Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.

When dealing with complicated algorithms in languages with support for nested functions (such as Python and D) I often write huge functions (because the algorithm is complicated) but mitigate this by using nested functions to structure the complicated code. Are huge (100+ line) functions still considered evil even if they're well-structured internally via the use of nested functions?

Edit: For those of you not familiar with Python or D, nested functions in these languages also allow access to the outer function scope. In D this access allows mutation of variables in the outer scope. In Python it only allows reading. In D you can explicitly disable access to the outer scope in a nested function by declaring it static.

share|improve this question
4  
I regularly write 400+ line functions/methods. Gotta put that 500 case switch statement somewhere :) –  Callum Rogers Sep 28 '10 at 13:20
6  
@Callum Rogers: In Python, instead of having 500 case switch statement, you'd use a lookup dictionary/mapping. I believe they're superior than having 500-or-so switch-case statement. You still need a 500 lines of dictionary definitions (alternatively, in Python, you can use reflection to dynamically enumerate them), but dictionary definition is data, not code; and there is much less qualms about having large data definition than large function definition. Additionally, data is more robust than code. –  Lie Ryan Sep 28 '10 at 23:15
    
I cringe every time i see functions with lots of LOC,makes me wonder what is the purpose of modularity.Its easier to follow the logic and debug with smaller functions. –  Aditya P May 6 '11 at 11:08
    
The ancient and familiar Pascal language also allows access to the outer scope, and so does the nested function extension in GNU C. (These languages only allow downward funargs only, however: that is, arguments which are functions that carry scope, can only be passed down, and not returned, which would require full lexical closure support). –  Kaz Jan 16 '13 at 22:52
    
A good example of functions that tend to be long are the combinators you write when doing reactive programming. They easily hit thirty lines, but would double in size if split apart because you lose closures. –  Strilanc Feb 26 at 3:40

11 Answers 11

Always remember the rule, a functions does one thing and does it well! If you can do so, avoid nested functions.

It hinders readability and testing.

share|improve this answer
8  
I've always hated this rule because a function that does "one thing" when viewed at a high level of abstraction may do "many things" when viewed at a lower level. If applied incorrectly, the "one thing" rule can lead to excessively fine-grained APIs. –  dsimcha Sep 28 '10 at 12:49
1  
@dsimcha: What rule can't be misapplied and lead to problems? When someone is applying "rules"/platitudes past the point where they're useful, just bring out another: all generalizations are false. –  Roger Pate Sep 29 '10 at 0:14
2  
@Roger: A legitimate complaint, except that IMHO the rule often is misapplied in practice to justify excessively fine-grained, overengineered APIs. This has concrete costs in that the API becomes harder and more verbose to use for simple, common use cases. –  dsimcha Sep 29 '10 at 0:40
2  
"Yes, the rule is misapplied, but the rule is misapplied too much!"? :P –  Roger Pate Sep 29 '10 at 5:18
1  
My function does one thing and it does that thing very well: namely, occupy 27 screenfuls. :) –  Kaz Jan 16 '13 at 22:53

Ideally, the entire function should be viewable without having to scroll. Sometimes, this isn't possible. But if you can break it up into pieces then it will make reading the code a lot easier.

I know that as soon as I push Page Up/Down or move to a different section of the code, I can only remember 7 +/- 2 things from the previous page. And unfortunately, some of those locations are going to be used when reading the new code.

I always like to think about my short-term memory like a computer's registers (CISC, not RISC). If you have the entire function on the same page, you can go to cache to get the required information from another section of the program. If the entire function cannot fit on a page, that would be the equivalent of always pushing any memory to disk after every operation.

share|improve this answer
2  
You just need to print it out on a matrix printer - see, 10 meters of code all at once :) –  user1249 Oct 17 '10 at 12:21
    
Or get a monitor with higher resolution –  frogstarr78 Nov 12 '10 at 6:45
    
And use it in vertical orientation. –  Calmarius Mar 14 '13 at 15:07

Some have argued that short functions can be more error-prone than long functions.

Card and Glass (1990) point out that the design complexity really involves two aspects: the complexity within each component and the complexity of the relationships among components.

Personally, I've found that well-commented straight-line code is easier to follow (especially when you weren't the one who originally wrote it) than when it is broken up into multiple functions that are never used elsewhere. But it really depends on the situation.

I think the main take-away is that when you split up a block of code, you are trading one kind of complexity for another. There is probably a sweet spot somewhere in the middle.

share|improve this answer
    
Have you read "Clean Code"? It discusses this at length. amazon.com/Clean-Code-Handbook-Software-Craftsmanship/dp/… –  Martin Wickman Oct 26 '10 at 6:49

Why use nested functions, rather than normal external functions?

Even if the external functions are only ever used in your one, formerly-big function, it still makes the whole mess easier to read:

DoThing(int x){
    x += ;1
    int y = FirstThing(x);
    x = SecondThing(x, y);
    while(spoo > fleem){
        x = ThirdThing(x+y);
    }
}

FirstThing(int x){...}
SecondThing(int x, int y){...}
ThirdThing(int z){...}
share|improve this answer
2  
Two possible reasons: cluttering the namespace, and what I'll call "lexical proximity". These might be good or bad reasons, depending on your language. –  Frank Shearar Sep 28 '10 at 11:28
    
But nested functions can be less efficient because they need to support downward funargs, and possibly full blown lexical closures. If the language is poorly optimized, it may do extra work to create a dynamic environment to pass to the child functions. –  Kaz Jan 16 '13 at 22:55

The answer is it depends, however you should probably turn that into a class.

share|improve this answer
    
Unless you're not writing in an OOPL, in which case maybe not :) –  Frank Shearar Sep 28 '10 at 11:24
    
+1: Stops namespace clutter. –  Callum Rogers Sep 28 '10 at 13:19
    
dsimcha did mention they were using D and Python. –  frogstarr78 Nov 12 '10 at 6:46
    
Classes are too often crutch for people who have never heard of things like lexical closures. –  Kaz Jan 16 '13 at 22:56

I don't like most nested functions. Lambdas fall in that category but usually don't flag me unless they have more than 30-40 characters.

