These languages use weaker monad syntax because of the difficulty conceptualizing programming in a way other than "Do arbitrary action A to object B in scope C". You can't gain from a more powerful monad abstraction without losing the constraints of impure functions, dynamic typing, and strict evaluation.
It is therefore incumbent upon Haskellers to make the superiority of powerful abstractions like the monad abundantly clear through the propagation of high-quality software using said abstraction. :)
Monads are not a superior abstraction. Their composition is not commutative and, worse, not even closed (a composition of two monads may not be a monad). Monad transformers are clumsy syntax-wise and have a performance penalty. When I used Haskell, monads were the most meh part: not a pain, but not anything too cool either. Just some syntax sugar to remove a repetitive wrapping function from a few lines of code.
You can't gain from a more powerful monad abstraction without losing the constraints of impure functions, dynamic typing, and strict evaluation.
Those languages that choose to use options or results with ?. instead of null or exceptions are already choosing to avoid impure functions and use static typing. Monads work fine with strict evaluation (indeed we see post-Haskell languages e.g. Idris using a strict evaluation model).
Monads do work fine in languages with strict evaluation.
In and of itself, that isn't really any more compelling a statement than observing that the strategy pattern works fine in languages with first-class functions. Just because you can do something in a certain way doesn't mean it's your best option. The argument's got to go a bit further than that.
Monads rose to prominence in Haskell in order to deal with some specific technical challenges that stemmed from the project's decision not to make any compromises about lazy evaluation. Using monads for specifically that purpose in a language with strict evaluation would be a case of cargo culting Haskell.
That is certainly not the only use for monads. Haskell uses them as a general-purpose solution for things that other languages typically handle with an absolute zoo of special-purpose features. In that respect, you could argue that monads are, at least in the abstract, better than all those features, because they're one thing that can do the work of many.
However, if you're working in a language that already has some mix of strict evaluation, mutable variables, async/await keywords, exceptions, elvis operators, null punning, etc., introducing monads to solve those problems runs the real risk of running afoul of the xkcd.com/927 problem. Especially if those existing language have a tendency to not understand and respect the fact that they're now being used in code that's supposed to be observing the monad laws. Even implementing `Maybe` in a language with nulls can be surprisingly tricky to get right.
Which is where I stop having much interest in holy wars about this stuff. If you're lucky enough to be working in Haskell or a language inspired by it, that's awesome, you've got a whole menagerie of category theoretic abstractions to work with, and a language that's actually designed to help you use them with confidence. If you're not, then I'd argue that their use needs to be justified in much more specific, pragmatic terms, including an eyes-wide-open perspective on what the ergonomics and technical challenges are going to be like. Because, believe me, it's no fun having to work with a Try monad that can both return and throw errors.
Monads rose to prominence in Haskell in order to deal with some specific technical challenges that stemmed from the project's decision not to make any compromises about lazy evaluation.
Disagree. To an outsider, IO is the most prominent monad because it's very visible in small example programs (and something that separates the Haskell versions of those small example programs from those in other languages). But to a practitioner it's not a particularly important or interesting example, most use of the monad abstraction is not about IO, and I would certainly hope that the language (in the broad sense) would have adopted it with or without IO.
Using monads for specifically that purpose in a language with strict evaluation would be a case of cargo culting Haskell.
Disagree; even in languages which define an evaluation order, it's still very much implicit to a human reader. Being able to distinguish between accidental ordering and intentional ordering is really helpful to enable fearless refactoring. Like I said, post-Haskell languages like Idris often use strict evaluation but still see value in managing IO explicitly.
However, if you're working in a language that already has some mix of strict evaluation, mutable variables, async/await keywords, exceptions, elvis operators, null punning, etc., introducing monads to solve those problems runs the real risk of running afoul of the xkcd.com/927 problem.
The special-case syntax is often more suitable to the special cases it was designed for (it would be pretty tragic if it weren't). But the general-case syntax is necessary for general-case code. It can work pretty nicely as long as there's a direct transformation between one and the other; compare e.g. different ways of iterating through a collection, where you'll generally have a fully general function/syntax but also sugar that's more appropriate to restricted special cases.
Especially if those existing language have a tendency to not understand and respect the fact that they're now being used in code that's supposed to be observing the monad laws.
