Larry Wall - Present Continuous, Future Perfect

This is a transcription of Larry Wall's talk "Present Continuous, Future Perfect".

Thanks to all the volunteer transcribers and editors.

The recorded talk can be downloaded from here: http://www.osdc.org.il/2006/html/larry-wall-2006-02-26.mp3 (52 Mb)

The slides for the lecture are also available online courtesy of Mr. Wall. It is advisable to follow them when listening to the audio or going over the transcript.

Update: (2006-06-03) Chris Dolan has prepared a video representation (slides along with audio) of the lecture.

Table of contents

Introduction - Where Perl Draws Ideas From

Good morning.

Many time I have shown this picture in many of my talks. Perl derives from many sources, these are probably the four biggest ones. You should not take this picture to mean that Linguistics is the opposite of common sense. (laughter) But there are many computer scientists who have complained that it should more like engineering and less like an art and I disagree, but that's OK.

But Perl also draws ideas from many other sciences - Ecology, Math, Sociology, etc., etc., etc. I have given talks about those too, but I'm not gonna do that today. Also Joel's ideas from the science of Golf. The Biology of Perl ... there's lots of different aspects of that, that Perl has borrowed metaphors from, which we could also talk about. You can feel free to ask me about any of these things later.

And you know what the Harvard Law is. That says that under controlled conditions of water, temperature, light, food - the organism will do as it damn well pleases (laughter).

How Perl Entered the Unix Universe (2:06)

Anyway, speaking of paleontology, I dug up this fossil. It's kind of a dinosaur egg. And when Perl was first starting out this was how people saw the universe. If you were a Unix programmer you either programmed in C or shell. And there really wasn't much in between. There were these little languages that we used on top of shell, but that was the big divide. The big revelation that hatched Perl, as it were, was that this opened up into a two-dimensional space. And C was good at something I like to call manipulexity, that is the manipulation of complex things. While shell was good at something else which I call whipuptitude, the aptitude for whipping things up.

So Perl was hatched. As a small egg. That was Perl 1. And it was designed from the very beginning to evolve. The fact that we put sigils in front of the variables meant that the namespaces were protected from new keywords. And that was intentional, so we could evolve the language fairly rapidly without impacting.

And it evolved... And it evolved... And finally we got to Perl 5. And... So... Perhaps the Perl 6 slogan should be "All Your Paradigms Are Belong To Us". We'll get to that.

Why Perl Evolved As It Did (3:54)

So we had these five successful organisms. If you look at the Perl culture right now, it is very healthy, very vibrant... Lots of stuff going on. Nobody can keep track of it all. Why? Well, there are evolutionary reasons, psychological reasons, linguistic reasons, anthropological reasons, and, of course, sometimes there is just no good reason. So I'd like to start off with my own irrationalities.

I don't think syntax should dangle in the wind. I'm with Aristotle. I think things should have a beginning, a middle, and an end. Which means I like K&R bracketing. I do not like the way that Python hangs stuff out there, with no end.

I think that ordinary people dislike abstraction. That's because I dislike abstraction and I think I'm ordinary. (laughter) I might be wrong about that, but I don't know.

I simultaneously believe that languages are wonderful and awful. You have to hold both of those. Ugly things can be beautiful. And beautiful can get ugly very fast. You know, take Lisp. You know, it's the most beautiful language in the world. At least up until Haskell came along. (laughter) But, you know, every program in Lisp is just ugly. I don't figure how that works.

I think visual metaphors are very important. How it looks. Different things should look different. Similar things should look similar. A language designer simultaneously has to care what other people think, and has to not care what other people think. Otherwise you go crazy. Well, crazier. (laughter)

And finally, I think God has free will. And therefore he created programmers with free will and that they ought to be given choices.

Irrationalities in Other Languages (5:54)

Now, I'm not the only language designer with irrationalities. You can think of some languages to go with some of these things.

"We've got to start over from scratch" - Well, that's almost any academic language you find.

"English phrases" - We'll that's Cobol. You know, cargo cult English. (laughter)

"Text processing doesn't matter much" - Fortran.

"Simple languages produce simple solutions" - C.

"If I wanted it fast, I'd write it in C" - That's almost a direct quote from the original awk page.

"I thought of a way to do it so it must be right" - That's obviously PHP. (laughter and applause)

"You can build anything with NAND gates" - Any language designed by an electrical engineer. (laughter)

"This is a very high level language, who cares about bits?" - The entire scope of fourth generation languages fell into this... problem.

"Users care about elegance" - A lot of languages from Europe tend to fall into this. You know, Eiffel.

"The specification is good enough" - Ada.

"Abstraction equals usability" - Scheme. Things like that.

"The common kernel should be as small as possible" - Forth.

"Let's make this easy for the computer" - Lisp. (laughter)

"Most programs are designed top-down" - Pascal. (laughter)

"Everything is a vector" - APL.

"Everything is an object" - Smalltalk and its children. (whispered:) Ruby. (laughter)

"Everything is a hypothesis" - Prolog. (laughter)

"Everything is a function" - Haskell. (laughter)

"Programmers should never have been given free will" - Obviously, Python. (laughter)

So my psychological conjecture is that normal people, if they perceive that a computer language is forcing them to learn theory, they won't like it. In other words, hide the fancy stuff. It can be there, just hide it.

