Pytek

Aqua Tart

00:00

Ooh, we have to change print to os.write(1,...

Everything else should be OK though.

El'endia Starman

...that's all?

Aqua Tart

Well, there's other things.

But I think that's the one thing that'll affect us.

El'endia Starman

uh-huh

Well, maybe when we get to writing out all the functions there'll be more to consider...

By the way, idea I had in the shower: type: (or spec:) for denoting something as a type specifier. Use like type:safe(...).

Traceback (most recent call last):
  File "/home/ubuntu/workspace/src/pytek_parser.py", line 293, in <module>
    T.run()
  File "/home/ubuntu/workspace/src/pytek_parser.py", line 31, in run
    item.execute()
  File "/home/ubuntu/workspace/src/pytek_parser.py", line 199, in execute
    return self.func(self.opts,self.args,self.code)
  File "/home/ubuntu/workspace/src/pytek_functions.py", line 5, in pPrint
    if type(arg).name == "string": os.write(1,arg.val)
TypeError: 'str' does not support the buffer interface

...uhhhh...

Aqua Tart

? that should be right

El'endia Starman

I tried running it in C9.

5

A: 'str' does not support the buffer interface Python3 from Python2

In Python 2, bare literal strings (e.g. 'string') are bytes, whereas in Python 3 they are unicode. This means if you want literal strings to be treated as bytes in Python 3, you always have to explicitly mark them as such. So, for instance, the first few lines of the encoding function should loo...

Aqua Tart

00:06

Ohhh

El'endia Starman

Alright, fixed that.

Aqua Tart

Somewhat annoyingly I can't find where the translator toolchain is

They've moved it

El'endia Starman

:/

I assume you're already Googling? :P

Aqua Tart

Yes

Oh, found it.

Dammit we have a problem

We can't use Python 3 for translating

It has to be Python 2

elendiastarman:~/workspace/bin $ python3 ../pypy/rpython/bin/rpython ../src/pytek_main.py
  File "../pypy/rpython/bin/rpython", line 17
    print __doc__
                ^
SyntaxError: Missing parentheses in call to 'print'

elendiastarman:~/workspace/bin $ python ../pypy/rpython/bin/rpython ../src/pytek_main.py
[translation:info] 2.7.6 (default, Jun 22 2015, 17:58:13)
[translation:info] [GCC 4.8.2]
[platform:msg] Set platform with 'host' cc=None, using cc='gcc', version='Unknown'
[translation:info] Translating target as defined by ../src/pytek_main
Traceback (most recent call last):
  File "../pypy/rpython/bin/rpython", line 20, in <module>
    main()
  File "/home/ubuntu/workspace/pypy/rpython/translator/goal/translate.py", line 217, in main

El'endia Starman

......nuuuuuuuuuuuuuu..........

Aqua Tart

00:14

:/

Well on the bright side translating Pytek to 2 won't be nearly as painful as translating PyPy to 3 :PP

El'endia Starman

Oh wait, there's such a thing as PyPy3...

Aqua Tart

!

Oh, didn't know they had ported it

OK, I'll pull that right away :P

El'endia Starman

:D

Aqua Tart

k looks like the branch is py3.3

cloning

Err... I think this might be the wrong branch??

ER yup

Oh, maybe it's py3k

Yeah, think so.

Mercurial is so slow

it's not even funny

5-10 minutes to check out a branch jeez

Dammit, this still isn't Python 3.

What is going on

OK screw this I'll just grab a source tarball

OK ragequit

This is for sure PyPy3 but the translator shell is throwing an error

Absurd.

@El'endiaStarman OK, we're going to have to just port it to Python 2. The translation tool chain doesn't appear to support Python 3 yet

Er... that shouldn't be too problematic right?

We don't have much code yet

Actually we can just 3to2 it

@El'endiaStarman You still there?

we needz you

PhiNotPi

00:57

How much progress has been made already? (You should add a link or something to the room description)

Aqua Tart

@PhiNotPi We don't have a GitHub yet.

PhiNotPi

Do you have any example code?

Aqua Tart

Tons.

One second.

func:isPrime(n) {
  return ([1..n]~\product,n->\mod) == n - 1
}

\print("Hello, World!")