I'd argue that if your special-case solution doesn't conform to the monad laws then you already have a problem, even if you don't realise it yet. E.g. the fact that in many languages code that uses a map will break if that map contains null values, and the fact that implementing a valid maybe monad on top of null is difficult-to-impossible, are both reflections of the same underlying problem.
Those languages that choose to use options or results with ?. instead of null or exceptions are already choosing to avoid impure functions and use static typing.
An impure "function" is any procedure which may have side-effects or produce different results for the same formal arguments depending on the runtime environment or evaluation order. Most languages with option or result types (e.g. Rust) still make extensive use of impure procedures. Haskell and its close relatives are pretty much the only well-known exceptions with first-class pure functions enforced through the type system.
Purity is always a spectrum - Haskell allows non-terminating "functions", for example. Choosing to treat certain kinds of failures as valid function evaluations that can be reasoned about under the normal rules of the language is a step in the direction of purity. More to the point, it's the aspect of purity that's salient when we're talking about whether it makes sense to regard these particular constructs as monads.
Haskell allows non-terminating "functions", for example.
Non-terminating functions are still pure functions in the mathematical sense. While it does somewhat complicate the use of programs as proofs, a pure function doesn't need to have a defined value for every possible input. A better example might have been unsafePerformIO which, like "unsafe" in Rust, is meant to be used to construct a pure interface to impure code but depends on the programmer to handle it properly. The difference is that Rust doesn't require "unsafe" around all side-effects, just those that may impact memory safety.
I would agree that Option and Result types represent "a step in the direction of purity", but even a little bit of impurity nullifies referential transparency and inhibits equational reasoning.
... it's the aspect of purity that's salient when we're talking about whether it makes sense to regard these particular constructs as monads.
A construct is a monad if it obeys the monad laws for all well-typed inputs. In languages like Rust or JS which lack any type-level enforcement of purity the constructs are only monads under the condition that the inputs happen to be pure, not in the general case. For example, for any functor (which includes all monads) we have the law "map f . map g == map (f . g)". However, if f and g may have side effects then substituting one side for the other will interleave the effects in a different order and potentially change the result, so the monad laws are not satisfied.
Non-terminating functions are still pure functions in the mathematical sense.
No they're not. Mathematically a function must evaluate to a value.
I would agree that Option and Result types represent "a step in the direction of purity", but even a little bit of impurity nullifies referential transparency and inhibits equational reasoning.
It doesn't nullify anything; the "Fast and Loose Reasoning is Morally Correct" result applies just as well to a language with, say, nulls, as it does to the impure parts of Haskell that it was originally addressed at. The more relevant these aspects of your language are to practical programs, the less useful reasoning one can do, but it's very much a spectrum rather than a binary.
In languages like Rust or JS which lack any type-level enforcement of purity the constructs are only monads under the condition that the inputs happen to be pure, not in the general case. For example, for any functor (which includes all monads) we have the law "map f . map g == map (f . g)". However, if f and g may have side effects then substituting one side for the other will interleave the effects in a different order and potentially change the result, so the monad laws are not satisfied.
Pure is not a binary; rather the law holds to the extent that the functions are pure (that is, the two sides of the law are equivalent to each other in the same sense that the function is equivalent to its evaluation). Even in Haskell you have cases of the same kind of law violation where two expressions should be equivalent (according to the monad laws) but one terminates and the other doesn't.
Mathematically a function must evaluate to a value.
For every input within its domain, yes. Whether this is a problem for partial functions will depend on how you define the domain: either exactly the set permitted by the function's type, or the subset of well-typed inputs with an associated value. As far as I can tell the most common definition of the domain for a relation or function is the set of inputs which have at least one associated result; on the other hand, those definitions are not concerned with the function's type. I would say that the input type is only an approximation (superset) of the function's domain, with better type systems permitting closer approximations. (For perspective, I've never known anyone to argue that the result type must perfectly capture the function's range, which is defined much the same way as the set of results associated with at least one input.)
The more relevant these aspects of your language are to practical programs, the less useful reasoning one can do...
There is indeed a spectrum of varying degrees of impurity among impure languages, depending on both language design and custom among its users. However, there is one key area where the classification is binary, and that is in the answer to the question: Does the language assume referential integrity or not? In Haskell the answer is "yes". The compiler will make substitutions under the assumption that evaluation does not have side effects; if you break that expectation, via unsafePerformIO or other means, the result is undefined. In Rust or Javascript the answer is "no", and various optimizations are prevented because the compiler cannot assume that the evaluation of an unknown function will not have side effects.