Linguistics (8:22)

Going from psychology to linguistics, I think that Perl succeeds primarily because it does behave like a natural language. Some of the things that underlying deep principles rather than cargo culting in phrases like Cobol did, we try to aim for deep principles of linguistics. With a natural language you learn as you go. You know, you learn it once and use it many times. So, you should optimize for expressiveness and not for ease of learnability. Learnability is OK, but expressiveness is more important.

Competence: We do not expect a fifty year old to speak with the same level of confidence as a five year old. And that's OK. You can talk baby talk.

There are multiple ways to say the same thing.

There is no shame in borrowing words. Unless you're French. (laughter)

Back to dimensionality. When you are saying something linguistically, it's like taking a trip. You know, when you take a trip from California to Netanya, you don't go straight south and then straight west and then straight north. It's not orthogonal. There are little bits at the beginning. Then you take bigger hops on the planes and then you take littler hops at the end. Language works the same way, it's fractal. There is little orthogonality. At least apparently; you can have orthogonal views of it, there are orthogonal subsets. But there are multiple orthogonal subsets. At first glance it just looks like a network, and you have to navigate the geography.

More natural language things, which we'll probably skip over most of them, but they have corresponding things in Perl.

Natural language develops, without an overriding theory. Your peers enforce your style. The language does not enforce your style.

Design is distributed.

Dialects will diverge, so you should plan for it.

It's useful to take things from multiple languages and jam them all into one language. You may have noticed some of that in Perl.

And finally, a language without a culture is dead. It's not just an intellectual exercise to design a language. You have to have the culture to go with it. Which leads us to my anthropology ideas, which is that Perl culture is just as important as the language itself because the culture can fix the language but not vice versa.

Anthropology (11:10)

Now in terms of the anthropology we try to welcome people into the tribe. We allow people to have their own little fiefdoms, where they are the ruler and can beat up on their followers.

We try to let people share with each other. We try to capture knowledge. Both of those things are why we have the CPAN, Comprehensive Perl Archive Network, which is arguably one of the greatest repositories of reusable crappy software in the world. (laughter).

And we have a culture of cooperating with other cultures too. We try to make Parrot so that other languages can ran on top of that. We've always tried to hook up Perl with everything. In kind of a humble sort of way. And finally it's culture of fun. At least we try to make it that way. And that's why I give weird talks.

Other Sciences (12:14)

What about the hard sciences? You know there is physics, especially quantum physics. That's pretty hard. (laughter) Chemistry might be a hard science if you think of it as physics. And we are not so sure. Golf is pretty hard though.

Now, I've often shown this waterfall chart, and many of you have probably seen it. If you turn it into a fractal feedback loop and if you run it the other direction, then you have extreme programming. But the point is, you're racing around like this, in a strange attractor of some sort or other. But how do you know you're done? You know, other than the tests tell you you're done. But are the tests right? Who knows?

Well, we all who have studied sciences have seen these potential energy curves. Now, if you're an economist you turn this upside down and you have hill-climbing algorithms. But it's just the same thing.

The ball wants to roll downhill. Are we optimal? No. Is it optimal now? Maybe. Maybe not. We could be in a false minimum. Now, we're happy where we are, but we don't yet know that we might be happier over there. It takes a certain amount of energy to climb over the hill and take a peek. You can do it that way (go over the peak), or you can find sneaky ways to get through (tunnel through). Arguably, this is the Parrot approach. And this is the Pugs approach. But however you get there, it's the Promised Land. I think you guys know something about the Promised Land. (laughter). You know, if you go back here, here is Jacob and Caleb wanting to sneak into the land, and here's the Children of Israel wanting to spend forty years in the desert. (laughter) Hopefully we can get Perl 6 out in less than forty years. (laughter) But if that's how long it takes, we're still gonna get there.

Fan Mail (14:42)

Q: "Dear Larry, I love Perl. It has saved my company, my crew, my sanity and my marriage. After Perl I can't imagine going back to any other language. I dream in Perl, I tell everyone else about Perl. How can you improve on perfection? Signed, Happy in Haifa."
A: "Dear Happy,
You need to recognize that Perl can be good in some dimensions and not so good in other dimensions. You also need to recognize that there will be some pain in climbing over or tunneling through the barrier to the true minimum."

Now Perl 5 has a few false minima. Syntax, semantics, pragmatics, (laughter), discourse structure, implementation, documentation, culture... Other than that Perl 5 is not too bad.

Q: "Dear Larry,
You have often talked about the waterbed theory of linguistic complexity, and beauty times brains equals a constant. Isn't it true that improving Perl in some areas will automatically make it worse in other areas? Signed, Terrified in Tel-Aviv."
A: "Dear Terrified,
...
No." (laughter)

You see, you can make some things so they aren't any worse. For instance, we changed all the sigils to be more consistent, and they're just the same length, they're just different. And you can make some things much better. Instead of having to write all this gobbledygook to dereference references in Perl 5 you can just do it straight left to right in Perl 6. Or there's even more shortcuts, so multidimensional arrays and constant hash subscripts get their own notation, so it's even clearer, at least once you've learned it. Again, we're optimizing for expressiveness, not necessarily learnability.

