How to read Haskell
(Small pattern matching example.)
m (Reformatted so people don't have to horizontally scroll.)
Revision as of 22:42, 3 July 2007
This (very incomplete) tutorial is aimed at the non-Haskeller who probably doesn't care too much about trying to write code, but wants to understand it. Our adopted format is a collection of tips and tricks broken down by category. It probably isn't very important what order you read it in, but it might be good to start with the general advice. Please feel encouraged to make any complaints about Haskell on the discussion page! It will help us to improve this tutorial.
Note: you should also consider having a look at Haskell for C Programmers. It might be a good way to get over the culture shock.
1 General advice
1.1 Tip: it's just very very concise
One thing that can make Haskell hard to read is that Haskell code is extremely succinct. One tiny little piece of code can say a lot, so many times, when you are faced with something you don't understand, the best thing you can do is to think about it for some time. It will usually make sense after a while. The good news is that because of this succinctness, Haskell functions tend to be very small, which means that when you're trying to understand a difficult piece of Haskell, you normally do not have to look very far. It's just two sides of the same coin:
- bad news: high density == spending more time per line of code
- good news: succinctness == fewer lines of code to spend time on
Spending on this time to get one tiny line of code may be frustrating, but it's well worth the effort, because the fact that a very small code is hard to understand probably means that it's very abstract, and the fact that it is abstract probably means that it's going to be used in many places. So understanding that one tiny line code, as painful as it may have been initially, can pay off in a big way.
1.2 Trick: use the haddock
When reading a long piece of Haskell code, one which is broken up into many modules, you should consider keeping a browser window open with the auto-generated API documentation on the side (if any).
2 What does this function do?
2.1 Trick: use type signatures
When you see stuff like this
map :: (a -> b) -> [a] -> [b]
...don't fight it! These are type signatures and they are an incredibly useful way of getting a rough idea what a function is supposed to do.
For example, the function above takes any function that turns an 'a' into a 'b' ('a -> b') and yields a function that takes a list of a's ('[a]') and produces a list of b's ('[b]'). For example, 'sqrt' takes a number and returns the square root of that number, so 'map sqrt' takes a list of numbers and returns a list of their square roots.
As another example,
swap :: (a,b) -> (b,a)
This takes a tuple ('(a,b)') and swaps its values around ('(b,a)').
Hint: If you see '->' inside brackets, then you have a high-order function. (A function that takes another function as a parameter.) These are extremely common in Haskell.
Hint: If the type has 'IO' in it anywhere, it does something to do with input/output. (Conversely, if there isn't any 'IO' in the type, the function cannot possibly perform any I/O of any kind.) Most usually the IO part comes at the end, either as 'IO ()' ("this function does some kind of input and/or output, and doesn't 'return' anything useful") or as something like 'IO Int' ("this function does input and/or output and returns an Int").
- Lots more hints could be added here. How much do we want to go into this? (Perhaps mention lists and tuples, since they have weird syntax?)
2.2 Tip: Haskellers love pattern matching
head [x] = x
This says that if 'head' is followed by a list containing only 1 item, label that item as 'x', and then return 'x'. Another example might be
fst (x,y) = x snd (x,y) = y
These functions fetch the first and second items in a tuple, respectively. It should be fairly obvious how they work.
2.3 Tip: a function may be defined in more than one piece
Remember math class, where functions would be defined like abs(x) = x if x >= 0 or -x otherwise? It's a bit like that in Haskell too. Sometimes, rather than writing one big if-then-else, Haskellers find it more convenient to define a function separately for each case, such as...
abs x | x >= 0 = x abs x = -x
What gets confusing is when you look at a definition like this...
foo x | blah = some enormous long thing foo x = some other enormously long thing
Especially looking at the bottom bit, it's hard to remember that
foo might have a another definition lurking around. Luckily, you never have to look very far, either immediately above or immediately below the other definition.
(Note: some programmers will perhaps write something like
foo x | otherwise = .... The
otherwise is redundant (and equal to
True), but useful as reminder that this isn't the entire definition of
2.4 Tip: pattern matching and guards can be mixed and matched
combine ((f,a,b,r):(f',a',b',r'):ss) | f == f' = combine ((f,a.+a',b.+b',r+r'):ss) combine ((f,a,b,r):ss) = (f,a,b,r) : combine ss combine  = 
3 What the heck is xyz?
One problem you might face when reading Haskell code is figuring out some cryptic entity like
3.1 Tip: the smaller the name, the smaller the scope
Do you hate the way Haskell code is littered with short, meaningless name like
xs? When Haskell programmers use names like that, it's often for good reason.
- elaborate: (1) scope size, (2) abstraction means it would be silly to give long names
3.2 Tip: the -s and m- habits
There is a variable name habit that sometimes comes with short names. Typically, if you have a thing you want to name
x, you'll sometimes want to name lists of these
xs. As in the plural of
x. So if you see a name like
foos, it's often good to mentally read that as "aeyes" (the plural of a), "bees" (the plural of b), and "foohs" (the plural of foo). It might seem obvious to some, but it took me a while to stop asking myself in situations like this, "
as? What the heck is aey-ess?"
Similarly, another habit you might see is people who begin variable names with m-. This is probably less common, but if you see a lot of m-, it might be because of the Maybe type. Sometimes we have
foo of type
mfoo of type
Maybe Whatever. Relax, this isn't Hungarian notation. It's not something that's used systematically, or rigidly in any way.
Both of these conventions are just helpful when you have both variants floating around in the same place, that is, when you have both Whatever and [Whatever] (that would be list of whatever),
xs is a good way to indicate that they are both the same thing, except one comes in a list. Likewise, when you have both Whatever and Maybe Whatever in the same function,
mx are too.
Finally, library functions are sometimes suffixed with "l", "r", "_", "M" or "'". What do these mean?
mapM -- the 'map' function lifted into a monad. An 'M' suffix implies that the function is a -- monadic version of an equivalent pure function mapM_ -- the '_' suffix indicates that the result of this computation is discarded, and () is -- returned (by analogy with the _ pattern). foldl -- a fold that traverses its structure left to right foldr -- a fold that traverses its structure right to left foldl' -- a fold that is strict in its accumulator, "'" is used to indicate a strict variant of -- a function
3.3 Tip: order doesn't matter
It doesn't matter what order functions are defined in. This:
foo x y z = ... bar a b = ... foo b ...
is exactly equivilent to this:
bar a b = ... foo b ... foo x y z = ...
Functions further up can call functions that are defined lower down, and vice versa. Functions can be written in any order at all. It doesn't matter.
Very important: The parts that make up a single function are in a particular order -- and that order is very important. The order in which separate functions are defined doesn't matter.
- scope in a nutshell
- last paragraph is weak
3.4 Trick: use grep
(This might seem really obvious, but it's sometimes easy to forget)
Or use the search feature of your favourite text editor. It's probably defined right there before your eyes, and if it's true to Haskell style, the definition is probably so small you blew right through it. In vi, for example, you could do
/= *xyz which searches for =, an arbirtary number of spaces, and then xyz.
xyz might be defined in some different module in the code you downloaded. You can look for telltale signs like
import Manamana (xyz)
But note that sometimes programmers get lazy, and they don't specify that
xyz should be imported. They just let rip with
So solution number 3 would be do something like
grep xyz *.lhs *.hs
(Note that literate programs sometimes use non-literate code, so search in both lhs AND hs)
A fourth idea, if you can't find something, is to look it up in Hoogle
A fifth idea, for Hugs/WinHugs users, is to use the ":find" command, ":find xyz" will open up your text editor with the appropriate module, jumped to the correct place.