The article you mention is about modularity in general, and it would apply equally to structured, functional, and object-oriented programs. I have heard of that article before from someone who was a big OOP guy, but I read it as an article about programming in general, not something OOP specific. There is a famous article about functional programming, Why Functional Programming Matters, and the first sentence of the conclusion states “In this paper, we’ve argued that modularity is the key to successful programming.” So the answer to (1) is no.

Well designed functions don’t assume more about their data than they need to, so the part about “intimately aware of the data” is wrong. (Or at least as wrong as it would be of OOP. You can’t program strictly at a high level of abstraction and ignore all details forever in any paradigm. In the end, some part of the program does actually need to know about the specific details of the data.)

Data hiding is an OOP specific term, and it isn’t exactly the same as the information hiding discussed in the article. Information hiding in the article is about design decisions that were hard to make or are likely to change. Not every design decision about a data format is hard or likely to change, and not every decision that is hard or likely to change is about a data format. Personally, I can’t see why OO programmers want everything to be an object. Sometimes, a simple data structure is all you need.

Edit:
I found a relevant quote from an interview with Rich Hickey.

Fogus: Following that idea—some people are surprised by the fact that Clojure does not engage in data-hiding encapsulation on its types. Why did you decide to forgo data-hiding?

Hickey: Let’s be clear that Clojure strongly emphasizes programming to abstractions. At some point though, someone is going to need to have access to the data. And if you have a notion of “private”, you need corresponding notions of privilege and trust. And that adds a whole ton of complexity and little value, creates rigidity in a system, and often forces things to live in places they shouldn’t. This is in addition to the other losing that occurs when simple information is put into classes. To the extent the data is immutable, there is little harm that can come of providing access, other than that someone could come to depend upon something that might change. Well, okay, people do that all the time in real life, and when things change, they adapt. And if they are rational, they know when they make a decision based upon something that can change that they might in the future need to adapt. So, it’s a risk management decision, one I think programmers should be free to make.
If people don’t have the sensibilities to desire to program to abstractions and to be wary of marrying implementation details, then they are never going to be good programmers.

… and is probably one of those articles that OOP was based on.

Not really, but it did added to the discussion, specially to practitioners that, at the time, were trained to decompose systems using the first criteria he describes in the paper.

First I want to know if my assessment is correct. Does the FP paradigm and this article philosophically disagree?

No. Moreover, to my eyes, your description of what an FP program looks like is not different from any other that uses procedures or functions:

Data gets passed from function to function, each function being intimately aware of the data and “changing it” along the way.

…except for the “intimacy” part, since you can (and often do) have functions operating on abstract data, precisely to avoid the intimacy. Thus, you do have some control over that “intimacy” and you can regulate it however you like, by setting up interfaces (ie. functions) for what you want to hide.

So, I see no reason why we wouldn’t be able to follow Parnas criteria of information hiding using functional programming and end up with an implementation of a KWIC index with similar pointed benefits as his second implementation.

Assuming they agree, I’d like to know what FPs implementation of data hiding is. It’s obvious to see this in OOP. You can have a private field that nobody outside the class can access. There’s no obvious analogy of this to me in FP.

As far as data is of concern, you can elaborate data abstractions and data type abstractions using FP. Any of these hide concrete structures and manipulations of these concrete structures using functions as abstractions.

EDIT

There is a growing number of assertions here stating that “hiding data” in the context of FP is not so useful (or OOP-ish (?)). So, let me stamp here a very simple and clear example from SICP:

Suppose your system needs to work with rational numbers. One way you might want to represent them is as a pair or a list of two integers: the numerator and denominator. Thus:

(define my-rat (cons 1 2)) ; here is my 1/2 

If you ignore data abstraction, most likely you will get the numerator and denominator using car and cdr:

(... (car my-rat)) ; do something with the numerator

Following this approach, all parts of the system that manipulate rational numbers will know that a rational number is a cons — they will cons numbers to create rationals and extract them using list operators.

One problem you may face is when you need to have a reduced form of the rational numbers — changes will be required across the entire system. Also, if you decide to reduce at creation time, you might find later that reducing when accessing one of the rational terms is better, yielding another full scale change.

Another problem is if, hypothetically, an alternative representation for them is preferred and you decide to abandon the cons representation — full scale change again.

Any sane effort in dealing with these situations will likely start to hide the representation of rationals behind interfaces. At the end, you might end up with something like this:

• (make-rat <n> <d>) returns the rational number whose numerator is the integer <n> and whose denominator is the integer <d>.

