Take the Arc Challenge(paulgraham.com)
paulgraham.com
Take the Arc Challenge
http://www.paulgraham.com/arcchallenge.html
34 comments
Note: I posted this at http://arclanguage.org/item?id=1043 as well. It's kind of weird having this topic on both YC news and Arc news.
Suppose instead of being a temporary state of affairs, YC/Arc news were to support cross-forum discussions. How simple would that be to support in Arc as it is today? I.e. no fair adding something like (crosspost) to the core libraries.
Suppose instead of being a temporary state of affairs, YC/Arc news were to support cross-forum discussions. How simple would that be to support in Arc as it is today? I.e. no fair adding something like (crosspost) to the core libraries.
It's true that this example will do more to distinguish Arc from other languages than from other Lisp dialects. But if I'd used a Lispy sort of problem as a challenge, everyone would have accused me of cheating by using an example that favored Lisp dialects.
I'll eventually post some comparisons of the same programs in Arc and other Lisp dialects.
I'll eventually post some comparisons of the same programs in Arc and other Lisp dialects.
I'm very interested in the Arc vs. Lisp comparisons. (especially vs. CL)
Here is my contrived example:
How would you, in your favorite language, implement a program that displays GDP information for every country for years from 1970-2007 using a nice graph. Countries should be selectable using a fancy popup menu that displays country name and country shape...
Using Mathematica 6:
How would you, in your favorite language, implement a program that displays GDP information for every country for years from 1970-2007 using a nice graph. Countries should be selectable using a fancy popup menu that displays country name and country shape...
Using Mathematica 6:
MenuView[Row[{Rasterize[
Show[CountryData[#, "Shape"], ImageSize -> {40, 40}],
Background -> None], #}] ->
Dynamic[DateListPlot[CountryData[#, {{"GDP"}, {1970, 2005}}]]] & /@
CountryData[], ImageSize -> Automatic]Languages are about a powerful set of orthogonal primitives (types) and I see pg trying to distill web primitives in a language. I haven't thought enough to say whether there is such a thing as a web primitive truly orthogonal to standard types, though. But, I think it's an interesting question.
I'm sure someone has addressed this very well, but I'm not sure where to look. Anyone happen to know?
I'm sure someone has addressed this very well, but I'm not sure where to look. Anyone happen to know?
> Though simple, as such tests have to be, this is not a contrived example. Web apps have to do this sort of thing all the time.
No. Web app prototypes do that sort of thing all the time. For production web apps, the problem ends up including such confounding factors as meeting accessibility policy, making the HTML surrounding the form elements match what a web designer came up, etc. The more web development work I do, the more wary I am of any system that writes HTML for me. Inevitably the system ends up putting too many constraints on what HTML can be output.
Witness the ASP.NET framework, which writes lots of HTML for you. Unlike Arc it was written with the best practices of the time in mind, but it's aged badly as the web has grown and changed. Doing accessible, cross-browser, standards-based work in ASP.NET today involves a certain amount of fighting the framework which offsets the initial productivity gains.
I understand that Arc is intended for exploratory work and rapid prototyping. But that's exactly why the Arc challenge is meaningless. We're being asked to compare Arc to languages that provide -- dare I say it -- a great deal more flexibility and expressive power when it comes to outputting HTML.
So here's my challenge: Take a sample Arc application and modify it to replace tables with appropriate block elements that have designer-specified IDs and classes. Use <label> elements for each input if you aren't already. Replace a few regular submit buttons with image buttons or vice versa. Reorder some elements, and throw a few more arbitrary containers in here and there. Compare that process against a conventional language and let's see where we get.
In my line of work, that kind of scut work is absolutely part of prototyping. Before I can build the real thing my prototype has to be approved by people who have a hard time looking past even minor visual inconsistencies with what they're used to. For that reason I'm very wary of any design that makes fine-grained control of HTML any harder than it has to be, and I've come to value clean view/everything else separation even in early prototypes.
I suspect that Arc is much more appealing to people who don't live with such constraints.
