Haskell
Yet Another Haskell Tutorial
Preamble
Introduction
Getting Started
Language Basics (Solutions)
Type Basics (Solutions)
IO (Solutions)
Modules (Solutions)
Advanced Language (Solutions)
Advanced Types (Solutions)
Monads (Solutions)
Advanced IO
Recursion
Complexity


As we mentioned earlier, it is difficult to think of a good, clean way to integrate operations like input/output into a pure functional language. Before we give the solution, let's take a step back and think about the difficulties inherent in such a task.

Any IO library should provide a host of functions, containing (at a minimum) operations like:

  • print a string to the screen
  • read a string from a keyboard
  • write data to a file
  • read data from a file

There are two issues here. Let's first consider the initial two examples and think about what their types should be. Certainly the first operation (I hesitate to call it a "function") should take a String argument and produce something, but what should it produce? It could produce a unit (), since there is essentially no return value from printing a string. The second operation, similarly, should return a String, but it doesn't seem to require an argument.

We want both of these operations to be functions, but they are by definition not functions. The item that reads a string from the keyboard cannot be a function, as it will not return the same String every time. And if the first function simply returns () every time, there should be no problem with replacing it with a function f _ = (), due to referential transparency. But clearly this does not have the desired effect.

The RealWorld Solution edit

In a sense, the reason that these items are not functions is that they interact with the "real world." Their values depend directly on the real world. Supposing we had a type RealWorld, we might write these functions as having type:

printAString :: RealWorld -> String -> RealWorld
readAString  :: RealWorld -> (RealWorld, String)

That is, printAString takes a current state of the world and a string to print; it then modifies the state of the world in such a way that the string is now printed and returns this new value. Similarly, readAString takes a current state of the world and returns a new state of the world, paired with the String that was typed.

This would be a possible way to do IO, though it is more than somewhat unwieldy. In this style (assuming an initial RealWorld state were an argument to main), our "Name.hs" program from the section on Interactivity would look something like:

main rW =
  let rW' = printAString rW "Please enter your name: "
      (rW'',name) = readAString rW'
  in  printAString rW''
          ("Hello, " ++ name ++ ", how are you?")

This is not only hard to read, but prone to error, if you accidentally use the wrong version of the RealWorld. It also doesn't model the fact that the program below makes no sense:

main rW =
  let rW' = printAString rW "Please enter your name: "
      (rW'',name) = readAString rW'
  in  printAString rW'                 -- OOPS!
          ("Hello, " ++ name ++ ", how are you?")

In this program, the reference to rW'' on the last line has been changed to a reference to rW'. It is completely unclear what this program should do. Clearly, it must read a string in order to have a value for name to be printed. But that means that the RealWorld has been updated. However, then we try to ignore this update by using an "old version" of the RealWorld. There is clearly something wrong happening here.







Suffice it to say that doing IO operations in a pure lazy functional language is not trivial.

Actions edit

The breakthrough for solving this problem came when Phil Wadler realized that monads would be a good way to think about IO computations. In fact, monads are able to express much more than just the simple operations described above; we can use them to express a variety of constructions like concurrence, exceptions, IO, non-determinism and much more. Moreover, there is nothing special about them; they can be defined within Haskell with no special handling from the compiler (though compilers often choose to optimize monadic operations).

As pointed out before, we cannot think of things like "print a string to the screen" or "read data from a file" as functions, since they are not (in the pure mathematical sense). Therefore, we give them another name: actions. Not only do we give them a special name, we give them a special type. One particularly useful action is putStrLn, which prints a string to the screen. This action has type:

putStrLn :: String -> IO ()

As expected, putStrLn takes a string argument. What it returns is of type IO (). This means that this function is actually an action (that is what the IO means). Furthermore, when this action is evaluated (or "run") , the result will have type ().

Note

Actually, this type means that putStrLn is an action within the IO monad, but we will gloss over this for now.

You can probably already guess the type of getLine:

getLine :: IO String

This means that getLine is an IO action that, when run, will have type String.

