Thanks for reminding me. Although I wonder if there is a difference between drawing metaphors (monads are burritos etc.) and drawing parallels. For example in the beginning it was useful for me to think of monads (and typeclasses really) as approximating Java interfaces. Now this parallel dies pretty quickly but it was something for me to hold onto while internalizing the concepts. Is there a danger in telling people this? -deech On Tue, Aug 3, 2010 at 5:36 PM, Dan Piponi wrote:

On Tue, Aug 3, 2010 at 2:51 PM, aditya siram wrote:

...

I am doing an "Intro To Monads" talk in September [1]. ...what would you stay away from.

Some things to stay away from:

http://byorgey.wordpress.com/2009/01/12/abstraction-intuition-and-the-monad-...

(Though I secretly harbour a belief that there is a correct metaphor that everyone should be using. :-) -- Dan

Hi, aditya siram wrote:

For example in the beginning it was useful for me to think of monads (and typeclasses really) as approximating Java interfaces.

Type classes are somewhat parallel to Java interfaces, but Monad is a *specific* type class, so it should be somewhat parallel to a *specific* Java interface, if at all. Type classes are somewhat parallel to Java interfaces because a Java interface interface Foo { Result method (Argument argument); } declares that there is a set of types so that every type T in that set has an operation (T, Argument) -> Result, with these operations all implemented specifically to particular type T. In Haskell, the type class class Foo t where method : t -> Argument -> Result expresses a similar concept. There are a number of differences though: Firstly, in Java, calls to the method are late bound, while in Haskell, they are early bound. However, a kind of late bound behavior can be achieved using existentials. Secondly, in Java, the receiver of the method has to be of type T, and T may not appear at other positions in the type of the method, while in Haskell, T may appear anywhere in the type of the method, even more then once. Finally, in Java, T has to be a proper type (of kind *), while in Haskell, it may be an improper type (of a kind involving ->). Already for the type class Functor, these differences become relevant. class Functor f where fmap :: (a -> b) -> f a -> f b f has kind (* -> *), and it is mentioned twice in the type of fmap. Conclusion: While Haskell type classes have some similarities to Java interfaces, the type class Functor (or Monad, if you like) is not that similar to any Java interface, because it uses features specific to Haskell type classes which are not available in Java interfaces. Nevertheless, it may be helpful for a Java developer to understand that Haskell type classes are more similar to Java interfaces than to Java classes. Tillmann

Hi, Please take a look at this video http://videoarch1.s-inf.de/FP.2005-SS-Giesl.(COt).HD_Videoaufzeichnung/2005-... Here Monad's are explained as "something" that helps making your program modular. The teacher gives an example implementation of an expression evaluator with and without monads. It takes a complete rewrite to incorporate changes in the program without monads where as only minor tweaks are required for the implementation with monads - also, its easier to identify the location where change needs to be done and the change is isolated. And the flow is pretty nice - as in, people will not doze off :) Regards, Kashyap On Tue, Aug 3, 2010 at 10:04 PM, Colin Paul Adams wrote:

...

...

...

...

...

"Alexander" == Alexander Solla writes:

Alexander> On Aug 3, 2010, at 2:51 PM, aditya siram wrote:

...

I am looking for suggestions on how to introduce the concept and its

...

implications. I'd also like to include a section on why monads exist and why we don't really see them outside of Haskell.

Alexander> Start with functors (things that attach Alexander> values/functions/functors to values in an algebra). Move Alexander> on to applicative functors (functors that can interpret Alexander> the thing that is getting things attached to it). Move Alexander> on to monads

Too late! The audience has already dozed off.

Alexander> (applicative functors where you can Alexander> explicitly control the order of Alexander> evaluation/interpretation).

-- Colin Adams Preston Lancashire () ascii ribbon campaign - against html e-mail /\ www.asciiribbon.org - against proprietary attachments _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

-- Regards, Kashyap

On Fri, Aug 06, 2010 at 10:17:26AM -0500, aditya siram wrote:

...

From my vantage point they are (in no particular order) : Reader, Writer, State, IO, ST, STM, Parsec (have I missed any?) and of course the transformer versions. I am debating whether or not to add [] to the bunch.

Not sure how much time you have budgeted, but I'd start with a simple one like Maybe, actually show how to implement it, then move on to list, and then finally talk about Reader, Writer, State, etc. from a more high level perspective. Hopefully people would grok the monad concept by that point, and should realize how useful it is to add logging, or state, etc. Then they may wonder about mixing in more than 1, so that could lead to transformers.

