iterators again

[David B. Held]

Jarrett Billingsley wrote:
> [...]
> Sounds like a good place for a D development blog.

A blog is an interesting idea, but Walter doesn't strike me as the 
blogging type.

> [...]
> What's a bit weird is that sometimes some feature will just come out of 
> nowhere, with no announcement, no pre-discussion, no communication 
> whatsoever besides a bullet on the "what's new with DMD 1.xxx" page.  And 
> they're not always trivial features either -- import() and mixin() 
> expressions, struct literals, scope() statements, and nested classes were 
> all "well OK, never saw that one coming" features.

Walter is trying to evolve the language as quickly as possible, and even 
when he sets priorities, he later changes them for various reasons. 
There has been discussion of a roadmap to help smooth things out, but 
you should harrass Brad Roberts about that.  Which would you prefer: 
vaporware announcements or surprise announcements?  I'll always take 
overdeliver to overpromise every day of the week and twice on Sundays.

> [...]
> Walter also seems to be influenced by certain people more than others, which 
> is more than a little disturbing. 

Is it more disturbing than the President being more influenced by his 
cabinet than his countrymen (say, yourself...assuming you aren't part of 
the Bush cabinet, of course)?  Is it more disturbing than Bjarne 
designing C++ with his local colleagues at AT&T?  More disturbing than 
James Gosling designing Java with a few friends at Sun?  More disturbing 
than Larry Wall single-handedly deciding what goes in Perl 6?

Pick a successful language, and I will show you a language that was not 
designed by democracy.  Unlike your government, you can pick your 
language, and I presume people have picked D so far because they thought 
it was designed well (or at least better than the alternatives for a 
sufficient range of cases).  If you can find a D-size language designed 
by a community to show that that is a Better Way, I'm sure we would all 
be both surprised and delighted to hear about it.  Until then, the proof 
of the pudding is in the eating, and everyone who has tasted the 
language design pudding has decided that a small group of individuals, 
however flawed they may be, produces a better product than a whole 
community.

At some point in a language's evolution, the weight of the community 
increases and becomes the dominant force.  COBOL, FORTRAN, Lisp, C and 
C++ are all too mature to be run by a maverick designer any more.  Even 
Java is moving towards a more heavily community-weighted input, and it's 
still a relatively young language.  However, D is barely out of the 
starting gate (and by some measures, has not even left the gate yet) and 
already you want to saddle it with bureaucracy?  Maturity leads to 
fossilization.  When everyone has an equal say, conflicting interests 
stifle progress.

All of the languages I mentioned could be seriously upgraded...if not 
for the fact that they all have millions of users whose code would get 
broken if you made more than minor changes.  Look at Java...despite 
being less than 10 years old, they couldn't even add first-class 
generics due to backwards compatibility.  C# is only a few years 
younger, but was able to do things the right way because its VM hadn't 
totally fossilized yet.

Walter knows and recognizes that everyone wants D to succeed.  And trust 
me, he reads the NGs as much as anyone, and takes the community quite 
seriously.  But at the end of the day, Walter only has one brain to 
design D with, and fully analyzing every input available to him would 
seriously overcommit the computing resources of that brain, however 
oversize it may already be.  I think he takes a very meritocratic 
approach while conceding the pragmatism of geography; and if you stack 
up his design approach against any other major language designer's, I 
think you'll see that few other designers are nearly as democratic at 
this stage in a language's development.

I personally wouldn't want a language designed by a committee of equals, 
because I've seen software produced that way, and it almost universally 
sucks.  If you don't trust the people who are designing a system, well, 
there's not much you can do about that but complain.  But if they've 
produced a decent product so far, perhaps they have earned some trust 
and should be given the benefit of the doubt.  If you can argue 
convincingly that the "cabal that meets behind closed doors" isn't doing 
a good job, I'm sure your argument would be given serious consideration. 
  But merely complaining about a process without any demonstration that 
it's a bad process is arguing on principles that are not well-established.

Let me give you an example from real life.  Consider The Simpsons.  When 
The Simpsons was written by Matt Groening and one or two other writers, 
the show was considered brilliant.  When Groening left to do other 
things and the show was handed over to a team of writers, the show was 
considered mediocre and continued merely on the weight of its momentum 
and the brand that Groening had established.  Now imagine if Groening 
had opened up The Simpsons writing to his fanbase, which would be the 
democratic thing to do.  How good would you think the episodes would be 
then?  Out of all the fans, there may be one or two really brilliant 
writers who could actually have improved the show.  But with "one man, 
one vote", their voices would have been lost in the cacophony of severe 
sub-mediocrity.

That's not to say that folks on the NGs are all mediocre or less.  I 
think the fact that they are willing to try out a new language puts them 
in a different class than your average coder.  But I will say that most 
of them are not committed programming language designers, and if you 
left them to design a language, they might create one or two really nice 
features, and then move on to other things.  That doesn't give you the 
kind of experience needed to understand how much a feature will cost to 
implement and how much value it provides relative to other ways of doing 
things.  Perhaps Walter is missing out on those one or two brilliant 
writers lurking out there.  But I don't think he is.

The people who push the language to its limits demonstrate the need for 
various features in quite palpable ways.  Those are the best arguments 
for change, and it doesn't take a committee to see that.  Do you want to 
know why Andrei has so much influence in D's design?  Because he uses D 
on a supercomputing cluster.  He showed that one of his programs was 
unnecessarily slow because of the design of the allocator and suggested 
an alternative that was known from C++.  When Walter implemented it, he 
found that it did indeed give enormous performance benefits.  Andrei 
argued for struct literals because it would seriously improve the 
readability of his code.

Andrei isn't just some "blessed expert".  He earns his clout through his 
code.  Is it fair that he has access to a cluster that most of us will 
never see?  No.  But do you want the design of D to be hamstrung by 
"fairness"?  I think Walter's approach to language design is fairly 
inspired because it is not so much about individual people as it is 
programs.  Walter cares about which programs are easiest to write and 
fastest to run, and it is these programs that guide his decisions. 
People turn out to be the representatives of the programs, even when the 
programs aren't actually written (because people have to write them 
after all), but it isn't so much the people themselves who influence the 
design so much as the resulting programs.

It's something like a genetic algorithm for language design.  The most 
compelling programs are assigned higher fitness and given more "survival 
tools" (features) for the next round of culling/evaluation.  Since 
Walter cannot directly inspect the space of all D programs in any 
meaningful way, he needs people to help him visualize the important 
parts of the space.  If he could do that without relying on people, I'm 
fairly certain that there would be nobody advising him.  Anybody can be 
that person that says: "Hey, look here!  This is a very interesting 
portion of the program space!"  But some people are naturally better at 
doing that than others, and that is why it appears that some people have 
more influence.  In reality, it is that some *programs* have more influence.

Perhaps Andrei seems like a Dick Cheney or Karl Rove, pulling Walter's 
strings like a political puppet.  But you quickly become disabused of 
that notion when you see Walter say to Andrei point-blank: "No, you are 
wrong!"  Even Andrei has to concede at the end of the day that Walter 
will decide for himself what is best for the language.  But if it 
appears that he has more influence, it is only because he has worked 
much, much harder for it than most.  Like I said, it's very 
meritocratic.  And like I said, while the people who meet with Walter 
face-to-face are constrained by geography, the people who have influence 
on the language's design are not.

I don't think Walter needs any defense of his approach to language 
design, but I hope that gives you a little insight from the perspective 
of someone who has seen both sides.

Dave