No. Web app prototypes do that sort of thing all the time. For production web apps, the problem ends up including such confounding factors as meeting accessibility policy, making the HTML surrounding the form elements match what a web designer came up, etc. The more web development work I do, the more wary I am of any system that writes HTML for me. Inevitably the system ends up putting too many constraints on what HTML can be output.
Witness the ASP.NET framework, which writes lots of HTML for you. Unlike Arc it was written with the best practices of the time in mind, but it's aged badly as the web has grown and changed. Doing accessible, cross-browser, standards-based work in ASP.NET today involves a certain amount of fighting the framework which offsets the initial productivity gains.
I understand that Arc is intended for exploratory work and rapid prototyping. But that's exactly why the Arc challenge is meaningless. We're being asked to compare Arc to languages that provide -- dare I say it -- a great deal more flexibility and expressive power when it comes to outputting HTML.
So here's my challenge: Take a sample Arc application and modify it to replace tables with appropriate block elements that have designer-specified IDs and classes. Use <label> elements for each input if you aren't already. Replace a few regular submit buttons with image buttons or vice versa. Reorder some elements, and throw a few more arbitrary containers in here and there. Compare that process against a conventional language and let's see where we get.
In my line of work, that kind of scut work is absolutely part of prototyping. Before I can build the real thing my prototype has to be approved by people who have a hard time looking past even minor visual inconsistencies with what they're used to. For that reason I'm very wary of any design that makes fine-grained control of HTML any harder than it has to be, and I've come to value clean view/everything else separation even in early prototypes.
I suspect that Arc is much more appealing to people who don't live with such constraints.
Do you consider News.YC a mere prototype? It's been around for a year, never gets runtime errors, and serves about 80k pages a day without breaking a sweat. And the source is full of code like this.
I think you're confusing types of software with organizations that sponsor them.
Arc certainly wasn't designed for use in the sort of environment where programs are 10% functionality and 90% the embodiment of the random bureaucratic processes of the organization paying for them. More for the opposite end of the spectrum, in fact. That's where you need power.
That said, if you care to make your challenge a precise one, it's likely I (or someone else) would try implementing it in Arc, if only out of curiosity.
I think you're confusing types of software with organizations that sponsor them.
Arc certainly wasn't designed for use in the sort of environment where programs are 10% functionality and 90% the embodiment of the random bureaucratic processes of the organization paying for them. More for the opposite end of the spectrum, in fact. That's where you need power.
That said, if you care to make your challenge a precise one, it's likely I (or someone else) would try implementing it in Arc, if only out of curiosity.
Bureaucracy isn't the only reason to care about, or have to modify, HTML. Things like accessibility and browser compatiblity matter too, and that ground is constantly shifting. Beyond that, there aren't a whole lot of people outside the world of hackers and scientists who don't care about aesthetics. That means that the overwhelming majority of us will end up working for people who care how stuff looks, even at the prototype stage. That's not bureaucracy, it's human nature.
Every time I've let somebody else's tools write HTML for me, it's caused me grief. Either because the tool author had an agenda that was at odds with my needs (Microsoft), because of bugs in the tool (everyone), or because the problem set shifted since the tool was written as browsers and techniques changed (everyone). Likewise, I'm wary of entertwining HTML output with the rest of the program the way Arc does because it makes it too much work to change things down the line.
It would be interesting to write a templating system that separates HTML output without making Arc programs take much longer to write. I suspect that it could be done, but it's not entirely clear to me what shape it would take. If done right it would give users the option to leverage Arc's existing HTML output functions where desired. I'll have to play with it.
Every time I've let somebody else's tools write HTML for me, it's caused me grief. Either because the tool author had an agenda that was at odds with my needs (Microsoft), because of bugs in the tool (everyone), or because the problem set shifted since the tool was written as browsers and techniques changed (everyone). Likewise, I'm wary of entertwining HTML output with the rest of the program the way Arc does because it makes it too much work to change things down the line.
It would be interesting to write a templating system that separates HTML output without making Arc programs take much longer to write. I suspect that it could be done, but it's not entirely clear to me what shape it would take. If done right it would give users the option to leverage Arc's existing HTML output functions where desired. I'll have to play with it.