Even in Haskell you have cases of the same kind of law violation where two expressions should be equivalent (according to the monad laws) but one terminates and the other doesn't.
If one side doesn't terminate then you can never get to the point of observing that they have different results. The point of "fast and loose reasoning" is that you only need to prove that the laws are never broken within the program. The condition "map f . map g == map (f . g)" cannot evaluate to False... but that doesn't mean it must evaluate to True. In Rust or JS, however, that condition could evaluate to false in the presence of side effects.
Comments
These languages use weaker monad syntax because of the difficulty conceptualizing programming in a way other than "Do arbitrary action A to object B in scope C". You can't gain from a more powerful monad abstraction without losing the constraints of impure functions, dynamic typing, and strict evaluation.
It is therefore incumbent upon Haskellers to make the superiority of powerful abstractions like the monad abundantly clear through the propagation of high-quality software using said abstraction. :)
Monads are not a superior abstraction. Their composition is not commutative and, worse, not even closed (a composition of two monads may not be a monad). Monad transformers are clumsy syntax-wise and have a performance penalty. When I used Haskell, monads were the most meh part: not a pain, but not anything too cool either. Just some syntax sugar to remove a repetitive wrapping function from a few lines of code.
Scala has monad abstractions (and even special syntax for it) without enforcing purity or having lazy evaluation by default.
Those languages that choose to use options or results with ?. instead of null or exceptions are already choosing to avoid impure functions and use static typing. Monads work fine with strict evaluation (indeed we see post-Haskell languages e.g. Idris using a strict evaluation model).
Monads do work fine in languages with strict evaluation.
In and of itself, that isn't really any more compelling a statement than observing that the strategy pattern works fine in languages with first-class functions. Just because you can do something in a certain way doesn't mean it's your best option. The argument's got to go a bit further than that.
Are you suggesting that languages with strict evaluation have some better alternative to monads? Because that's simply not true.
Talking specifically about strict evaluation:
Monads rose to prominence in Haskell in order to deal with some specific technical challenges that stemmed from the project's decision not to make any compromises about lazy evaluation. Using monads for specifically that purpose in a language with strict evaluation would be a case of cargo culting Haskell.
That is certainly not the only use for monads. Haskell uses them as a general-purpose solution for things that other languages typically handle with an absolute zoo of special-purpose features. In that respect, you could argue that monads are, at least in the abstract, better than all those features, because they're one thing that can do the work of many.
However, if you're working in a language that already has some mix of strict evaluation, mutable variables, async/await keywords, exceptions, elvis operators, null punning, etc., introducing monads to solve those problems runs the real risk of running afoul of the xkcd.com/927 problem. Especially if those existing language have a tendency to not understand and respect the fact that they're now being used in code that's supposed to be observing the monad laws. Even implementing `Maybe` in a language with nulls can be surprisingly tricky to get right.
Which is where I stop having much interest in holy wars about this stuff. If you're lucky enough to be working in Haskell or a language inspired by it, that's awesome, you've got a whole menagerie of category theoretic abstractions to work with, and a language that's actually designed to help you use them with confidence. If you're not, then I'd argue that their use needs to be justified in much more specific, pragmatic terms, including an eyes-wide-open perspective on what the ergonomics and technical challenges are going to be like. Because, believe me, it's no fun having to work with a Try monad that can both return and throw errors.
Disagree. To an outsider, IO is the most prominent monad because it's very visible in small example programs (and something that separates the Haskell versions of those small example programs from those in other languages). But to a practitioner it's not a particularly important or interesting example, most use of the monad abstraction is not about IO, and I would certainly hope that the language (in the broad sense) would have adopted it with or without IO.
Disagree; even in languages which define an evaluation order, it's still very much implicit to a human reader. Being able to distinguish between accidental ordering and intentional ordering is really helpful to enable fearless refactoring. Like I said, post-Haskell languages like Idris often use strict evaluation but still see value in managing IO explicitly.