Unfortunately we haven't quite decided whether for will take an anonymous function or a code block yet :P

Anyways, the spec is (almost) done and the parser is working but only supports strings, functions and natural numbers

So Hello World works but nothing else

PhiNotPi

> [1..n]~\product,n->\mod

Aqua Tart

Yeah.

It's called a chain.

PhiNotPi

01:08

That took me a while to understand.

Aqua Tart

It's the same as \mod(\product(1..n),n)

But it looks cleaner once you understand how it works.

Also this is a thing:

{x | 0 < x <= 3 and x in \Z}

Set-builder notation.

The current planned types are: sets dictionaries lists strings tuples chains lambdas numbers and complex numbers

Here's quickref.md:

##Sequences

Sets: {1,2,3} or {x | 0 < x <= 3 and x in \Z}
Alternative: `{x | x in 1..3}`

Strings: "123", '123'

Lists: [1,2,3]
Another way: `\first[3](\N)`

Tuples: (1,2,3)

Dictionaries: {"foo":1,"bar":2,"baz":3}

##Functions

Chains: func1,func2,...->func3,func4,...->...

(Usage: args~chain)

Anonymous Functions: <args:code>

(Usage: (<args:code>)[args...])

Named Functions: func:name[options](args) {code}

(Usage: \name[options](args))

##Control Flow

<> denote anonymous functions.

\if(condition,<body>)

PhiNotPi

What is the difference between a list and a tuple?

Lists are of the same type with variable size, while tuples can be of many types but are of fixed size?

That's how I think I've seen it done before.

Aqua Tart

Lists are mutable tuples are immutable

Just like Python.

PhiNotPi

I guess I'll have to brush up on some Python 101.

Aqua Tart

Not really, Pytek won't be all that similar to Python.

PhiNotPi

01:15

One thing I've kinda wanted to do is make an actual non-golfing language.

But looking at the stuff so far, it is definitely going is a completely different direction than how I would probably take it.

El'endia Starman

01:48

@AquaTart Sorry, was chatting with parents.

Aqua Tart

Anyways the translation toolchain doesn't seem to work with Python 3 for whatever reason

I tried PyPy3 and it still didn't work

El'endia Starman

That's a bummer.

Aqua Tart

Yeah. :/

How painful will porting be?

There is a 3to2 tool.

El'endia Starman

Personally, I'm thinking about not bothering with PyPy.

Aqua Tart

Really?

El'endia Starman

01:51

@PhiNotPi Perhaps you can still benefit from the parser.

Aqua Tart

Er... is not being able to use Python 3 that big a deal?

El'endia Starman

I haven't used Python 2 in years.

Aqua Tart

I personally think that we need RPython if we want the language to be usable speed wise

El'endia Starman

hmm

Aqua Tart

We can always port what we have now and do a benchmark.

El'endia Starman

01:54

I'm not against porting it to Python 2, I just don't want to do that myself.

Speed is not a big deal for me, in large part because I know computers will continue to get faster.

And if we really cared about speed, we would be writing this in C, and I am not doing that. :P

Aqua Tart

hahaha

True enough

OK, so forget about PyPy.

I think having a way to compile it into an executable of some sort would be nice though.

cx_freeze is still a thing right?

You know, just to have it nice and bundled.

El'endia Starman

I don't know what cx_freeze is.

Aqua Tart

It (used to be) a popular module for taking a bunch of Python files and compiling them into a standalone executable.

OK, it's still a thing. Good.

El'endia Starman

ahh, okay

Hmm.

Aqua Tart

That's what we were trying to do :P

ETHproductions

02:16

So, what's happened with Pytek recently?

El'endia Starman

Quartata - excuse me, @AquaTart - tried to use PyPy to make the Pytek interpreter run faster, but it apparently doesn't support Python 3 (at least, not easily).

I'm not sure if I got that right.

ETHproductions

Oh, bummer

Do you have an official home for the spec somewhere?

PhiNotPi

How much of the syntax has been decided upon? Like, I know you haven't decided how for loops would work, but besides that.

El'endia Starman

@ETHproductions We have one on Cloud 9. Ish. Scroll up a couple pages to see the current quickref.md.

@PhiNotPi Most of it, I think.

PhiNotPi

Any other documentation I'm not aware of?