Q: "Dear Larry,
I've heard a disturbing rumor that Perl 6 is turning into Java, or Python, or (whispered:) Ruby, or something. What's the point of using Perl if it's just another object-oriented language? Why are we changing the arrow operator to the dot operator? Signed, Nervous in Netanya."
A: "Dear Nervous,
First of all, we can do object orientation better without making other things worse. As I said. Now, we're changing from arrow to dot, because ... because ... Well, just 'cuz I said so!"

You know, actually, we do have some good reasons - it's shorter, it's the industry standard, I wanted the arrow for something else, and I wanted the dot as a secondary sigil. Now we can have it for attributes that have accessors. I also wanted the unary dot for topical type calls, with an assumed object on the left and finally, because I said so. Darn it.

Future Perfect (17:25)

So, let's go from the Present Contiunous, imperfect though it may be, to the Future Perfect, to at least Perfecter.

Now, when we started off this Perl 6 adventure, we announced - "The community is going to rewrite Perl 6. It's gonna rewrite the language, the culture, the development, everything." So we opened it up, and I expected maybe twenty RFC's for changes. We got three hundred and sixty one. And they were all over the map. They contradicted each other, they had some really bogus solutions, and they all suffered from the same problem, which is: each of them assumed that this was the only change you are going to make to Perl 5. And otherwise everything else will be the same.

Well, that's not how you do a redesign. If you do a redesign that way, you end up with a mishmash. So, one of the things I discovered, for instance, is that the First Law of Language Design is that "Everyone wants the colon" for their particular syntax. And they can't all have it because it's contradictory. So the Second Law of Language Redesign is "Larry gets the colon for whatever he wants". (laughter)

I have to take a Winnie the Pooh approach. This was just way too much information for one person. You know, even somebody as smart as Audrey could not comprehend all of this at once. And I'm stupider than Audrey, at least in some ways. So I took the ways in which I am stupid as a sort of a model of what Perl programmers like. So I figured if I couldn't understand something very easily it probably wouldn't be very understandable to anybody else, so I would be very careful about introducing that sort of a feature. You know, we call it the "Bear of Very Little Brain approach". It works for me because that's the only thing I can do.

Right at the beginning we said, "There is never time to do it right, but there is always time to do it over". So we're just gonna take the time to do it right. You've heard this saying "Good, Fast, Cheap. Pick two.". Well, this is Open Source. We have to do it cheap. Therefore, it reduces to a problem of "Good or Fast. Pick one." We chose Good. We did not choose Fast. You may have noticed this. (laughter) Considering we announced Perl 6 five and half years ago. We're still going to do it right. As long as it's a converging process, we'll get there. It is really simple. Just keep everything good and throw out everything bad... that's easier said than done.

Goals and Meta-Goals (20:33)

So our goals, which we'll cover in the rest of our time here:

But behind those are some meta-goals.

Pick defaults that allow for evolution: We found out the problem for Perl 5. It allowed for evolution. It put object orientation completely orthogonal ... orthogonally - I'm an English speaker and I can't even say the word - into the language. It's wonderfully orthogonal but it's too orthogonal because it doesn't set up any good defaults, and everyone does it differently and nothing can interoperate. So in Perl 6 we'll pick good a default object orientated way to do things, but allow for evolution.

Declarative multiculturalism: Multiculturalism is a very sort of postmodern idea. Multiculturalism is great as long as you know that you are in a multicultural situation. Perl 6 is not going to be a single language. You start off with standard Perl 6 at the top, and every time you say "use" or declare a macro or something you are changing into a different language, a different culture. As long as you declare it, all is fair if you predeclare.

Context dependency: I don't really have time to explain that.

Better encapsulation: But also ex-capsulation. You shouldn't have to worry about the innards of a module, but there are ways in which the innards of a module shouldn't have to worry about the outside world. Unless it wants to. Not just modules, but any kind of scoping mechanism. Any kind of lexical scope, package scope... There are lots of different scopes. Different ways to isolate things dimensionally.

DWIMmery: That's "Do What I Mean"-ery. When we design a dwimming feature in Perl 6, we try to put it in one spot and then reuse it everywhere, so you know which table to look up to find out for instance how a smart match operator works.

No arbitrary limits round two: Perl started off with the idea that strings should grow infinitely, if you have memory. Just let's get rid of those arbitrary limits that plagued Unix utilities in the early years. Perl 6 is taking this in a number of different dimensions than just how long your strings are. No arbitrary limits - you ought to be able to program very abstractly, you ought to be able to program very concretely - that's just one dimension.

Behind that is another yet more-meta-goal, and that is we want to manage complexity in the future so that you people who are interested in changing Perl to do what you want can do that.

And there's a meta-meta, a meta-meta-meta goal, which is, of course, to enjoy life. Now when you're young, "enjoy life" means, "I enjoy life". When you start getting a little older like me, the way you enjoy life is by helping other people enjoy life. And there's a balance in there.

Simplifications (24:00)

So let's un-meta-meta-meta. Which of course means we're back to bikeshed painting here so all of you people who are not interested in Perl at all can go to sleep now.

I don't hear anyone snoring.

(I caught a cold before I came so this is an American cold not an Israeli cold. Don't worry about that.)