The special-case syntax is often more suitable to the special cases it was designed for (it would be pretty tragic if it weren't). But the general-case syntax is necessary for general-case code. It can work pretty nicely as long as there's a direct transformation between one and the other; compare e.g. different ways of iterating through a collection, where you'll generally have a fully general function/syntax but also sugar that's more appropriate to restricted special cases.
I'd argue that if your special-case solution doesn't conform to the monad laws then you already have a problem, even if you don't realise it yet. E.g. the fact that in many languages code that uses a map will break if that map contains null values, and the fact that implementing a valid maybe monad on top of null is difficult-to-impossible, are both reflections of the same underlying problem.
An impure "function" is any procedure which may have side-effects or produce different results for the same formal arguments depending on the runtime environment or evaluation order. Most languages with option or result types (e.g. Rust) still make extensive use of impure procedures. Haskell and its close relatives are pretty much the only well-known exceptions with first-class pure functions enforced through the type system.
Purity is always a spectrum - Haskell allows non-terminating "functions", for example. Choosing to treat certain kinds of failures as valid function evaluations that can be reasoned about under the normal rules of the language is a step in the direction of purity. More to the point, it's the aspect of purity that's salient when we're talking about whether it makes sense to regard these particular constructs as monads.
Non-terminating functions are still pure functions in the mathematical sense. While it does somewhat complicate the use of programs as proofs, a pure function doesn't need to have a defined value for every possible input. A better example might have been unsafePerformIO which, like "unsafe" in Rust, is meant to be used to construct a pure interface to impure code but depends on the programmer to handle it properly. The difference is that Rust doesn't require "unsafe" around all side-effects, just those that may impact memory safety.
I would agree that Option and Result types represent "a step in the direction of purity", but even a little bit of impurity nullifies referential transparency and inhibits equational reasoning.
A construct is a monad if it obeys the monad laws for all well-typed inputs. In languages like Rust or JS which lack any type-level enforcement of purity the constructs are only monads under the condition that the inputs happen to be pure, not in the general case. For example, for any functor (which includes all monads) we have the law "map f . map g == map (f . g)". However, if f and g may have side effects then substituting one side for the other will interleave the effects in a different order and potentially change the result, so the monad laws are not satisfied.
No they're not. Mathematically a function must evaluate to a value.
It doesn't nullify anything; the "Fast and Loose Reasoning is Morally Correct" result applies just as well to a language with, say, nulls, as it does to the impure parts of Haskell that it was originally addressed at. The more relevant these aspects of your language are to practical programs, the less useful reasoning one can do, but it's very much a spectrum rather than a binary.
Pure is not a binary; rather the law holds to the extent that the functions are pure (that is, the two sides of the law are equivalent to each other in the same sense that the function is equivalent to its evaluation). Even in Haskell you have cases of the same kind of law violation where two expressions should be equivalent (according to the monad laws) but one terminates and the other doesn't.
For every input within its domain, yes. Whether this is a problem for partial functions will depend on how you define the domain: either exactly the set permitted by the function's type, or the subset of well-typed inputs with an associated value. As far as I can tell the most common definition of the domain for a relation or function is the set of inputs which have at least one associated result; on the other hand, those definitions are not concerned with the function's type. I would say that the input type is only an approximation (superset) of the function's domain, with better type systems permitting closer approximations. (For perspective, I've never known anyone to argue that the result type must perfectly capture the function's range, which is defined much the same way as the set of results associated with at least one input.)
There is indeed a spectrum of varying degrees of impurity among impure languages, depending on both language design and custom among its users. However, there is one key area where the classification is binary, and that is in the answer to the question: Does the language assume referential integrity or not? In Haskell the answer is "yes". The compiler will make substitutions under the assumption that evaluation does not have side effects; if you break that expectation, via unsafePerformIO or other means, the result is undefined. In Rust or Javascript the answer is "no", and various optimizations are prevented because the compiler cannot assume that the evaluation of an unknown function will not have side effects.
If one side doesn't terminate then you can never get to the point of observing that they have different results. The point of "fast and loose reasoning" is that you only need to prove that the laws are never broken within the program. The condition "map f . map g == map (f . g)" cannot evaluate to False... but that doesn't mean it must evaluate to True. In Rust or JS, however, that condition could evaluate to false in the presence of side effects.
The real difficulty with having “proper” support for monads is having a type system that supports higher-kinded types of some description
Or alternatively no (static) type system at all.