El'endia Starman

02:27

Not really. It's basically all in the chat transcript.

I guess I can work on getting a spec up tonight.

RikerW

02:46

Thanks.

El'endia Starman

No problem. ..... What are you saying Thanks for? :P

PhiNotPi

03:17

My idea for a language feature: do{}while(){} the first {} is performed before the condition is evaluated, but each iteration travels across both the first and last {}, so you can have do{take input}while(input is invalid){print "ur wrong, try again"}

^ probably doesn't fit in with Pytek at all

El'endia Starman

What does fit for sure is wanting to reduce programmer work so far as bookkeeping goes.

That particular feature might not make it in that exact notation, but I'd like to include it somehow.

PhiNotPi

Maybe you could even have do{}while(){}while(){} (or alternatively do{}while()do{}while()do{} for maybe some added consistency)

El'endia Starman

That looks like it'd be a pain to read.

PhiNotPi

probably

El'endia Starman

Naw, I'm thinking of somehow using \func[opts](args){code}.

PhiNotPi

03:26

Actually, I think my first example is equivalent to for(take input; input is valid; take input){print "ur wrong, try again"}

El'endia Starman

@PhiNotPi Mind elaborating on exactly why this feature would be useful? I can think of a couple scenarios, but I'm not sure what you have in mind.

PhiNotPi

Ehh, I don't really have any strong justification, other than a literal "try to take input from a user" example.

El'endia Starman

yeah

I know there have been situations where I want to read input in the conditional so I can leave the loop when it's empty, but that means I don't get to use it in the body.

PhiNotPi

Would that be fixed by simply allowing assignment in the conditional?

... if I understand the issue correctly

El'endia Starman

I could make assignment operators "return" the value that was assigned.

PhiNotPi

03:31

That could make sense for "chained" assignment.

El'endia Starman

while input():
    print(input())

PhiNotPi

a = b = c could be interpreted as a = (b = c) with the assignment returning the value.

El'endia Starman

possible

PhiNotPi

An alternative is to special-case stuff, like Perl's beautifully hacky while(<>){}

El'endia Starman

That alternative doesn't sound particularly appealing to me. :P

El'endia Starman

03:48

Crazy notation idea: 1 [] 5 for [1,2,3,4,5], 1 [) 5 for [1,2,3,4], 1 (] 5 for [2,3,4,5], 1 () 5 for [2,3,4].

PhiNotPi

How do you tell the difference between [] and an empty list?

El'endia Starman

Context, maybe?

Aqua Tart

Hmm we will have to figure out how to parse binary operators

El'endia Starman

I think I'll be implementing the shunting yard algorithm, in essence.

PhiNotPi

I feel like it's easy to run out of symbols to use.

El'endia Starman

03:53

yeah

Aqua Tart

@El'endiaStarman Wouldn't it just be isPossible if two approriate nodes surrounding it and isComplete if it matches <any binary operator> followed by something?

El'endia Starman

@AquaTart Think about 2 + 5 * 3 - 4. Also, I don't have a provision for isPossible or isComplete to look backwards.

Aqua Tart

What about isPossible if matches any binary operator followed by something and always assumed to be complete?

El'endia Starman

hmm

Aqua Tart

You couldn't have = and == though hrmph

Colon for assignment? Or too weird?

El'endia Starman

04:00

I always disliked the extra shift+key for that kind of thing.

Aqua Tart

Actually wait you could have them both

But == would have to be checked first

OK, I think that works.

I'm away from C9 but I'll take a crack at that later perhaps

El'endia Starman

I'm gonna put another dict key for shunting infix operators in there.

PhiNotPi

Horrible idea, inspired by the "use colons for assignment": Since we already have {"a":val1, "b":val2, "c":val3} for dicts, but there's also the idea {a=val;b=val;c=val} where the {} is a code block, why can't the two be the same?

El'endia Starman

hmmmm

It may not be good to have two completely distinct ways of notating the same thing.

PhiNotPi

I'm saying, have them be the same, and notate them to be the same.

As in, the "keys" to a dict are like the local variables of a code block.

You could do something like data = {a=7; b=2; c=b+a} and then data is a dict where data{c} == 7

El'endia Starman

04:17

....that could be really interesting and/or backfire horribly, and I'm not sure which it'll be. :P

PhiNotPi

In Perl, objects are dicts (well, generally speaking).

