Monday, August 25, 2014

Dylan: the harsh realities of the market

So, this is a little case study.

I did everything for Dylan. And when I say everything, I mean everything.  Here's my resumé:


  • I got excited about Dylan as a user, and I used it. I bought an old Mac that I don't ever remember the designation for, it's so '90's old, and got the floppies for the Dylan IDE from Apple research.
I'm not joking.

  • I integrated Dylan into my work at work, building an XML parser then open-sourcing it to the community under the (then) non-restrictive license. I think mine was the only XML parser that was industrial-strength for Dylan. Can't claim originality: I ported over the Common-LISP one, but it was a lot of (fun) work.
  • I made improvements to the gwydion-dylan compiler, including some library documentation (you can see my name right there, right in the compiler code), including some library functionality, did I work on the compiler itself? The Dylan syntax extensions or type system? I don't recall; if not in those places, I know I've looked at those guts: I had my fingers all over parts of the compiler.
I was in the Dylan compiler code. For you ll-types ('little language') that's no big deal.

But ask a software developer in industry if they've ever been in their compiler code. I have, too: I've found bugs in Java Sun-compiler that I fixed locally and reported up the chain.
  • I taught a course at our community college on Dylan. I had five students from our company that made satellite mission software.
  • I effing had commercial licenses bought when the boss asked me: what do we have to do to get this (my system) done/integrated into the build. I put my job on the line, for Dylan. ... The boss bought the licenses: he'd rather spend the $x than spending six weeks to back-port down to Java or C++.
  • I built a rule-based man-power scheduling system that had previously took three administrative assistants three days each quarter to generate. My system did it, and printed out a PDF in less than one second. I sold it, so that means I started a commercial company and sold my software.
I sold commercial Dylan software. That I wrote. Myself. And sold. Because people bought it. Because it was that good.

Hells yeah.

Question: what more could I have done?

I kept Dylan alive for awhile. In industry. For real.

So why is Dylan dead?

That's not the question.

Or, that question is answered over and over and over again.

Good languages, beautiful languages, right-thing languages languish and die all the time.

Dylan was the right-thing, and they (Apple) killed it in the lab, and for a reason.

Who is Dylan for?

That's not the question either. Because you get vague, general, useless answers.

The question is to ask it like Paul Graham answered it for LISP.

Lisp is a pointless, useless, weird language that nobody uses.

But Paul and his partner didn't care. They didn't give a ...

Something.

... what anybody else thought. They knew that this language, the language they loved, was built and designed and made for them. Just them and only them, because the only other people who were using it were college kids on comp.lang.lisp asking for the answers for problem-set 3 on last night's homework.

That's what Lisp was good for: nothing.
That's who Lisp was good for: nobody.

Same exact scenario for Erlang. Exactly the same. Erlang was only good for Joe Armstrong and a couple of buddies/weirdos like him, you know: kooks, who believed that Erlang was the right-thing for what they were doing, because they were on a mission, see, and nothing nobody could say could stop them nor stand against them, and all who would rise up against them would fall.

All.

What made Lisp and Haskell and Erlang and Scala and Prolog (yes, Prolog, although you'll never hear that success story publicly, but $26M and three lives saved? Because of a Prolog system I wrote? And that's just one day in one month's report for data? I call that a success) work when nobody sane would say that these things would work?

Well, it took a few crazy ones to say, no, not: 'say' that it would work, but would make it work with their beloved programming language come hell or high water or, worse: indifferent silence, or ridicule, or pity from the rest of the world.

That is the lesson of perl and python and all these other languages. They're not good for anything. They suck. And they suck in libraries and syntax and semantics and weirdness-factor and everything.

But two, not one, but at least two people loved that language enough to risk everything, and ...

They lost.

Wait. What?

Did you think I was going to paint the rosy picture and lie to you and say 'they won'?

Because they didn't.

Who uses Lisp commercially? Or Haskell, except some fringers, or Scala or Clojure or Erlang or Smalltalk or Prolog

... or Dylan.

These languages are defined, right there in the dictionary.

Erlang: see 'career wrecker.'

Nobody uses those languages nor admits to even touching them with a 10-foot (3-meter) pole. I had an intern from college. 'Yeah, we studied this weird language called ML in Comp.sci. Nobody uses it.'

She was blown away when I started singing ML's praises and what it can do.

A meta-language, and she called it useless? Seriously?

Because that's what the mainstream sees.

Newsflash. I'm sorry. Dylan, Haskell, Idris: these aren't main-stream, and they never will be.