The question immediately arises: "how do you `run' an action?". This is something that is left up to the compiler. You cannot actually run an action yourself; instead, a program is, itself, a single action that is run when the compiled program is executed. Thus, the compiler requires that the main function have type IO (), which means that it is an IO action that returns nothing. The compiled code then executes this action.

However, while you are not allowed to run actions yourself, you are allowed to combine actions. In fact, we have already seen one way to do this using the do notation (how to really do this will be revealed in the chapter Monads). Let's consider the original name program:

main = do
  hSetBuffering stdin LineBuffering
  putStrLn "Please enter your name: "
  name <- getLine
  putStrLn ("Hello, " ++ name ++ ", how are you?")

We can consider the do notation as a way to combine a sequence of actions. Moreover, the <- notation is a way to get the value out of an action. So, in this program, we're sequencing four actions: setting buffering, a putStrLn, a getLine and another putStrLn. The putStrLn action has type String -> IO (), so we provide it a String, so the fully applied action has type IO (). This is something that we are allowed to execute.

The getLine action has type IO String, so it is okay to execute it directly. However, in order to get the value out of the action, we write name <- getLine, which basically means "run getLine, and put the results in the variable called name."

Normal Haskell constructions like if/then/else and case/of can be used within the do notation, but you need to be somewhat careful. For instance, in our "guess the number" program, we have:

    do ...
       if (read guess) < num
         then do putStrLn "Too low!"
                 doGuessing num
         else if read guess > num
                then do putStrLn "Too high!"
                        doGuessing num
                else do putStrLn "You Win!"

If we think about how the if/then/else construction works, it essentially takes three arguments: the condition, the "then" branch, and the "else" branch. The condition needs to have type Bool, and the two branches can have any type, provided that they have the same type. The type of the entire if/then/else construction is then the type of the two branches.

In the outermost comparison, we have (read guess) < num as the condition. This clearly has the correct type. Let's just consider the "then" branch. The code here is:

              do putStrLn "Too low!"
                 doGuessing num

Here, we are sequencing two actions: putStrLn and doGuessing. The first has type IO (), which is fine. The second also has type IO (), which is fine. The type result of the entire computation is precisely the type of the final computation. Thus, the type of the "then" branch is also IO (). A similar argument shows that the type of the "else" branch is also IO (). This means the type of the entire if/then/else construction is IO (), which is just what we want.

Note

In this code, the last line is else do putStrLn "You Win!". This is somewhat overly verbose. In fact, else putStrLn "You Win!" would have been sufficient, since do is only necessary to sequence actions. Since we have only one action here, it is superfluous.

It is incorrect to think to yourself "Well, I already started a do block; I don't need another one," and hence write something like:

    do if (read guess) < num
         then putStrLn "Too low!"
              doGuessing num
         else ...

Here, since we didn't repeat the do, the compiler doesn't know that the putStrLn and doGuessing calls are supposed to be sequenced, and the compiler will think you're trying to call putStrLn with three arguments: the string, the function doGuessing and the integer num. It will certainly complain (though the error may be somewhat difficult to comprehend at this point).

We can write the same doGuessing function using a case statement. To do this, we first introduce the Prelude function compare, which takes two values of the same type (in the Ord class) and returns one of GT, LT, EQ, depending on whether the first is greater than, less than or equal to the second.

doGuessing num = do
  putStrLn "Enter your guess:"
  guess <- getLine
  case compare (read guess) num of
    LT -> do putStrLn "Too low!"
             doGuessing num
    GT -> do putStrLn "Too high!"
             doGuessing num
    EQ -> putStrLn "You Win!"

Here, again, the dos after the ->s are necessary on the first two options, because we are sequencing actions.