• (numer <x>) returns the numerator of the rational number <x>.

• (denom <x>) returns the denominator of the rational number <x>.

and the system will no longer (and should no longer) know of what rationals are made of. This is because cons, car and cdr are not intrinsic to rationals, but make-rat, numer and denom are. Of course, this could easily be an FP system. So, “data hiding” (in this case, better known as data abstraction, or the effort to encapsulate representations and concrete structures) comes as a relevant concept and a technique widely used and explored, whether in the context of OO, functional programming or whatever.

And the point is…though one may try to make distinctions between what “kind of hiding” or encapsulation they are doing (whether they are hiding a design decision, or data structures or algorithms — in the case of procedural abstractions), all of them have the same theme: they are motivated by one or more points Parnas made explicit. That is:

• Changeability: whether required changes can be made locally or are spreaded through the system.
• Independent Development: to what degree two parts of the system can be developed in parallel.
• Comprehensibility: how much of the system is required to be known to understand one of its parts.

The example above was taken from SICP book so, for the full discussion and presentation of this concepts in the book, I highly recommend checking out chapter 2. I also recommend getting familiar with Abstract Data Types in the context of FP, which brings other issues to the table.

Your belief that functional programming lacks data hiding is wrong. It just takes a different approach to hiding data. One of the most common ways of hiding data in functional programming is through the use of polymorphic functions that take a function as an argument. For example, this function

map :: (a -> b) -> [a] -> [b]
map _ [] = []
map f (x:xs) = f x : map f xs

can only see the outermost structure of the data (i.e. that it is a list) it can’t see anything about the data that the list contains and can only operate on the data through the single function that is passed to it.

The function that is passed as an argument is analogous to a public method on the data type that the list contains. It provides a limited way of operating on the data, but does not expose the internal workings of the data type.

I’m going to strike out on a limb here and say that the concept is just not relevant in FP the way it is in OO.

tl;dr;
The point of data hiding is to ensure responsibilities are maintained where they should be, and you don’t have outside actors messing with data they don’t have the knowledge for. In FP data is generated by expressions, and in this way you can’t mess with the data because it’s not mutable properties so much as composable computations which completely changes the rules of the game.

In my experiences with FP; which are admittedly insignificant, I tend to find one stark contrast with OO in what denotes good/common data modeling.

This contrast is that in OO in general you model things to represent your data. Obligatory car analogy:

OO

• You have a car Object, that correctly hides details about the car such as the AC implementation (is it belt driven, or air-pressure-driven? Consumers shouldn’t need to know, so hide it).
• This car object has many properties and methods that delineate all the facts about the car as well as ways that you can work with a car.
• This car object has properties which are components of a car that further hide from the overall Car their particular implementations, and their data facts allowing the components of the car to be interchangeable.

The thing to note here is when you’re modeling things in an OO format, it’s all about representing things as Data. You have objects with properties, many of those properties are objects with more properties. You have a couple methods here and there attached to those objects, but all they really do is usually jigger the properties of the objects this way and that, again it’s very data-centric modeling; that is you model your data to be interacted with focussing on structuring it to make available all the points of your data so consumers may change the data this way and that.

FP

• You have a great deal of computations that allow you to describe behaviours
• These expressions of behaviour are related in a way that can be translated to the way a cars behaviours are related to eachother, such as a car having accelerate/decelerate, there are two behaviours which oppose eachother in a similar way.

The big difference between OO and FP that constantly strikes me is as I said above the way you model data. In OO as mentioned above you model data as data, in FP you model data as computations, expressions, algorithms, it’s more about modeling the activities of your data than the facts of it. Think about basic data modeling in math, it’s always about getting an equation that can generate your data, which models your data as the activity that causes it, as opposed to OO is modeling is coming up with a way to represent the data you have. That is much of the distinction between FP and OO.

Remember, for a great deal of time LISP, one of the foundational FP languages, lived with a very small amount of primitive data types. This works because the approach isn’t about modeling complex representations of your data so much as computations that generate and express the behaviours of your system.

When I start writing some code in FP, I start by writing code that does something, where as when I start writing code in OO, I start by writing models that describe something. The doing of things is hidden in FP by being expressions, the doing of things is exposed in OO by being described with data, hiding this data limits said exposure.

Back to the question at hand, what does FP say about data hiding, does it appreciate it or disagree with it or what not?