Algebraic types? Dependent types? You'll never see them. They're too ... research-y. They stink of academe, which is: they stink of uselessness-to-industry. You'll be dead and buried to see them in this form, even after they discover the eternity elixir. Sorry.

Or you'll see them in Visual Basic as a new Type-class form that only a few Microserfs use because they happened to have written those extensions. Everybody else?

Nah.

Here's how Dylan will succeed, right now.

Bruce and I will put our heads together, start a company, and we'll code something. Not for anybody else to use and to love and to cherish, just for us, only for us, and it will blow out the effing doors, and we'll be bought out for $40M because our real worth is $127M.

And the first thing that Apple will do, after they bought us, is to show us the door, then convert the code into Java. Or Swift. Or Objective-C, or whatever.

And that's how we'll win.

Not the $40M. Not the lecture series on 'How to Make Functional Programming Work in Industry for Real' afterwards at FLoC and ICFP conferences with fan-bois and -girls wanting to talk to us afterwards and ask us how they can get a job doing functional programming.

Not that.

We'll win because we made something in Dylan, and it was real, and it worked, and it actually did something for enough people that we can now go to our graves knowing that we did something once with our lives (and we can do it again and again, too: there's no upper limit on the successes you're allowed to have, people) that meant something to some bodies. And we did that. With Dylan.

Nyaah!

I've done that several times already, by my counting: the Prolog project, the Dylan project, the Mercury project, and my writing.

I'm ready to do that, again.

Because, actually, fundamentally, doing something in this world and for it ... there's nothing like it.

You write that research paper, and I come up to you, waving it in your face, demanding you implement your research because I need it to do my job in Industry?

I've done that to three professors so far. Effing changed their world-view in that moment. "What?" they said, to a person, "somebody actually wants to use this?" The look of bemused surprise on their faces?

It was sad, actually, because they did write something that somebody out there (moiself) needed, but they never knew that what they were doing meant something.

And it did.

Effing change your world-view. Your job? Your research? Your programming language?

That's status quo, and that's good and necessary and dulce and de leche (or decorum, I forget which).

But get up out of the level you're at, and do something with it so that that other person, slouched in their chair, sits up and takes notice, and a light comes over their face and they say, 'Ooh! That does that? Wow!' and watch their world change, because of you and what you've done.

Dylan is for nothing and for nobody.

So is everything under the Sun, my friend.

Put your hand to the plow, and with the sweat of your brow, make it yours for this specific thing.

Regardless of the long hours, long months of unrewarded work, and regardless of the hecklers, naysayers, and concerned friends and parents, and regardless of the mountain of unpaid bills.

You make it work, and you don't stop until it does.

That's how I've won.

Every time.