We made many simplifications, which we will go through here.

In Perl 5 it started off without nested data structures, because it was just a scripting language, so we sort of waved our hands and rationalized how these weird sigils were, but they were inconsistent. We've made them consistent: if it's an array it always has an at on the front, if it's a hash it always has a percent, regardless of how you use it. And this gives us referential transparency.

I've mentioned the constant subscripts, we've stolen angle brackets for constant subscripts so they're very visually distinct. Rather than having a weird rule that if you have a bareword inside curlies, we now give it a separate construct, and there's just lots of ways in which this clarifies your intent. So, inside curlies is now always an expression, no special rules. We're trying to get rid of all these special rules.

A my variable in Perl 5 extends to the end of its closing block - except that if a my variable is declared, blah blah, blah, ...it only extends, blah, blah, blah... and only for some times.... My eyes glaze over. If I were Damian I would put gibberish in there.

In Perl 6 the rules are going to be a little bit simpler: it extends to the end of the enclosing block and no further. Rule 2: There is no Rule 2.

There's no more double parsing. Perl 5 was always looking for the end of the string, and then trying to figure out what was in the string, which is a mistake. There's no problem with having an expression containing quotes inside a double-quote string, even though it's the same type of quote. It knows what it's looking for at that particular time.

Now, Perl 5, Perl 5 would blow up on this pattern, and the reason it would blow up on it, actually it should probably have a /x on the end. But leave that aside. How many can see the problem there?

See, the slash - well, maybe you can't see it. Now you really can't see it! The slash in the comment there matches the beginning slash. So Perl 5 thinks that's the end of the regular expression, and gets very confused. That's fixed in Perl 6, because of this one-pass parsing. When you're in a comment, it knows it's in a comment.

Simpler precedence. We reduced the number of precedence levels a little bit - by one or two. It went from 24 to 22 or something.

Regularizations (27:49)

There are fewer gotchas. You know, you had to put parens some places you didn't expect to. We had to unify the relational and equality precedence levels so that we could do chained equality operators the way mathematicians like to do.

Perl 5 is just all full of these strange gobbledygooky variables which we all know and love - and hate. So the error variables are now unified into a single error variable. These variables have been deprecated forever, they're gone! These weird things that just drive syntax highlighters nuts (laughter) now actually have more regular names. The star there, $*GID, that's what we call a secondary sigil, what that just says is this is in the global namespace. So we know that that's a global variable for the entire process. Similarly for uids.

A lot of these things ought to have been object properties of some object or other such as a file handle, and that has a global name. In Perl 5 there was this list of names that just magically was the same variable in every package, and you had to memorize that. You don't have to memorize that anymore because we have this star notation which says which things are globals. You can leave the star out and it will find the global one anyway, but you don't have to keep track of it. Likewise, STDIN and its friends were these mysterious everywhere-variables, and we even managed to shorten them while we were making them more regular.

Now, people are always saying, you know I program in C and I like to be able to leave the curlies off of my expression. Why can't we make the curlies optional? And that's always gone back to my irrationality of not liking dangling syntax. But I'm kind of glad now that I never made those optional, because it turns out that if you keep the curlies mandatory you can make the parentheses optional on the expressions. And it removes a lot of the visual clutter. Makes it less like Lisp. (laughter)

And those blocks, because the curlies are mandatory around a block, those blocks are all now logically closures. Especially if they use lexical variables outside, as this closure does. That's how you write in Perl 5, but in Perl 6 you don't have to say the "sub". Any block that's like this is just automatically a closure. (Unless it's a hash constructor.)

Now, another regularization was, in C and Perl and many C-derived languages you don't know which blocks require a semicolon on the end and which ones don't require a semicolon on the end, you just have to memorize it. Or just put semicolons everywhere and people think you're stupid. (laughter) In Perl 6 we've regularized that: any line-ending curly has an implicit semicolon on the end, so you just don't have to worry about it.

Things which were special cases in Perl 5 now just naturally fall out of these rules. (Aside: Gaal likes these.) Now this is just a little bit different; the comma there is now required after the curly or it would think it was the end of the list. But that's a small price to pay for the regularization that we now have.

Likewise, map was a special case in Perl 5; it is not a special case. sort was a special syntactic case. And instead of $a and $b being magical variables that you should avoid, we now have a placeholder syntax that works just the same only it's generalized. And it just says we have two parameters to this closure, they happen to be named a and b, and they will be passed in that order. And we always get the question saying, well, what if gets too complicated, how will I keep track of the alphabetical order? That's not what it's for! If it gets that complicated you should use a regular parameter list.

So, when you see something like a for loop, which is like foreach in Perl, that actually is logically a closure, a subroutine on the end there. It has an implicit parameter, $_, so it still works just the same way you think of it in Perl 5. But you could pass, using one of these explicit placeholder variables - that means exactly the same thing: I have one parameter. Print it.

33:28

Or we have another notation, which we call the pointy sub, which is another way of declaring a block with parameters. Now, if you know Ruby - it puts the parameter names inside vertical bars - I didn't like that. I like the way this reads, because you say "for" list and then bind it to this formal parameter for that block and print it. It's a lambda. It's just sort of unsigned or something. (laughter)

34:00

So, because we're just binding to a formal parameter list, it's just like you passed a sub as the last argument, only it reads a lot nicer. And because it's a formal parameter list you can have more than one parameter. We can have two parameters, so it takes the list two at a time. Or three at a time! No, the next slide's not four at a time.