I say it doesn’t matter, in OO your data is the guts and important pieces in your program that should be hidden from being meddled with. In FP the guts and knowledge of your system is all hidden away in the algorithms and computations that express the system. These are by definition more or less immutable, the only way to mutate computation expressions are things like macros, but even then you’re mutations definitions are expressions themselves which cannot be further meddled with.

There is a bit of a paradox here. Even though functional programming focuses on, well, functions, and frequently has functions that work directly on primitive data types, it tends to have more data hiding than object oriented programming.

How is this so? Think about a nice OO interface that hides underlying data – perhaps collections (I’m trying to pick something nearly ubiquitous). You may not need to know the underlying type of the objects in the collection or type of the object implementing the collection, as long as you know the collection implements, say, IEnumerable. So you have data hiding.

In functional programming, you might write a function that effectively works with an IEnumerable interface, but operates on a primitive data type (or on any data type). But what if the type didn’t ever implement the IEnumerable methods? Here’s the key, you can always have the “methods” that form the needed pieces of the “interface” be parameters passed in to your function. Or you can put functions together with data and do things in an OO-like way.

Notice that either way you don’t have any less data hiding than you have in OO. My general function that works on any type clearly isn’t accessing the data in that type – that happens within the functions passed as parameters to the general function, but the general function never peeks inside those functions to see the data.

So, as far as your point 1, I don’t think FP and the article really disagree. I don’t think your characterization of FP not hiding data is correct. One could implement the design the author preferred in FP, certainly.

As far as point 4 (2 and 3 don’t make sense to answer given what I’ve said for point 1) it varies. It also varies in OO languages, and in many private fields are private by convention rather than enforced by the language.

First, thanks for the link to this great article, I did not know this so far, and it gave me some great input about some things which I was discussing with some other software designers from the community in the last years. Here is my opinion about it:

First I want to know if my assessment is correct. Does the FP paradigm and this article philosophically disagree?

FP design focuses very much on data-flow (which is IMHO not so bad as the article may imply). If this is a complete “disagreement” is argueable.

Assuming they disagree, how does FP “compensate” for its lack of data hiding? Perhaps they sacrifice data hiding but gain X, Y and Z. I’d like to know the reasoning for why X, Y and Z are considered more beneficial than data hiding.

IMHO it does not compensate. See below.

Or, assuming they disagree, perhaps FP feels that data hiding is bad. If so, why does it think data hiding is bad?

I don’t think most of the FP users or designers do feel or think in this way, see below.

Assuming they agree, I’d like to know what FPs implementation of data hiding is. It’s obvious to see this in OOP. You can have a private field that nobody outside the class can access. There’s no obvious analogy of this to me in FP.

Here is the point – you have probably seen so many OOP systems implemented in a non-functional way that you believe OOP is non-functional. And that’s a fallacy, IMHO OOP and FP are mostly orthogonal concepts, and you can perfectly build functional OO systems, which gives you an obvious answer to your question. The classic “object” implementation in FP is done by utilizing closures, and if you want objects to be used in a functional system, the key point is to design them immutable.

So for creating bigger systems, IMHO you can create modules, classes and objects using OO concepts, exactly the way described under “Modularization 2” in the article without leaving the “FP path”. You will use the module concept of your favorite FP language, make just all your objects immutable and use the “best of both worlds”.

TL;DR: No

Does the FP paradigm and this article philosophically disagree?.

No it doesn’t. Functional Programming is declarative which is “a style of building the structure and elements of computer programs, that expresses the logic of a computation without describing its control flow.” It’s less about following the flow chart and more like creating the rules that let the flow arise on it’s own.

Procedural programming is much closer to a encoding of a flow chart than Functional programming. It follows that transforms that occurs, and encodes of those transforms into procedures that are executed in order, exactly like the flow in a flow chart describes.

Whereas procedural languages model execution of the program as a sequence of imperative commands that may implicitly alter shared state, functional programming languages model execution as the evaluation of complex expressions that only depend on each other in terms of arguments and return values. For this reason, functional programs can have a freer order of code execution, and the languages may offer little control over the order in which various parts of the program are executed. (For example, the arguments to a procedure invocation in Scheme are executed in an arbitrary order.)

Data Hiding

• Functional Programming has it’s own methods of data hiding, for example think closures. That is data hiding by encapsulation in a closure. It’s hard for fields to be any more private data that has been closed over since only the closure has a reference to the data and you cannot refer to it external to the closure.
• One of the reasons for data hiding is to stabilize the programming interface by hiding mutating data. Functional programming doesn’t have mutating data, therefore it doesn’t need as much data hiding.