Friday, August 1, 2014

1HaskellADay July 2014 problems and solutions


  • July 1st, 2014: (text:) "Hi, all! @1HaskellADay problems are back! #YAY First (renewed) problem: a verbose way of saying, 'Give me a deque!' http://lpaste.net/106748" Deque, last, and all that (verbose version with hints) (solution: Deque the halls (with my solution): Data.Deque)
  • July 2nd, 2014: (text:) "Today's Haskell exercise: Vectors, length in constant time, and (bonus) reverse return in constant time. http://lpaste.net/106843" Vector (solution: Vector: Magnitude, and Direction, OH, YEAH! Data.Vector)
  • July 4th, 2014: (text:) "Today's exercise(s). pack/unpack. encode/decode. Cheer up, sweet B'Genes! http://lpaste.net/106912" Cheer up, Sweet B'Genes (solution: GATTACA)
  • July 7th, 2014: (text:) "#haskell daily exercise: ROLL with it, Baby! http://lpaste.net/107047  ('cause I'm feeling a little #forth'y')" Roll (solution: Rollin' on the (finite) river)
  • Bonus problem: July 7th, 2014: (text:) "For those who found the 'roll'-exercise trivial; here's (a more than) a bit more of a challenge for you to play with. http://lpaste.net/107023" Acid rules! (solution: "A solution set to today's challenge exercise: BASIC ... http://lpaste.net/107069  ... and Acidic http://lpaste.net/107071  ... WHEW! That was fun!" BASIC ... and Acitic)
  • July 8th, 2014: (text:) "Today's #Haskell exercise: LOTTO! Powerball! Mega-whatever! Who's the big winner? http://lpaste.net/107104" Lotto (solution: "And the big winner today is ... solution-set to today's #Haskell lotto exercise http://lpaste.net/107130" ... and the winner is ...)
  • Bonus problem: July 8th, 2014: (text:) "#bonus #haskell exercise: Well, that was RND... Randomness, and such (or 'as such') http://lpaste.net/107105" Well, that was RND (solution: For YESTERDAY's bonus question of roll-your-own-rnd-nbr-generator, here's one as comonadic cellular automata (*WHEW*) http://lpaste.net/107205: Data.Random)
  • July 9th, 2014: (text:) "Okay, ... WHERE did yesterday and today GO? :/ #haskell exercise today: "Hey, Buddy!" http://lpaste.net/107181  I will post solution in 4 hours" Hey, Buddy! Distinct sets-of-an-original-set. (solution: "Here's a story ..." A(n inefficient) solution to bunches and cliques. http://lpaste.net/107273" Brady Bunch)
  • July 10th, 2014: (text:) "Today's #haskell list-exercise: "Get out of the pool!" http://lpaste.net/107286  Will post a solution at 9 pm EDT (which is what time CET? ;)" (solution: "She's a supa-freak! She's supa-freaky! (Bass riff) A solution to today's #haskell exercise about list-length-ordering http://lpaste.net/107308")
  • July 11th, 2014: (text:) ""It's Friday, Friday!" So, does that mean Rebecca Black wants to code #haskell, too? Today is a Calendar #exercise http://lpaste.net/107328" (solution: ""In a New York Minute": a solution to today's #haskell exercise that took WAAAY more than a minute to complete! #WHEW http://lpaste.net/107370")
  • July 14th, 2014: (text:) "Today's #haskell exercise: isPrime with some numbers to test against. They aren't even really Mp-hard. ;) http://lpaste.net/107463" First stab at primality-test (solution: "A simple, straightforward stab at the test for primality. #haskell #exercise http://lpaste.net/107483" The start of a primal inquiry
  • July 15th, 2014: (text:) "Primes and ... 'not-primes.' For a prime, p, a run of p-consecutive 'plain' numbers is today's #haskell exercise: http://lpaste.net/107536" (solution: "So ya gotta ask yerself da question: are ya feelin' lucky, punk? Run of p non-primes in linear time http://lpaste.net/107549  #haskell exercise." Alternate solution by Gautier: http://lpaste.net/107549)
  • July 16th, 2014: (text:) "Difference lists? We no need no steenkin' Difference lists!" http://lpaste.net/107593  DList in #haskell for today's exercise. (solution: "DLists? We got'cher DList right here! A solution to today's #haskell exercise is posted at http://lpaste.net/107607")
  • July 17th, 2014 (text:) "http://lpaste.net/107652 Prélude à l'après-midi d'un Stream ... I thought that last word was ... something else...? #haskell exercise today." Comonads for lists and Id. (solution: "Control.Comonad: That was easy! http://lpaste.net/107661 #haskell exercise #solution" Learn you a Comonad for Greater GoodFunny story, bro'! id is not necessarily Id. (I knew that.) http://lpaste.net/107662 #haskell solution")
  • Bonus exercise: July 17th, 2014 (text:) "Streams are natural, streams are fun, streams are best when ... THEY'RE BONUS QUESTIONS! #bonus #haskell exercise http://lpaste.net/107655" LET'S GET THIS PARTY STARTED! (solution: "Take this Stream and ... it! #solution to today's #haskell #bonus exercises http://lpaste.net/107665")
  • July 18th, 2014: (text: "Today's #haskell exercise: Frère Mersenne would like a prime, please. http://lpaste.net/107700") (see solution next bullet)
  • Bonus exercise: July 18th, 2014 (text: "#bonus prime-time! Frère Mersenne would be pleased with a partial proof of a prime ... in good time. http://lpaste.net/107701") (solution: "A #haskell #solution for (monadic?) primes and the #bonus interruptible primes. http://lpaste.net/107708") Primary primes.
  • Bonus-bonus exercise: July 18th, 2014 (text: "Ooh! π-charts! No. Wait. #bonus-bonus #haskell exercise. http://lpaste.net/107702") (solution: "#bonus-bonus: a #solution http://lpaste.net/107718")