To explain monads (now that I have Timothy's awesome blog post to reference) I'll be drawing the parallel between monads and interfaces in Java. And thanks to Tillman for showing me where the analogy breaks down. Are there any such parallels in other languages like Perl and Python?

I get the type class / Java interface analogy, but trying to draw a parallel between Java interface and monads seems likely to just create confusion IMHO.

I'm still a little iffy on why the monad concept isn't used in other languages. From where I sit it seems as though monads really let you play with the order of evaluation - just because one statement is executed after another doesn't mean they are executed in that order. I think other languages don't make this easy.

I first encountered monads in OCaml. And the concept exists in other languages, although maybe not always explicitly by that name. Good luck, should be a good talk, Alex

At version 2 Parsec was an amalgamation of a state and error monad - by amalgamation I mean the data types and Monad instance encoded the combination directly, it wasn't made from transformers. Version 3 of Parsec complicates things a quite a bit. If you're addressing Perl programmers, you could try and encode the monads in Perl and see what it looks like. Maybe Mark Jason Dominus the author of Higher Order Perl has done this already? On 6 August 2010 16:17, aditya siram wrote:

[SNIP] I think I want to dive right into the most common and useful monads. From my vantage point they are (in no particular order) : Reader, Writer, State, IO, ST, STM, Parsec (have I missed any?) and of course the transformer versions. I am debating whether or not to add [] to the bunch.

On Fri, Aug 6, 2010 at 11:17 AM, aditya siram wrote:

I'm still a little iffy on why the monad concept isn't used in other languages.

The greatest feat that monads have accomplished, in my opinion, is providing the right mathematical abstraction for declaring side-effect and stateful computation in a pure functional language. For the first time it was now possible for a pure functional programming language to be a general purpose programming language. That's was quite an accomplishment! The second most important feat of monads was their close offspring, the monad transformers. These guys gave pure functional programmers the ability to maintain invariants on which side-effect/stateful computations their end users are going to use. This enhanced verifiability by a measured mix of pure and impure. The other languages that you mention are impure. So feat #1 is pointless since they were already general purpose without monads, and feat #2 is not realistically achievable since impure languages generally don't have a way of restricting a function from having arbitrary side-effects. There have been some clever things done with monads aside from #1 and #2. Parsec is one, but it seems applicative functors are a better match for the parsing domain. Other things are "neat", but not killer features that would turn the head of a pragmatist of the impure type in my opinion. So, that is why I think the monad concept isn't generally used in other languages. David -- David Sankel Sankel Software www.sankelsoftware.com 585 617 4748 (Office)

It might be a little late at this point, but here's my take on monads: In most imperative languages sequencing of statements is a feature that is hard-coded into the language to act in a certain way, e.g. to have a particular implicit state (the global state plus possibly the fields available from a class, if the code is in a method). In Haskell, sequencing of statements is a first-class feature of the language that we can define to work however we want. For example, the Maybe monad allows us to define the sequencing operation to abort whenever we access a Maybe value that is Nothing. In most languages they have to introduce a special keyword "return" to get this kind of early-exit functionality, but in Haskell we don't need such a keyword because Monads allow us to extend the sequencing operation to *add* this kind of functionality. Similarly, in most languages you cannot completely change the implicit state available to code; the most that you can do is to use "Object-Oriented programming", which is an additional feature to the language, to add additional implicit state that is available to code (when it is inside a method) that stacks on top of the global state. By contrast, in Haskell defining an implicit state for code is trivial, and furthermore we can do additional things like forcing this state to be read-only, all without having to add new features to the language itself. Not only is sequencing is a first-class operation that we can define at will, but it is also possible to compose the functionality of multiple sequencing operations so that, for example, we can get access to both a read-only state, a global mutable state, *and* have the ability to perform an early exit from our code, all (again) within the language. This technique is known as stacking "Monad transformers", and there are multiple libraries for doing it. Finally, here is something that is trivial to do in Haskell: create a sequencing operation based on continuations that allows code to perform operations asynchronously while writing code in a synchronous style. That is, you can define a monad that lets you write code that looks like do result <- request x case result of A -> request y B -> request z which has the feature that rather than blocking a thread while we wait for a request, we instead actually implicitly creates a callback that runs the rest of the code as soon as the result as ready. In most other languages you'd have to invent a special language features such as continuations in order to allow for a coding style like this, but in Haskell we don't need them because *overriding the sequence operation itself* is a first-class feature of the language that allows us to do this. So again, the moral of this story is that just like having access to continuations as a first-class object allows one to do powerful things, so too does having access to the sequencing operation as a first-class object let us do stuff that makes our job as programmers easier. Cheers, Greg On 08/06/10 08:17, aditya siram wrote:

Thanks all for you suggestions! Upon further reflection I realized that my audience is more pragmatic than theoretical. Instead of emphasizing how monads are constructed and the monad laws I think I want to dive right into the most common and useful monads. From my vantage point they are (in no particular order) : Reader, Writer, State, IO, ST, STM, Parsec (have I missed any?) and of course the transformer versions. I am debating whether or not to add [] to the bunch.