The basic reason is that it becomes a highly locally dense function with internal semantic recursion, meaning that it's hard for me to wrap my brain around, and it's just easier to push some stuff out to a helper function that doesn't clutter the code space.

I consider that a function should Do Its Thing. Doing Other Things is what other functions do. So if you have a 200-line function Doing Its Thing, and it all flows, that's A-OK.

share|improve this answer
    
Sometimes you have to pass so much context to an external function (which it could just have convenient access to as a local function) that there is more complication in how the function is invoked than what it does. –  Kaz Jan 16 '13 at 22:57

I don't have the book in front of me right at this moment (to quote), but according to Code Complete the "sweetspot" for function length was around 25-50 lines of code according to his research.

There are times where is ok to have long functions:

  • When the cyclomatic complexity of the function is low. Your fellow developers might get a little frustrated if they have to look at a function that contains a giant if statement and the else statement for that if is not on the screen at the same time.

The times where it's not ok to have long functions:

  • You have a function with deeply nested conditionals. Do your fellow code readers a favor, improve the readability by breaking up the function. A function provides an indication to it's reader that "This is a block of code that does one thing". Also ask yourself if the length of the function indicates that it's doing too much and it needs to be factored out to another class.

The bottom line is that maintainability should be one of the highest priorities on your list. If another dev can't look at your code and get a "gist" of what the code is doing in less than 5 seconds, ur code doesn't provide enough "metadata" to tell what it's doing. Other devs should be able to tell what your class is doing just by looking at the object browser in your chosen IDE instead of reading 100+ lines of code.

Smaller functions have the following advantages:

  • Portability: It's much easier to move functionality around (either within the class on refactoring to a different one)
  • Debugging: When you look at the stacktrace it's much faster to pinpoint an error if you are looking at a function with 25 lines of code rather than 100.
  • Readability - The name of the function tells what an entire block of code is doing. A dev on your team might not want to read that block if they aren't working with it. Plus, in most modern IDE's another dev can have a better understanding what your class is doing by reading the function names in an object browser.
  • Navigation - Most IDE's will let you search on the name of functions. Also, most modern IDE's have the ability to view the source of a function in another window, this gives other devs to look at your long function on 2 screens (if the multi-monitors) instead of making them scroll.

The list goes on.....

share|improve this answer

Is it acceptable? That's really a question only you can answer. Does the function achieve what it needs to? Is it maintainable? Is it 'acceptable' to the other members of your team? If so, then that's what really matters.

Edit: I didn't see the thing about the nested functions. Personally, I'd not use them. I'd use regular functions instead.

share|improve this answer

No, multi-page functions are not desirable, and should not pass code review. I used to write long functions too, but after reading Martin Fowler's Refactoring, I stopped. Long functions are hard to write correctly, hard to understand and hard to test. I have never seen a function of even 50 lines that wouldn't be more easily understood and tested if it were split into a set of smaller functions. In a multi-page function there are almost certainly entire classes that should be factored out. It's hard to be more specific. Maybe you should post one of your long functions to Code Review and someone (maybe me) can show you how to improve it.

share|improve this answer

When I'm programming in python, i like to step back after I've written a function and ask myself if it adheres to the "Zen of Python" (type 'import this' in your python interpreter):

Beautiful is better than ugly.
Explicit is better than implicit.
Simple is better than complex.
Complex is better than complicated.
Flat is better than nested.
Sparse is better than dense.
Readability counts.
Special cases aren't special enough to break the rules.
Although practicality beats purity.
Errors should never pass silently.
Unless explicitly silenced.
In the face of ambiguity, refuse the temptation to guess.
There should be one-- and preferably only one --obvious way to do it.
Although that way may not be obvious at first unless you're Dutch.
Now is better than never.
Although never is often better than right now.
If the implementation is hard to explain, it's a bad idea.
If the implementation is easy to explain, it may be a good idea.
Namespaces are one honking great idea -- let's do more of those!

share|improve this answer
1  
If explicit is better than implicit, we should code in machine language. Not even in assembler; it does nasty implicit things like automatically fill branch delay slots with instructions, calculate relative branch offsets, and resolve symbolic references among globals. Man, these dumb Python users and their little religion ... –  Kaz Jan 16 '13 at 23:00

A long "straight-line" function can be a clear way to specify a lengthy sequence of steps that always occur in a particular sequence.

However, as others have mentioned, this form is prone to having local stretches of complexity in which the overall flow is less evident. Placing that local complexity into a nested function (ie: defined elsewhere within the long function, perhaps at top or bottom) can restore clarity to the mainline flow.

A second important consideration is controlling the scope of variables that are intended to be used only in a local stretch of a long function. Vigilance is required to avoid having a variable that was introduced in one section of code unwittingly referenced somewhere else (for example, after cycles of editing), as this kind of mistake will not show up as a compile or runtime error.

In some languages, this problem is easily averted: a local stretch of code can be wrapped in its own block, such as with "{...}", within which any newly-introduced variables are only visible to that block. Some languages, such as Python, lack this feature, in which case local functions can be useful to enforce smaller scope regions.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.