  • July 21st, 2014: (text: "Demonstrating coprimality of two integers with examples #haskell exercise http://lpaste.net/107819") (solution: "A coprimes solution #haskell problem is at http://lpaste.net/107843")
  • July 22nd, 2014: (text: "The prime factors of a number (and grouping thereof) as today's #haskell exercise. http://lpaste.net/107878") (solution: "OKAY, THEN! Some prime factors for ya, ... after much iteration (torquing) over this #haskell exercise solution. http://lpaste.net/107939")
  • Bonus exercise: July 22nd, 2014: ("For today's #bonus #haskell exercise you'll find a Bag 'o gold at the end of the rainbow http://lpaste.net/107881") (solution: "Second things first: a definition for the Bag data type as today's #bonus #haskell exercise. http://lpaste.net/107815")
  • July 23rd, 2014: (text: "Today's #haskell exercise, two variations of Euler's totient function http://lpaste.net/107955") (solution: "And, for a very small φ ... http://lpaste.net/107972  is a solution-set to today's #haskell exercise.")
  • July 24th, 2014: (text: "WEAKSAUCE! Goldbach's conjecture irreverently presented as a #haskell exercise. http://lpaste.net/108019") (solution: "That solution to today's #haskell exercise will cost you one Goldbach (*groan!*) http://lpaste.net/108059")
  • July 25th, 2014: LOGIC! Peano series: it's as easy as p1, p2, p3 ... http://lpaste.net/108099 ... in today's #haskell exercise. "Excuse me, Miss, where do we put this Grande Peano?" A solution to today's #Haskell exercise in the HA!-DSL http://lpaste.net/108140
  • Bonus: July 25th, 2014: http://lpaste.net/108108 Bonus #haskell problem for today. But not as easy as λa, λb, λc ... Church numerals and booleans. Correction: Ooh! forall! Church encodings and Haskell have a funny-kind of relationship. Updated the #bonus #haskell exercise with rank-types and forall. Solution: "Gimme that olde-time Church encoding ... it's good enough for me!" A solution to today's #bonus #haskell exercise http://lpaste.net/108114
  • July 28th, 2014: George Boole, I presume? Today's #haskell exercise: http://lpaste.net/108272. Solution: This and-or That ... a NAND-implementation of today's #haskell exercise at http://lpaste.net/108295
  • July 29th, 2014: Readin' 'Ritin' 'Rithmetic: today's #haskell exercise http://lpaste.net/108358 Solution: That's alotta NANDs! A solution to today's exercise at http://lpaste.net/108387
  • July 30th, 2014: ACHTUNG! BlinkenLights! Today's #haskell exercise http://lpaste.net/108420. Solution: Let it Snow! Let it Snow! Let it (binary) Snow! A solution to today's exercise is at http://lpaste.net/108429 
  • July 31st, 2014: π-time! http://lpaste.net/108485 Today's #haskell exercise. BLARG! UPDATE! Please read the update attached to the problem statement, simplifying the calculation quite a bit: http://lpaste.net/108485 Solution: Apple or coconut π? A solution to today's problem http://lpaste.net/108494

Notes on the problems
  • July 9th, 2014. I didn't quite know how to go about this, so I made several attempts with the State pattern. But how to describe it? It's the base pool from which you draw, and each (sub-)choice-point affects it, what's that type? I spent way too much time trying to discern the type, and failing. But now, a much simpler approach suggests itself to me (after experiencing the New York Minute exercise): this is simply a permutation of the list, and then that permutation is partitioned by the sizes of the groups! Implementing permute-then-partition is a much simpler approach than tracking some monster monadic state transformer.

    No, that didn't work, either. A permutation will give you [[1,2], ...] and[[2,1], ...] That is, all solutions, even the redundant ones. So, I reworked the problem simply following the data. With takeout feeding the iterative-deepening function, I finally got a guarded state-like thingie working fast and correctly. The new solution is on the same page as the old one.
  • July 11th, 2014. The New York Minute problem demonstrates the implementation of a rule-based classifer. It takes unclassified numeric inputs, and based on the cues from the rules, either types each number into day, month, year, hour, minute, or rejects the input data as unclassifiable. I was pleased to have implemented this entire system in less than two days of work! Sweet!
  • July 22nd, 2014. So, I've been running up against the double-exponential cost of computing a stream of primes for some time now since I gave the solution to the question of demonstrating the Prime Number Theorem. So now I have to tackle of bringing down that extraordinary, or unreasonable, cost down to something useable, and along those lines (of feasibility), I'm thinking of instead of regenerating and re-searching the primesish stream that we have some kind of State-like thing of ([already generated primes], indexed-primesish) ... something like that. Solution: "General improvement of problem-solving modules in anticipation of solving today's #haskell exercise, including primes: http://lpaste.net/107480"