If you're used to programming in an imperative language like C or Java, you might think that return will exit you from the current function. This is not so in Haskell. In Haskell, return simply takes a normal value (for instance, one of type Int) and makes it into an action that returns the given value (for instance, the value of type IO Int). In particular, in an imperative language, you might write this function as:

void doGuessing(int num) {
  print "Enter your guess:";
  int guess = atoi(readLine());
  if (guess == num) {
    print "You win!";
    return ();
  }

  // we won't get here if guess == num
  if (guess < num) {
    print "Too low!";
    doGuessing(num);
  } else {
    print "Too high!";
    doGuessing(num);
  }
}

Here, because we have the return () in the first if match, we expect the code to exit there (and in most imperative languages, it does). However, the equivalent code in Haskell, which might look something like:

doGuessing num = do
  putStrLn "Enter your guess:"
  guess <- getLine
  case compare (read guess) num of
    EQ -> do putStrLn "You win!"
             return ()

  -- we don't expect to get here unless guess == num
  if (read guess < num)
    then do putStrLn "Too low!";
            doGuessing num
    else do putStrLn "Too high!";
            doGuessing num

will not behave as you expect. First of all, if you guess correctly, it will first print "You win!," but it won't exit, and it will check whether guess is less than num. Of course it is not, so the else branch is taken, and it will print "Too high!" and then ask you to guess again.

On the other hand, if you guess incorrectly, it will try to evaluate the case statement and get either LT or GT as the result of the compare. In either case, it won't have a pattern that matches, and the program will fail immediately with an exception.

Exercises

Write a program that asks the user for his or her name. If the name is one of Simon, John or Phil, tell the user that you think Haskell is a great programming language. If the name is Koen, tell them that you think debugging Haskell is fun (Koen Classen is one of the people who works on Haskell debugging); otherwise, tell the user that you don't know who he or she is.

Write two different versions of this program, one using if

statements, the other using a case statement.

The IO Library edit

The IO Library (available by importing the System.IO module) contains many definitions, the most common of which are listed below:





data IOMode  = ReadMode   | WriteMode
             | AppendMode | ReadWriteMode

openFile     :: FilePath -> IOMode -> IO Handle
hClose       :: Handle -> IO ()

hIsEOF       :: Handle -> IO Bool

hGetChar     :: Handle -> IO Char
hGetLine     :: Handle -> IO String
hGetContents :: Handle -> IO String

getChar      :: IO Char
getLine      :: IO String
getContents  :: IO String

hPutChar     :: Handle -> Char -> IO ()
hPutStr      :: Handle -> String -> IO ()
hPutStrLn    :: Handle -> String -> IO ()

putChar      :: Char -> IO ()
putStr       :: String -> IO ()
putStrLn     :: String -> IO ()

readFile     :: FilePath -> IO String
writeFile    :: FilePath -> String -> IO ()

bracket      ::
    IO a -> (a -> IO b) -> (a -> IO c) -> IO c

Note

The type FilePath is a type synonym for String. That is, there is no difference between FilePath and String. So, for instance, the readFile function takes a String (the file to read) and returns an action that, when run, produces the contents of that file. See the section on Synonyms for more about type synonyms.

Most of these functions are self-explanatory. The openFile and hClose functions open and close a file, respectively, using the IOMode argument as the mode for opening the file. hIsEOF tests for end-of file. hGetChar and hGetLine read a character or line (respectively) from a file. hGetContents reads the entire file. The getChar, getLine and getContents variants read from standard input. hPutChar prints a character to a file; hPutStr prints a string; and hPutStrLn prints a string with a newline character at the end. The variants without the h prefix work on standard output. The readFile and writeFile functions read an entire file without having to open it first.

The bracket function is used to perform actions safely. Consider a function that opens a file, writes a character to it, and then closes the file. When writing such a function, one needs to be careful to ensure that, if there were an error at some point, the file is still successfully closed. The bracket function makes this easy. It takes three arguments: The first is the action to perform at the beginning. The second is the action to perform at the end, regardless of whether there's an error or not. The third is the action to perform in the middle, which might result in an error. For instance, our character-writing function might look like:

writeChar :: FilePath -> Char -> IO ()
writeChar fp c =
    bracket
      (openFile fp ReadMode)
      hClose
      (\h -> hPutChar h c)

This will open the file, write the character and then close the file. However, if writing the character fails, hClose will still be executed, and the exception will be reraised afterwards. That way, you don't need to worry too much about catching the exceptions and about closing all of your handles.