So, if you really wanted to, you could turn this into a way to do OOP.

El'endia Starman

That's pretty much what they're like in JS too.

PhiNotPi

dict data = {a = 7; func:inc(){a = a +1}};

data{a} returns 7, data{inc()} modifies the object, data{a} returns 8.

except, now we have a weird nested {a=a+1} thing.

El'endia Starman

Yeah. I'll think on that.

PhiNotPi

Which, I suppose by extension of the current theme, is itself a dict with no keys (since the a isn't a local variable at that level), which is only evaluated once the function inc is called.

It's getting kinda weird at this point.

I need to catch some sleep.

IDK where I was going with the above suggestion.

El'endia Starman

04:32

haha

@Mego: want an invitation to the Cloud9 VM we're developing on?

Anonymous

@El'endiaStarman Sure

Anonymous

But if I see you've made folders for version control like VTC I'm gonna rm -rf --no-preserve-root / it

El'endia Starman

hahaha

I'll actually use GitHub releases this time. :P

Anonymous

Good

El'endia Starman

I do plan on hosting an online interface on my website like I did with Minkolang, and putting the source code up on GitHub.

Okay, you have a C9 account, right? How should I invite you?

(By which I mean, what's your username?)

Anonymous

04:45

Just mego

El'endia Starman

Alright, access granted.

Feel free to look around.

Anonymous

Neato

Anonymous

I'll do that in a little bit

Anonymous

Playing minecraft right now

El'endia Starman

alrighty

El'endia Starman

06:37

Justification for not letting anyone else work on the parser! :P

Anonymous

El'endia Starman

Pretty much, yeah. :P

El'endia Starman

07:41

define function eval-polynomial(x, coefs)
  let sum = 0;
  for (i from 0, c in coefs)
    sum := sum + c * x ^ i;
  end;
  sum;
end;

I really like this idiom of i from 0.

Required reading: Design Patterns. A bunch of useful ideas in here.

El'endia Starman

08:08

> As an illustration of what I mean about the relative power of programming languages, consider the following problem. We want to write a function that generates accumulators-- a function that takes a number n, and returns a function that takes another number i and returns n incremented by i.

func:foo(n){ \return func:bar(i){ n += i }}

El'endia Starman

08:48

Idea: figure out a way to let users define syntax for a new object.

Say, @polynomial[ <isPossible>, <isComplete> ]{<parsing code>}.

I have no idea how I'd make this work, but wouldn't it be cool to define the syntax for a polynomial and then just use that later in the program and it just works?

There are far too few delimiter symbols... :(

If we want to stick to ASCII, that is. Which I do.

Aqua Tart

14:45

@El'endiaStarman OK I had a good idea last night about control flow syntax

Basically we make {} the delimiters for anonymous functions

And we make the preprocessor automatically replace {} with ({})

And we use [] args for the conditions and what not

That way \while[True]{ ... } is really \while[True({ ...})

I know this sounds weird but I've thought it through and I cannot think of a single situation where this replacement would break anything

oops I meant \while[True]({...})

too late to edit

Nathan Merrill

15:11

@El'endiaStarman syntax for an object leads to an explosion of messiness. However, it you allow objects to define certain operators (and give no intrinsic meaning to those operators), then I think that approaches what you want

also, if you require all functions to start with a backslash, and you allow first class functions, then you must have static typing

or, at least, typing between functions and other variables

Aqua Tart

The syntax he proposed above appears to allow direct modification of the parser

Nathan Merrill

right

and I think that leads to an explosion of messiness

Aqua Tart

I find that intriguing but a little too complex for pure objects, yea.

But perhaps for defining shorthand ways of creating certain objects...

Nathan Merrill

also, one of the greatest strengths of Python is that it doesn't include certain common idioms in programming

such as do-while, switch statements, so on

having a language that does everything and anything isn't a powerful language, because it makes it difficult to build on top of preexisting code

PhiNotPi

15:27

Is this currently a c9 project? If so, my username is phinotpi

Nathan Merrill

I think your inital statement says it well: "identifying and leveraging common patterns". a do-while (for example) isn't a common pattern, and support for such an idiom will hold back your language IMO

Aqua Tart

16:02

@PhiNotPi yes

Aqua Tart

18:07

FINALLY got cxFreeze to install

Some stupid bug with Ubuntu...

oops I installed the py 2.7 version

oh wait I'm stupid

called cxFreeze wrong

Aqua Tart

18:35

OK, got it working.