How about releasing the source to news.yc so we can see how terse it is for a non-trivial application :)
I will eventually, but first I have to abstract out all the news.yc-specific stuff.
Until it supports Unicode instead of translating my unicode into random entities/replacements, it's still a prototype ;)
A great, usable prototype :)
A great, usable prototype :)
How about this: make news.yc look right in Dillo (http://www.dillo.org).
It looks reasonable in "links" (could use an alt-tag or two)
PHP:
if ($S['foo']) echo "you said {$S['foo']}";
elseif (PtoS('foo')) link('said','click here');
else form('foo');
S[] contains all session variables
P[] contains all posted variables
PtoS copies an entry from P to S. Returns false if P was not set.
link() prints a link
form() shows a form with the given fields. target is the current url.Just counting the tokens in the codetree is not an acceptable measure of brevity, because it implies that adding primitives to a language always makes it more powerful as long it makes your application shorter just once.
But adding primitives has a cost too. You could just have let, or you could have a let for one value, one for more values, one for more values initialized as nil, or as 0, or as an empty string, etc.
In Ruby I write expressions like this a lot: condition ? string : "" I'd like to say: condition &&? string
This would make my applications shorter, but, I claim, less readable and no more powerful.
To measure power, it is not enough to look at succinctness - you must hold something else constant too.
But adding primitives has a cost too. You could just have let, or you could have a let for one value, one for more values, one for more values initialized as nil, or as 0, or as an empty string, etc.
In Ruby I write expressions like this a lot: condition ? string : "" I'd like to say: condition &&? string
This would make my applications shorter, but, I claim, less readable and no more powerful.
To measure power, it is not enough to look at succinctness - you must hold something else constant too.
Sure. There are lots of pathological counterexamples to the length test. I took that as understood.
Just as an aside about your Ruby example:
(condition && string).to_s
(string if condition).to_s
n_to_generate = (ARGV.shift || "0").to_iPaul, in emphasising code brevity I think you are missing a greater good. I believe most of us subconsciously look for a language that is frictionless -- one that allows us to express our ideas without the language getting in the way. To me this is a far more appropriate measure of a language's power than the length of a program. I'd love a language that would allow me to stay in flow* and express my creativity without having to stop and think about the actual language itself. That would be the ultimate Zen power tool.
To be more specific here are three attributes that seem to aid frictionless flow. (perhaps in order of importance) -
--> Expressiveness -- i.e. simple, explicit and unambiguous abstractions.
--> Obviousness and readability. -- i.e. no need to decipher tokens, and the code structure illuminates the intention and flow of the process.
--> Terse. -- i.e. apart from the obvious benefit of requiring less work it also helps me 'keep the whole program in my head'.
There's nothing wrong with brevity of course, but when pursued for it's own sake it seems to be counter productive, in that when you focus on terseness alone there can be a tendency to ride roughshod over the more important attributes: expressiveness and obviousness. You then end up with code that's not as enjoyable to create and whose beauty is lost, except perhaps initially to it's creator.
BTW Paul, this is a great first shot, so a big thank you. I, for one, am routing for Arc to be a widespread success, if only so I can do most of what I need to do in a language that I really enjoy.
* On Flow -> http://www.life2point0.com/2006/06/the_little_book.html
To be more specific here are three attributes that seem to aid frictionless flow. (perhaps in order of importance) -
--> Expressiveness -- i.e. simple, explicit and unambiguous abstractions.
--> Obviousness and readability. -- i.e. no need to decipher tokens, and the code structure illuminates the intention and flow of the process.
--> Terse. -- i.e. apart from the obvious benefit of requiring less work it also helps me 'keep the whole program in my head'.
There's nothing wrong with brevity of course, but when pursued for it's own sake it seems to be counter productive, in that when you focus on terseness alone there can be a tendency to ride roughshod over the more important attributes: expressiveness and obviousness. You then end up with code that's not as enjoyable to create and whose beauty is lost, except perhaps initially to it's creator.
BTW Paul, this is a great first shot, so a big thank you. I, for one, am routing for Arc to be a widespread success, if only so I can do most of what I need to do in a language that I really enjoy.