You can take two arrays in parallel, and this each and the semicolon there is a way of piping two arrays in a single construct, and they're piping constructs, which I don't have time to talk about, but that's essentially a multi-dimensional list. It knows to take that, in parallel, so this allows you to read two arrays in parallel.

But on the other hand that looks a little bit, with the semicolons in there, like C's for loop, if you ignore the each, there's a little confusion there, so instead of using "for" which is a leftover from C, we've changed that keyword to "loop". But the three arguments still work the same way as they do in C, so you still have the generalized loop.

But having a different keyword is great there, because... we have this funny idiom for an infinite loop in Perl 5; we got it from C ["for(;;)"]. In Perl 6 you just say "loop" navigate. Now this would be like the control loop for a cruise missile, and when we speak of bombing out of the loop we mean it literally. [Laugh]

35:48

Perl 5 had this problem with "do" loops because they weren't real loops - they were a "do" block followed by a statement modifier, and people kept wanting to use loop control it them. Well, we can fix that. "loop" now is a real loop. And it allows a modifier on it but still behaves as a real loop. And so, do goes off to have other duties, and you can write a loop that tests at the end and it is a real loop. And this is just one of many many many things that confused new Perl 5 programmers.

And if you go and visit Perl Monks, and read through all the questions that people ask, you know, in about 9 out of 10 of them, you could, we could - we don't, because people get very tired of it, people already get tired of our hyping Perl 6. You could go through about 9 out of 10 of them, and say "This is going to be fixed in Perl 6", "This situation won't arise in Perl 6", "We've regularized that". And yet it's still Perl, it's Perl underneath, it's got a different syntactic sugar on the top.

Got rid of typeglobs. Instead of using typeglobs for aliasing, we have a new mystery operator, which I'll get to later, and there it is on the second line (wow! that was quick!). We'll talk about, not only about the simplifications, but these new powerful things and we can only cover a few of them.

We've got full type signatures now, on variables, in parameter lists, and they behave just as you would expect them to. Presuming you know what to expect.

37:53

There are also some lowercase ones, which are native types, rather than object types which are uppercase. Just by convention uppercase types are object-oriented types, whereas if I say my int Array, I mean a packed array of ints, or a packed bit array. So we can now talk about, optionally, very compact storage, and the VM is free to optimize this greatly. So we have the capacity of doing some things and using a lot less memory than Perl 5 does.

So these types are very useful, particularly with arrays - they can be packed.

Perl 5, another place where it was too orthogonal - we defined parameter passing to just come in as an array. You know arrays, subroutines - they're just orthogonal. You just happen to have one called @_, which your parameters come in, and it was wonderfully orthogonal, and people built all sorts of stuff on top of it, and it's another place where we are changing.

So instead of having these weird prototype pills, in Perl 6 you can just say "I'm expecting an array as the first argument" and then put the rest of the arguments into what we call this slurpy array, because slurps the rest of the arguments. So 1, 2 and 3 end up in args because we don't enclose them up there.

That's all well and good, but we'd like to unify these things. I said these RFCs proposed many different solutions. The hard design goal was to find the unifying concepts underneath all these RFCs and to treat the underline diseases rather than the symptoms. [40:00] So, we unified our argument passing which is a form of binding, with an operator which is used for binding - :=. That does exactly the same thing as passing this list into the argument list. There's no copying involved it just changes the names, so you can swap two names. We left equals (=) completely untouched because we think Perl 5 programmers will rise up and revolt if we change that.

Another powerful thing is the thing called hyper-operators. You know if you add two arrays together in Perl 5 - what do you get? Well, you end up with the length of the two arrays added. Well, guess what? It does that in Perl 6, by default, because it still makes the scalar/list context distinction, so it's a little bit smart about that. However, you can explicitly say: "I don't want to do it as in Perl, what I really mean by that plus (+) is add each element of @a to the corresponding element of @b and producing a list."

And for those explicit parallel operations we have what we call the hyper-operator - it's a meta operator, like assignment operators and some more things which modifies this operator to do a different operation. So, not only can you do that, but you can also have a scalar on one side - it's a little bit smart. So it just adds one to each element of @a and return that. Or you can do it in place, just increment each element of @a.

[Question from the audience: "How do you type it?"] How do you type it? With your keyboard? Do you use vi or Emacs? In vi it's Ctrl+K >>. You can also write it with two regular greater-than's (">>"). We're getting to the age of Unicode, and we want to make Unicode programming possible, so we're trying to shove people a little bit in that direction. So the actual operators we're encouraging people to use are all in the Latin-1 range, so this is in the Latin-1 range. But everything has an ASCII workaround to the Unicode, so it's not a big problem.

42:48.

Likewise, if you turn them inside out - the french quotes - you can use the regular angle brackets, and yes, we did change here-docs so it does not conflict, then that's the equivalent of "qw". This qw interpolates, with single-angles it does not interpolate - that is the exact "qw".