Aqua Tart

18:52

The cx_freeze exe is much faster.

So that's nice.

El'endia Starman

19:36

@AquaTart I think this is essentially unnecessary, because you can just write the pLoop function to use what's inside the {}, which will be passed in to the code parameter.

@PhiNotPi I wouldn't mind inviting you, but considering that I want to avoid bloat and such, I'm just wondering how you plan on contributing.

PhiNotPi

@El'endiaStarman Oh, I wasn't really planning on changing anything.

Found it: ide.c9.io/elendiastarman/pytek

El'endia Starman

yup, that's it

Aqua Tart

20:54

@El'endiaStarman But this way we don't have to write special code to parse it

It can just be a normal function

Anyways I made a small build script for cxfreeze

El'endia Starman

What I'm say is that there is no extra parsing work. It's all in pLoop in pytek_functions. (At least, I think that'd work.)

Aqua Tart

Ohhh.

Oh, I see how you did that.

That's nice.

@El'endiaStarman I added a little factorial function so we can test numbers and performance

El'endia Starman

okay okay I'm on the VM now

Aqua Tart

oops syntax error

I can't Python for beans today apparently

El'endia Starman

lol

Aqua Tart

21:07

Hmm although

Pytek's print doesn't do numbers yet

We should just screw type checking and make it alias directly to print

El'endia Starman

I noticed that yesterday and just realized why.

Well, but I'm not doing that because I want to make print more powerful/useful. Like centering text.

Aqua Tart

Oh. OK.

We should add a \str then.

(for now)

Anonymous

/me watches the language discussion going by, not following most of it

El'endia Starman

Have you had a chance to look around the code yet?

Anonymous

Not yet

El'endia Starman

21:11

I can explain its structure and answer any questions.

Aqua Tart

@El'endiaStarman it no work

outputs nothin

Oh I know why

El'endia Starman

But it has the name "number".

Aqua Tart

Oh.

So it should be number?

El'endia Starman

gimme a moment

Aqua Tart

k

Just so you know, you can do builds with bash build.sh

I presume you noticed that already

Anonymous

21:15

Why not make the building process a run conf?

El'endia Starman

YUS

Okay, I really should write an AST_tree_print function.

Aqua Tart

21:31

@Mego so that when we put it on le github other people can build

Anonymous

No I mean keep the build script, but add a run conf to build it for ezpzness

Aqua Tart

eh

not that much effort to cd out and bash build.sh

elendiastarman:~/workspace/bin $ time ./pytek ../tests/test3.pytek
3628800

real    0m0.045s
user    0m0.036s
sys     0m0.004s

El'endia Starman

cool

Aqua Tart

hmm running with python is one millisecond faster

Hmm this bothers me

The executable version always seems to be slower

And I think I know why

Anonymous

What about PyPy?

Aqua Tart

21:43

We'd have to use Python 2

Translation toolchain doesn't work for 3

El'endia Starman

Well, we can run Python3 code with PyPy3.

El'endia Starman

21:56

Really annoying fact: we can't use AST_node constructors in pytek_functions because they're not defined yet.

Aqua Tart

:/

El'endia Starman

Solution: shunt all the AST_node stuff into its own file.

Aqua Tart

It would be nice if we could just deal purely in either AST nodes or Python types

Unwrapping and wrapping just screams bad performance.

El'endia Starman

I'd rather stick with Pytek objects because that way, all of the capabilities are present.

Aqua Tart

OK.

So we need to make these functions return AST nodes

El'endia Starman

21:58

Hmm. Consider closures. How would you make those work solely in Python types?

Aqua Tart

Yeah.

If we make the functions return AST nodes we won't have to do all these shenanigans

but we can't construct them yet :/

Anonymous

@El'endiaStarman Decorators :P

Anonymous

lambda a: lambda b: a+b is a simple closure in Python. Its type is just function.