* On Flow -> http://www.life2point0.com/2006/06/the_little_book.html
I think by aiming toward brevity, what he is actually aiming for is the ability to refactor code into something more brief.
The idea is that by ensuring the progression of long code becoming shorter, and by upholding his second principle (axiomaticity), he is making sure that the code he does factor out will be useful to future programmers. He is removing the need to write common boilerplate code -- and giving the ability to remove boilerplate code to all programmers.
And that supports the idea that PG likes about Lisp -- that the language evolves with the program. The reason it seems to only be good for web programs is that his principle application is a web program. If he were to work on some other kind of program, say a word processor, you would see different kinds of abstractions.
The point is that the language can be adapted to ANY kind of application. That is what he is aiming for. And that's why he doesn't want you to have to count library import statements. He wants to compare the code you have to write repeatedly, not the once-off code. He is claiming that the code you have to write repeatedly indicates how expressive a language is. If you have to repeat yourself, then the language does not adequately represent the abstraction you want to capture.
That's what _On Lisp_ is about. That's what basically all of his Lisp essays are about. Paul Graham is aiming for brevity so that he knows that any programmer can make their program brief by evolving the language. Brevity is a sign of a tight fit between language and the concepts expressed in the program.
The idea is that by ensuring the progression of long code becoming shorter, and by upholding his second principle (axiomaticity), he is making sure that the code he does factor out will be useful to future programmers. He is removing the need to write common boilerplate code -- and giving the ability to remove boilerplate code to all programmers.
And that supports the idea that PG likes about Lisp -- that the language evolves with the program. The reason it seems to only be good for web programs is that his principle application is a web program. If he were to work on some other kind of program, say a word processor, you would see different kinds of abstractions.
The point is that the language can be adapted to ANY kind of application. That is what he is aiming for. And that's why he doesn't want you to have to count library import statements. He wants to compare the code you have to write repeatedly, not the once-off code. He is claiming that the code you have to write repeatedly indicates how expressive a language is. If you have to repeat yourself, then the language does not adequately represent the abstraction you want to capture.
That's what _On Lisp_ is about. That's what basically all of his Lisp essays are about. Paul Graham is aiming for brevity so that he knows that any programmer can make their program brief by evolving the language. Brevity is a sign of a tight fit between language and the concepts expressed in the program.
I have generally felt that short programs feel more elegant and enjoyable -- as long as the "flow" of the program is clear immediately by looking at it. So PG's emphasis on brevity is correct, I think.
The other problems people talk about (make this refactorable, etc. etc.) are essentially problems which arise when we transition from a small code base to a large one. And to have any insight about that, I suspect we will actually need to first build something truly large in Arc. Without doing that, arguments from both sides will just be fanciful and sensationalistic blog crud.
New languages are designed so that we can think differently. Let's first try to honestly think about such an artifact and see if we can express it easily using Arc. Hack first, folks, then talk.
The other problems people talk about (make this refactorable, etc. etc.) are essentially problems which arise when we transition from a small code base to a large one. And to have any insight about that, I suspect we will actually need to first build something truly large in Arc. Without doing that, arguments from both sides will just be fanciful and sensationalistic blog crud.
New languages are designed so that we can think differently. Let's first try to honestly think about such an artifact and see if we can express it easily using Arc. Hack first, folks, then talk.
I think it is quite mistaken to put brevity above everything else. For a 100-year language, presumably with corresponding 100-year codebases, surely the promotion of readable and refactorable source code would be the key. Macro systems already provide a means toward brevity - whatever the language itself does becomes inconsequential given that.
Here's a Scheme version. (Gauche, with a web framework Kahua):
Now, beyond a certain point, optimization becomes specialization. It's like compression. The more you know about the domain, the shorter you make the frequent patterns, and the net amount of code gets smaller. But how do you know you're not overspecializing?
Some of Lispers, including me, have this fear for overspecialization. Reading arc does have an effect to cure that. I see some nice abstractions in arc which I'd like to steal from.
I'm more curious, though, to see wider range of applications in arc and how the core abstractions will live up.
(btw, Paul, if you read this could you reset my password in Arc forum and send it to [email protected]?)