We have properties which you can put on variables and onto values. These are generalizations of things that were special code in Perl 5, but now we have general mechanisms to do the same things, they're actually done using a mix-in mechanism like Ruby.

Smart match operators is, like Damian say, equal-tilda ("=~") on steroids. Instead of just allowing a regular expression on the right side it allows basically anything, and it figures out that this wants to do a numeric comparison, this wants to do a string comparison, this wants to compare two arrays, this wants to do a lookup in the hash; this wants to call the closure on the right passing in the left argument, and it will tell if you if $x can quack. Now that looks a little strange because you can just say "$x.can('quack')". Why would you do it this way? Well, you'll see.

You can get match classes which will tell you whether this is a mammal, or whether it does the mammal interface. And another thing is against junctions. If any of you has seen Damian's quantum superpositions, this is just his any of 1, 2, 3. But we stole the short bit operator operations, because we think in Perl people don't actually use 2-bit operations that often - they now have something that is regularized differently. But now you can write this, and just means that it matches as 1 or 2, or 3. Junctions are used for smart operations, but they can be used anywhere.

Another thing we've done is - people always wanted the switch statement - so we now have a switch statement. But instead of saying "switch" and "case" which is talking about the constructs, we do it more like a natural language, where we have words that sound like it's in English. "given" one of these, when I have one or two or three, then argle, when B then bargle.

[Question from the audience: does it break by default?] Yes it breaks by default. If you want to fall into the next test, you would say "continue", and if you want to go into the body of the next test, as is the default in C, you'll need to use "goto"... we try to discourage that. But you don't have to say case, case, case, and all those fall through all these things, because we have these junctional things.

And these are what we call topicalizers, and that's another natural linguistic thing, but this is unified with the ~~ operator. The smart match is the same semantics under here, that's why we said we try to unify things. So a topicalizer is anything that says $_, and so when you're looping through an array, each element is set to $_ and when is just doing the smart match against that.

And, by the way, $_ is a lexical variable in Perl 6.

47:08

any block that sets $_, and that includes exception handlers. This catch - it's just a funny way to set $_ - from this block. And then there's a regular switch inside - we don't have special catch syntax like many other languages. It just naturally falls out with a little tweak.

Which is, other languages tend to put the "catch" outside of the block, and then you lose visibility to any of the lexical variables that were in the block. I don't refer to $x or $y here, but if I did, I could see it from inside the "catch" block there, which is inside the try. But the fact is, another advantage of putting it inside, is that it can turn its outer block into a try block. So we don't actually need that "try", it's redundant. We just write it like that - close it up. That is still a try-block, the sub is behaving as a try block, because it has a "catch" block in it.

48:27

There are many kinds of "catch"-like blocks that will capture control at various locations: when you're entering or leaving a block, when you're doing the next, the first time through this block; there's a keep and an undo block which is like a "leave" block except that it's transactional - it knows when the block is exiting successfully or unsuccessfully, so you can do transactional processing with it. "pre" and "post" blocks are for design by contract. But they're all the same thing - they're all just "begin" blocks. And "begin" blocks are all just properties on the blocks that they are.

Object Orientation

Moving on to object orientation... [49:15] We'll talk about some of these things. The syntax rather than setting run-time variable like "@ISA", is probably still doing things like that underneath, running bits of codes, but it looks declarative. And that's important, and the default declarations just map naturally to your typical object-oriented vocabulary.

So if you say the "dog is a mammal", that's how you write it. And you say "it has a tail" - "has"! We still use methods. Now this dot there allows us to auto-write our accessors for the attrs [= attributes] that have accessors. So you don't have to write those.

Here's a complete class, you can write it either this way if it's an entire file. Or you can write a class as a block with the block notation which is the more standard notation. And those are supposed to be the same thing. If it's an inner class, it will be installed on a different place.

50:30

So, as I mentioned, it auto-generates not only the accessors but also the constructor method, or defaults to a generic one that it can use. So I can say "Give me a new dog with a tail and legs", and you don't have to write those methods. Or the corresponding INIT methods which we call "build methods", which you can write if you want to, but you don't have to, if you just want the default.

Private variables, simply don't have the dot as a secondary sigil. We got tired of saying "secondary sigil", because we have a lot of them and they're very useful for indicating weird namespaces. It's "Weird things should look weird" is one of the principles in Perl 6, and the attributes are a little bit weird so they look a little bit weird.

The dot means that I'm autogenerating an accessor. If you leave that out, there isn't an accessor, and it's a private variable. It's lexically scoped to the class, and it can't be seen outside. And you have to use the accessors outside. And we have a way to specifying, finalizing, or not, on an application-wide basis, that we'll hopefully make that efficient without closing or finalizing classes prematurely.

Multi-methods are another important aspect which I really don't have time to get in to, but the entire Perl 6 run-time system is built on methods, so you can overload any of the built-in operators with more specific types and it will just find your version.

52:25

Delegation is supported, the natural syntax. Inner classes - you know about those, we talked about those. Roles are our idea of interfaces, but it's more than just interfaces. If any of you read the Smalltalks trait paper - it's more than just traits, but it's closely modelled on it. The idea is that it's an interface with a default implementation that you can pull in at compile-time and it will figure out if there are any method collisions.