Aqua Tart

Not the point

The point is that it is inconsistent.

We'd have to type check for AST nodes and Python types.

No, we need to make all functions return AST nodes.

Orrrrr make it so that when we call a function with an AST node we just give the value

Two options.

The former is better obviously

But the latter is easier

El'endia Starman

floopin circularity problems...

Aqua Tart

22:05

Did you make the AST_node constructor?

El'endia Starman

AST_function needs functions, which needs to come after pFac, which needs to come after AST_node...

Aqua Tart

oh lmao

Not usually a problem in Java :P

El'endia Starman

Oh hey, worked it out.

Aqua Tart

@El'endiaStarman By the way, I hate to say this...

But I think you should give some serious thought to PyPy again

El'endia Starman

You mean porting to Python 2?

Aqua Tart

22:10

I know Python 2 isn't your forte but it isn't too big of a difference (and I know Python 2 much better than Python 3)

I'll use 3to2 to port it later and show you the benchmarks

But I'm sure it'll be faster.

I'm worried about the times we're getting right now.

El'endia Starman

I think I've said before that I don't mind if you do the work.

Aqua Tart

OK.

Anonymous

Python 2 is actually super easy coming from Python 3 in everything except Unicode

Aqua Tart

When we're done fixing this I'll make a copy, 3to2 it and see what happens

El'endia Starman

Besides, if PyPy has been used for several other languages, then why not Python 3? That just doesn't make sense.

Anonymous

22:12

@El'endiaStarman Implementation takes time

Aqua Tart

PyPy3 is a thing

Just not the toolchain

I think they wrote PyPy3 in Python 2

El'endia Starman

hmm, okay

Anonymous

Main differences: everything returns a list unless it's in itertools, print and exec are statements, / is integer division with int arguments, backticks are repr, and longs are annoying in golf.

El'endia Starman

Oh, by the way, idea I had last night in bed: foo.{x,y,z} as a shortcut for [foo.x, foo.y, foo.z].

Anonymous

Wouldn't foo.[x,y,z] make more sense?

El'endia Starman

22:15

That'd work too, yeah.

Aqua Tart

@Mego I hate to say this but I like all of those differences :P

Having to do list( all the time kinda annoys me

Anonymous

I like Python 2 for golf, and Python 3 for real development

Anonymous

The nice thing is, a lot of the time, you don't need to cast to a list

Anonymous

Only really for printing

Aqua Tart

I know that's what they say but I always find myself casting anyways :/

Anonymous

22:17

Memory efficiency dawg

Aqua Tart

As in it always throws errors without it...

@Mego You should really join the C9 and watch us do stuff

@El'endiaStarman OK I put in an isNode call to factorial so that should work now

El'endia Starman

Yeah, it works because I moved the isNode function definition.

Aqua Tart

Yep, all good.

Passes test #4

But we will need to do something about that

This is just temporary

El'endia Starman

pFac still only returns a Python number though.

Aqua Tart

Right.

But we don't have an AST_node constructor I thought

El'endia Starman

22:25

I don't know how to AST_number it, and now that I think about it, that'd be a bit hard to do since AST_nodes expect lists. Hmm.

grrrrrr dependency issues...

Aqua Tart

? wat happened

El'endia Starman

Just in reference to the difficulty of using node constructors.

I think I have a solution, though it's a bit hacky...

Wait, is it possible to do circular imports?

Aqua Tart

In Python? Don't think so.

Should import only once.

El'endia Starman

yeah

Wait, we DO already have a circular import and it didn't barf!

Aqua Tart

Yeah.

I don't think it barfs...

El'endia Starman

22:30

It doesn't because pytek_objects is only imported with import ....

Nathan Merrill

22:41

@El'endiaStarman do you have a formal definition for your syntax?

El'endia Starman

Uh, read the transcript? :P

(No, I don't, not yet.)

Nathan Merrill

sorry, I read back a while, but not all of it

El'endia Starman

Yeah, there is a lot.

One of these days I'll get around to putting up a subdomain for Pytek, which will have the spec up there.

Nathan Merrill

also: jetbrains.com/mps

El'endia Starman

huh

El'endia Starman

23:21

Alright. I'm off to youth group, and I'll be back within 3 hours.

Transcript for