A File Reading Program edit

We can write a simple program that allows a user to read and write files. The interface is admittedly poor, and it does not catch all errors (try reading a non-existent file). Nevertheless, it should give a fairly complete example of how to use IO. Enter the following code into "FileRead.hs," and compile/run:

module Main
    where

import System.IO
import Control.Exception

main = do
  hSetBuffering stdin LineBuffering
  doLoop

doLoop = do
  putStrLn "Enter a command rFN wFN or q to quit:"
  command <- getLine
  case command of
    'q':_ -> return ()
    'r':filename -> do putStrLn ("Reading " ++ filename)
                       doRead filename
                       doLoop
    'w':filename -> do putStrLn ("Writing " ++ filename)
                       doWrite filename
                       doLoop
    _            -> doLoop

doRead filename =
    bracket (openFile filename ReadMode) hClose
            (\h -> do contents <- hGetContents h
                      putStrLn "The first 100 chars:"
                      putStrLn (take 100 contents))

doWrite filename = do
  putStrLn "Enter text to go into the file:"
  contents <- getLine
  bracket (openFile filename WriteMode) hClose
          (\h -> hPutStrLn h contents)

What does this program do? First, it issues a short string of instructions and reads a command. It then performs a case switch on the command and checks first to see if the first character is a `q.' If it is, it returns a value of unit type.

Note

The return function is a function that takes a value of type a and returns an action of type IO a. Thus, the type of return () is IO ().

If the first character of the command wasn't a `q,' the program checks to see if it was an 'r' followed by some string that is bound to the variable filename. It then tells you that it's reading the file, does the read and runs doLoop again. The check for `w' is nearly identical. Otherwise, it matches _, the wildcard character, and loops to doLoop.

The doRead function uses the bracket function to make sure there are no problems reading the file. It opens a file in ReadMode, reads its contents and prints the first 100 characters (the take function takes an integer   and a list and returns the first   elements of the list).

The doWrite function asks for some text, reads it from the keyboard, and then writes it to the file specified.

Note

Both doRead and doWrite could have been made simpler by using readFile and writeFile, but they were written in the extended fashion to show how the more complex functions are used.

The only major problem with this program is that it will die if you try to read a file that doesn't already exists or if you specify some bad filename like *\^\#_@. You may think that the calls to bracket in doRead and doWrite should take care of this, but they don't. They only catch exceptions within the main body, not within the startup or shutdown functions (openFile and hClose, in these cases). We would need to catch exceptions raised by openFile, in order to make this complete. We will do this when we talk about exceptions in more detail in the section on Exceptions.

Exercises

Write a program that first asks whether the user wants to read from a file, write to a file or quit. If the user responds quit, the program should exit. If he responds read, the program should ask him for a file name and print that file to the screen (if the file doesn't exist, the program may crash). If he responds write, it should ask him for a file name and then ask him for text to write to the file, with "." signaling completion. All but the "." should be written to the file.

For example, running this program might produce:

Example:

Do you want to [read] a file, [write] a file or [quit]?
read
Enter a file name to read:
foo
...contents of foo...
Do you want to [read] a file, [write] a file or [quit]?
write
Enter a file name to write:
foo
Enter text (dot on a line by itself to end):
this is some
text for
foo
.
Do you want to [read] a file, [write] a file or [quit]?
read
Enter a file name to read:
foo
this is some
text for
foo
Do you want to [read] a file, [write] a file or [quit]?
read
Enter a file name to read:
foof
Sorry, that file does not exist.
Do you want to [read] a file, [write] a file or [quit]?
blech
I don't understand the command blech.
Do you want to [read] a file, [write] a file or [quit]?
quit
Goodbye!