We can also do run-time mixins. The problems with mixins is they overlay the previous definitions and you can hide things and not realize it. Whereas with roles - at compile time, if you pulled them in, and say what roles this objects fullfills, it can figure out if there's a conflict and warn you. Which is a good.

53:23

We support not only types, but constrained sub-types with subset notation, so we can say $x is an Int between 0 to 15, define odd numbers, define Japanese strings, define what a paddit is? If that's not on your list I'm sorry.

You can redefine subsets using a role, and these are assertions that the values of these types have to fullfill, the where clauses. And again, that's just a smart match operation on the right.

Functional Programming

Moving away from Object orientation, we have put a lot of thought into also supporting Functional programming. We're supporting many different paradigms to the extent that it's not exclusive to the other paradigms. So we can do continuations, and currying and lazy lists, etc. Pugs is written in Haskell currently, though we'll re-host the compiler in Perl 6 really soon. And Parrot already supports continuations - that's the low-level VM.

Continuations - if you know what they are - I don't need to explain them. If you don't know what they are - you don't want to know. But basically they are like backtracking points in a regular expression, only without the regular expression. It's like the cat without the cat.

And they will be hidden from mere mortals, you'd have to use a big uppercase thing to get it, so you'd know you're doing something weird.

55:14

Currying is idea from Haskell and other languages, only we make it explicit, rather than doing it implicitly. And we say, take a function, and assume that the $x argument is 1, or assume that the $y argument is 2, and that gives us functions with more limited functionality, and hopefully gives the optimizer a chance to optimize some of that.

Lazy lists - in Perl 6, scalars are not lazy, but lists are lazy, by default - you can tweak that. This means that the list in the for loop can be an infinite list, it doesn't have to generate all those integers, it just runs until the loop exits. 0, 1, 2, 3.

You can do fancier thing if you're into PDL-type programming - mapping one thing to another.

Argument lists - because we have argument lists and we have multiple dispatch, we can do a trick... this is pretty much a transcription of something you'll probably see in Audrey's Haskell talk later. This is quick sort, and basically it says: if you pass a null list, return a null list. And if it doesn't mathc that, it matches a list which has a head and a tail. Partition them into a list that's greater than the head and a list that's less than the head, and return the sorted sublist, with the pivot in the middle. And that's all it is.

It's very powerful to be able to do that signature matching. And we are also planning to do this signature matching embedded in our regular expression rule engine, matching against things that are not strings, but look more like argument lists.

57:20

Regular Expressions

Now, the thing I really wanted to get to, which we think will influence the world outside of Perl is we saw how everyone borrowed Perl 5 compatible regular expressions, and we figured - well, you know, they're a really big mess, and we're sorry, but we're changing them now, now that you've just borrowed them.

So everyone is gonna go from PCRE to P6RE or something like that.

There's a lot of cruft that we inherited from the UNIX culture and we added more cruft, and we're cleaning it up. So in Perl 5 we made the mistake of interpreting regular expressions as strings, which means we had to do weird things like back-references are \1 on the left, but they're $1 on the right, even though it means the same thing. In Perl 6, because it's just a language, (an embedded language) $1 is the back-reference. It does not automatically interpolate this $1 from what it was before. You can also get it translated to Euros I guess.

In Perl 5, if you wanted to interpolate a string, but did not want it to be treated as a regular expression, you had to do this weird quotemeta thing. In Perl 6, that is the default, if you interpolate a string it matches it literally, which is why you can match the back-reference $1 literally.

The same is true of arrays. This is an array of literal strings that just match an alternation, or you can look up, by the longest token rule, the longest key in this hash that just matches this bit of string. And that is used heavily in the Perl 6 parser - or will be.

In Perl 5 if you put a bare variable, it treated it as a regular expression. Well that's literal now, so we changed the default around. If you want something treated as a regular expression, as an embedded regular expression, interpolated, then we have this new notation, which can interpolate a single rule or an alternation of rules, or again, look up a token key and then call up to another rule, and this allows us to build recursive grammars.

60:11

In Perl 5, the problem is that it was completely irrational how the meta-characters worked. Parens might or might not capture: that does grouping without capture, that calls a closure, that calls an indirect rule. The square brackets are under utilized. So are the curlies. And the angle brackets were unused so I stole them. They are now meta-characters. We just don't have enough bracketing characters in ASCII. Which is why we stole the French brackets.

In Perl 6 it's very consistent. Parens capture, brackets group, braces call a closure; meta-syntactics always with angle brackets. Use of whitespace is encouraged - in fact, there is no /x modifier anymore. That is now the default, the mandatory default.

61:13

Postfix bottom-up (?) modifiers are bad, because you have to go all the way to the end, and it changes the meaning of what's before, so we pulled them out front. And we have a consistent, what we call an adverbial notation, the colon on the front. It's basically a named argument, that colon (FILL IN) a named argument to the substitution function. These trailing switches are gone.

Things which are modal and change the meaning of ".", "^" and "$" - now they have separate tokens, so if you want to match interior newlines within a string, you double the "^" or the "$". "." now always matches any character, and if you want the old default meaning, you actually say, I want to match "not a newline". "\n" matches a newline so "\N", following the pattern we've already established, matches a character which is not a newline.