To explain monads (now that I have Timothy's awesome blog post to reference) I'll be drawing the parallel between monads and interfaces in Java. And thanks to Tillman for showing me where the analogy breaks down. Are there any such parallels in other languages like Perl and Python?

I'm still a little iffy on why the monad concept isn't used in other languages. From where I sit it seems as though monads really let you play with the order of evaluation - just because one statement is executed after another doesn't mean they are executed in that order. I think other languages don't make this easy.

-deech

On Wed, Aug 4, 2010 at 6:21 PM, Daniel van den Eijkel mailto:dvde@gmx.net> wrote:

For me, the following two things did the magic, so I'll suggest them:

1. Writing a recursive function that takes a binary tree and returns the same tree, but with its leaves enumerated. Each function call takes the tree and the counter and returns the resulting tree and the new counter value. The pattern that emerges is similar to the state monad. The pattern shows that the order of the recursive calls is not ambiguous, unlike in a function that just counts the leaves, for example. Changing the order of the recursive calls changes the result. (code below)

2. Putting the above pattern into a datatype and rewriting the apply-funtion for this datatype (>>=) allows only to apply funtions in a non-ambiguous way. Not giving a deconstructor for the IO monad forces the programmer to decide in which order calls to IO functions have to be done.

I hope this is clear enough; I was able to use the IO monad at the moment I realized that Haskell uses this kind of "trick" to ensure that the order of execution of function arguments is always well-defined and never ambiguous. Of course, there is much more about monads, but this was my entry point.

Best regards Daniel

code (tree enumeration):

data Tree a = Leaf a | Node (Tree a) (Tree a) deriving Show

enumTree n (Node a b) = let (n', a') = enumTree n a in let (n'', b') = enumTree n' b in (n'', Node a' b')

enumTree n (Leaf x) = (n+1, Leaf n)

aditya siram schrieb:

...

Hi all, I am doing an "Intro To Monads" talk in September [1]. The audience consists of experienced non-Haskell developers but they will be familiar with basic functional concepts (closures, first-class functions etc.).

I am looking for suggestions on how to introduce the concept and its implications. I'd also like to include a section on why monads exist and why we don't really see them outside of Haskell.

Has anyone here done a talk like this? And if so what parts of your presentation were successful and what would you stay away from.

Thanks for the feedback. -deech

[1] It's in St.Louis, Missouri at the St.Louis Perl Mongers meeting http://St.Louis%20Perl%20Mongers%20meeting so come on by if you're around! ------------------------------------------------------------------------ _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org mailto:Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Yes I think that showing the Maybe and List implementation of monads is essential. They're practical and in a lot of ways they represent two completely different types of computation demonstrating the flexibility of the Monad abstraction. Thanks for that suggestion. -deech On Mon, Aug 9, 2010 at 4:33 AM, wren ng thornton wrote:

aditya siram wrote:

...

Thanks all for you suggestions! Upon further reflection I realized that my audience is more pragmatic than theoretical. Instead of emphasizing how monads are constructed and the monad laws I think I want to dive right into the most common and useful monads.

...

From my vantage point they are (in no particular order) : Reader, Writer,

State, IO, ST, STM, Parsec (have I missed any?) and of course the transformer versions. I am debating whether or not to add [] to the bunch.

Whether you add [] or not, you should definitely include Maybe. Maybe captures the most basic kind of fallible computation, so it shows up all over the place with pragmatic coding. Compare against null pointers, returning -1 to signal error when a positive number is the expected return, using 0 to express an infinite limit on some kind of resource, etc. Maybe does the same thing, except it does them cleanly and correctly because we express the possibility of failure in the type system instead of relying on "magic values" to express them. Magic values are even worse than magic numbers and other magic constants, IMO.

Once you've explained Maybe, you can mention (Either a) in passing; they should figure out the generalization immediately.

-- Live well, ~wren

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe