal values.
It doesn't treat functions differently from other values. You can say (defvar fn (lambda (x) (+ x 2))) and put a function in a value cell if you want, and then to pass it to a function you just say fn, same as if it were a non-function value. The distinction you're talking about is about symbols being treated differently depending on context, not anything about functions being treated differently from other values.
I'm perfectly well aware that the difference is only in how names are resolved into values, and not in the values themselves. This is a distinction that doesn't matter to most people. What they care most about is what they have to type to convince the compiler to do what they want, and when there are exceptional circumstances that make them think harder or remember more details.
Telling explicitly what one wants to do actually helps the compiler and it helps with the maintenance. The compiler then can help me to find calls which are just accidents, name clashes, etc.
In a Lisp-2 in the expression (sin 2) then SIN is always a function name. It can't be anything else. It's not possible to bind a name in the function namespace to a non-function (or macro).
Using something like FUNCALL is a clear sign that a function value is being called. This makes the source code longer, but it helps me to understand the code even months/years later.
Personally I don't think the distinction between Lisp-1 and Lisp-2 is a very important one, in terms of practical engineering. I just wish I could use any Lisp more often; Lisp-1 vs Lisp-2 doesn't really enter in to it.
Now, I have to go back to working on this Automake/Autoconf mess I'm working on. mumble mumble M4 mumble...
In my TXR Lisp, which is a Lisp-2 dialect with strong ties to Common Lisp, I reveal a design which gives the programmer all of the advantages of Lisp-1, without losing Lisp-2 and its advantages.
If strings in Emacs are mutated so rarely that making them immutable would cause almost no breakage, it logically follows that strings in Emacs are mutated so rarely that using them from Clojure as if they were immutable would cause almost no problem either.
Except that the resulting breakage would be _surprising_. It would be especially surprising to Clojure developers, who are are even more likely to want to rely on immutable data structures.
The timestamp shows that this was written some years ago (2012). I know Guile Emacs has made some progress in the meantime, but is still not done. I wonder if there have been any other changes that would allow for this to happen.
Where are you getting this information from? Phil works for CircleCI, who have been very open about how much Clojure they use, and his GitHub profile indicates anything but: https://github.com/technomancy .
"At the heart of Clojure and ClojureScript’s implementation is #equiv that is in turn based off of Henry Baker’s egal operator introduced in this paper. Briefly, equality in Clojure is defined by equality of value, which is facilitated by pervasive immutability. Equality in the presence of mutability has no meaning."
> It turns out there is very little code that would break if Emacs strings became immutable
Which means that very little Clojure code would break if they just went ahead and used Emacs strings.
> Common Lisp fans claim that lisp-2s have first-class functions, but the way they are kept separate from other first-class values in their own namespace ghetto brings to mind claims of "separate but equal"—at best it is Jim Crow functional programming.
Jim Crow functional programming?? The virtue signalling is strong in this one. He's wrong on the facts, too.
A Lisp-2 doesn't keep the function inside the ghetto, it keeps everyone else out. Functions can be everywhere, but a non-function can't be the value of a function binding. Function's aren't second class in this system, they're royal class.