(define-entry (said)
(define (page . s) `((html (body ,@s))))
(page
(form/cont:
(@@: (cont (entry-lambda (:keyword foo)
(page (a/cont: (@@: (cont (entry-lambda () (page "You said: " foo))))
"click here")))))
(input: (@: (name "foo")))
(input: (@: (type "submit"))))))
Now, beyond a certain point, optimization becomes specialization. It's like compression. The more you know about the domain, the shorter you make the frequent patterns, and the net amount of code gets smaller. But how do you know you're not overspecializing?
Some of Lispers, including me, have this fear for overspecialization. Reading arc does have an effect to cure that. I see some nice abstractions in arc which I'd like to steal from.
I'm more curious, though, to see wider range of applications in arc and how the core abstractions will live up.
(btw, Paul, if you read this could you reset my password in Arc forum and send it to [email protected]?)
I'm interested in this part: "Steele and Sussman tried to start over when they first began working on Scheme, but they seem to have been practically the only ones. And they made, at least from the point of view of brevity/power, some serious mistakes early on."
Paul, out of curiosity, what are those serious mistakes?
Paul, out of curiosity, what are those serious mistakes?
I'm also curious.
Take the C++ Challenge (entries will be judged by quality rather than fewness of lines.)
http://en.wikipedia.org/wiki/Id_Tech_4
Take the Java Challenge (entries will be judged by quarterly revenue, rather than fewness of lines)
http://www.ebay.com
http://en.wikipedia.org/wiki/Id_Tech_4
Take the Java Challenge (entries will be judged by quarterly revenue, rather than fewness of lines)
http://www.ebay.com
The idea behind Arc seems to be to create a new language to stand the test of time. The problem I'm having with this challenge, and Arc in general, is that it's very domain specific.
Now, lisp /scheme is really great at making DSLs. This example makes me think that Arc isn't a new all-purpose language, but rather a DSL language that can easily be modified to withstand criticism. "It doesn't have feature x!!!" ...2 weeks later... "Well, it does now."
I'm not saying that DSLs are insignificant, or that Arc's creation was trivial, but when examples like this are used, it feels more like a framework than a new paradigm. I think most Python/Ruby/Haskell/(modern language) hackers could fairly easily implement a library to do what Arc does in about as many lines, but they wouldn't call it a new language.
Now, lisp /scheme is really great at making DSLs. This example makes me think that Arc isn't a new all-purpose language, but rather a DSL language that can easily be modified to withstand criticism. "It doesn't have feature x!!!" ...2 weeks later... "Well, it does now."
I'm not saying that DSLs are insignificant, or that Arc's creation was trivial, but when examples like this are used, it feels more like a framework than a new paradigm. I think most Python/Ruby/Haskell/(modern language) hackers could fairly easily implement a library to do what Arc does in about as many lines, but they wouldn't call it a new language.
is program shortness really that important?
what percentage or time developing an app do you guys spend typing? For me it is quite low but maybe i'm not a superhacker or i just type fast
what percentage or time developing an app do you guys spend typing? For me it is quite low but maybe i'm not a superhacker or i just type fast
It's not so much about typing as thinking. It's not just to save typing that you use high level languages instead of machine languages, right?
Suggestion to Arc fans: do not do XML examples, as C# 3/VB 9 and LINQ to XML will leave you in the dust.
Everyone loves a short girl (particularly Asian), you can throw them around the room and generally have a good time with them.
A short, hideous girl (Arc)? I wouldn't hit it.
I'm sticking to the short and beautiful (Python/Ruby).
A short, hideous girl (Arc)? I wouldn't hit it.
I'm sticking to the short and beautiful (Python/Ruby).
When I look at something like
what I see is more or less ordinary lisp that uses a library to solve an issue with applications that rely on http. I'm sure plenty of us could do the equivalent with the appropriate Python/Perl/Php/Ruby library, but then the obvious criticism would be that you're not really comparing lanugages. You would be comparing libraries.
Plenty of examples over here: http://arclanguage.com/item?id=722
Is it really the case that when you boil it down, Arc (language+libraries) is about making small web programs? If not, why this challenge and if so I suspect it won't live up to the 100-year idea.