the biggest trouble in writing code...

[juan_gandhi]

is not with algorithms; there's hardly any algorithm that is hard to implement, once it's clearly formulated.
And it is probably not even performance, since it is another algorithm issue.

It is our struggle with monads. With monads that don't commute.

That's what makes Haskell so good (and so hard): it makes us to mention monads explicitly.

Comments

Or maybe

[cema.livejournal.com]

With monads that don't commute.

With monads that don't compute.

:-)

[antilamer.livejournal.com]

Care to elaborate? With all due respect to monads, I've no idea what you mean.

[wizzard]

He's probably using monads to abstract out small-scale 'software components', like algorithm pieces.

[antilamer.livejournal.com]

But I still don't get why their commutation is so important.

[ivan-gandhi.livejournal.com]

People hide monadic nature of their pieces of code because they don't know how to deal with combinations of monads.

E.g. you have an io, and it throws an exception, and some data you input are not numbers as you expect, and some attributes are NULL, and some strings are too short.

[antilamer.livejournal.com]

Now it makes more sense, though I'm still not sure whether it is *that* important, and what we'd have if the monads commuted (and what that would even mean).

[ivan-gandhi.livejournal.com]

By commuting I mean converting a list of partial values to a partial list of values and vice versa, for instance.

Their non-commuting makes life harder, but more interesting... maybe this is the part of our profession is less formalizable than most.

[ivan-gandhi.livejournal.com]

No, I'm talking on a much smaller scale. Non-total functions being used as if they were total, for instance.

[wizzard]

Not using them in such a way would require a total description of the surrounding environment, which is impossible.

I think Erlang is just the opposite (reducing complexity by reducing the range of possible inputs)

[ivan-gandhi.livejournal.com]

I mean, we hardly ever have pure functions, right? But in the code, say, in Java, people pretend to have pure functions, sighing and doing something when it turns out kaboom those are not pure functions. String.charAt, Integer.valueOf(), etc, they are not functions. I mean, they are arrows in a Kleisli category for a certain monad, but people pretend not to notice it.

Haskell, on the other hand, at least ideally, disambiguates between pure, total functions and the arrows that make sense only in a certain Kleisli category, forcing people to explicitly use something monadic.

To me, this is similar to Java generics (which Java people kind of hate) that force people to write a more type-safe code.