62:15

Whitespace is the default. We no longer need a switch that says interpolate the right hand side as an expression, because we can interpolate expressions. As I said, everywhere, curlies mean a closure, even in a string. And we have the new modifiers - some of which are familiar and some of which are not: look for overlapping or exhaust all the possiblities - pretend you're prolog, I guess. Find third match, do the first three matches, do auto-whitespace matching, or don't do auto-whitespace matching. And of course the most important one is: "I don't understand all of this new stuff, just give me a Perl 5 regular expression here." And it will do that too.

The meta-syntax - instead of trying to guess what horizontal whitespace is, you just say "\h" that's horizontal whitespace, including Unicode whitespace. And we have vertical whitespace, which I suppose include vertical tab, whatever that is.

Alpha - if you say "A-Z" and want to match Alphas it's probably wrong, for most of Europe and anywhere that is not the U.S.. Now we tend to use named rules to match character classes, and this tends to free up the square brackets for other uses. So if you actually want to match a character class, you put it in the this angle brackets padded notation, so we sort have de-Huffmanized it. Or you can make it readable.

The negation is doing set addition and subtraction, so this just says the set of all characters with those subtracted out. Or you can say like this - it's a lot more readable. In Perl 5, we had this gobbleygock, you can just see that it's a lot clearer. And again, we got rid of those switches and now you can read the expression left to right and know what it means.

Simplified a lot of the really crufty meta-syntactic things.

The null match is now illegal. That prevents a class of errors where you say "a|b|c" and forgot to put the next thing, and suddenly it matches the null string and you don't know why. Or even worse if the null is on the front. Also there's the situation where you say "//", you don't know in Perl 5 whether it means match the null string, or match the previous match. You have to be explicit about that, and it's readable and prevents a class of errors.

Newline is logical newline, so if you mounted a Windows partition on your UNIX box and say "match a newline" - it figures out the right thing.

65:48

Assertions are now more readable, either positive or negative assertions. Slightly de-Huffmanized, the back-tracking control now can be applied to any preceding token - the same way a quantifier can. The conditional syntax looks a lot less like LISP. Here's how you match a conditional now in Perl 6: basically if it's this then do that, if it's this then do that - that's how you read this. The colons are being a back-tracking control. Only the double one says if you back-track over this back-track out of the entire alternative set.

Now if you wrote that in Perl 5, it would have to look like this, which as you can see is a lot worse. Especially all the trailing parens there. Of course, if you didn't use the /x modifier there to use whitespace, you'd have to write it like this. I'm sure you all agree that's a slight improvement.

Buy you don't even have to keep track of your captures here - you just say: "Give me the previous one". So it's a pattern-action pattern-action and each of them stands independently.

Now, that's not all - we have full grammars. So you want to match a floating point number - here's the grammar for it. Now it looks much more like YaCC or BNF that it does like regular expressions, doesn't it? We don't even call them regular expressions anymore, we call them rules. And they just build a parse tree for you, and using the names of these things you can parse a number. And then you say "give me the sign", "give me the mantissa", "give me the exponent", "give me the sign of the exponent", and it's all nested nicely just as a data structure.

You can even use the constant subscript notation, like that, and because you get tired of seeing those slashes, you can even get rid of them. But we got another problem here which is you might use digits in more than one place, and how you differentiate them. Well you can bind them to different names. And then after you did that, you can just say - well return the integer part of the mantissa and the fractional part of the mantissa.

69:00

Now, if you say evaluate something, that's really going to do something underneath like take the Perl 6 grammar as the top rule, and then parse this thing, and compile that, and run it.

And a grammar is just a strange kind of class, and these rules are just funny kind of methods, and since if you can overwrite methods, why not overwrite rules. So let's take the Perl grammar, and derive a grammar from it, and we'll call it "Rubbish"... I mean "Rubyish". I will redefine identifiers to allow an optional "?" or "!" on the end. And off we go. And install it as your current grammar using a "use" statement or something, and suddenly you're programming in Rubyish.

And this is how natural languages evolve, they just make little changes to the grammar and the dictionary all the time. It's also how people learn language. You learn the simple thing and then you learn other things in terms of those things.

It's also how Audrey is defining Pugs. He takes this syntactic sugar, and de-sugars it into this is defined in terms of that, and this is defined in terms of that, until at the low level there's just a small number of primitives.

Now, we had some of this "an A is just a B" in Perl 5, we could say, "an object is just a blessed reference", "a class is just a package", "a method is just a subroutine". Sometimes we got confused because we used the same keywords for all of these things.

The same things carries over to Perl 6. Though in some cases, we changed the keyword, but it's still the same thing semantically underneath. So we can still do a lot of this define one thing in terms of another.

An object attribute is just a weirdly scoped variable. A class attribute - well maybe that's not true anymore. A grammar is kinda like a class, a grammar rule is like a method... types are properties that are interpreted as contracts, attributes just bind a list, hyper-operators - you can define all of these things just in terms of lower level things.

A control structure is just a subroutine that can call closures - that's an important one.

All control flow to exit a block is just a strange kind of exception, and you know Perl 6 is just a new version of Perl.

And that's the talk.

I guess I have gone into your break time by 5 minutes so I don't have time for questions, but I'll be here all the rest of the time and feel free to ask me anything. [Applause]

END

 :

72:39