Interaction. Assuming Python is configured properly, interaction should look something like the following. You can run this any way you like: in IDLE, from a shell prompt, and so on:

% python
...copyright information lines...
>>> "Hello World!"
'Hello World!'
>>>                     # Ctrl-D or Ctrl-Z to exit, or window close

Programs. Your code (i.e., module) file module1.py and shell interactions should look like:

print 'Hello module world!'

% python module1.py
Hello module world!

Again, feel free to run this other ways—by clicking its icon, by IDLE’s Edit/RunScript menu option, and so on.

Modules. The following interaction listing illustrates running a module file by importing it.

% python
>>> import module1
Hello module world!
>>>

Remember that you need to reload the module to run again without stopping and restarting the interpreter. The questions about moving the file to a different directory and importing it again is a trick question: if Python generates a module1.pyc file in the original directory, it uses that when you import the module, even if the source code file (.py) has been moved to a directory not on Python’s search path. The .pyc file is written automatically if Python has access to the source file’s directory and contains the compiled byte-code version of a module. See Part V for more on modules.

Scripts. Assuming your platform supports the #! trick, your solution will look like the following (although your #! line may need to list another path on your machine):

#!/usr/local/bin/python          (or #!/usr/bin/env python)
print 'Hello module world!'

% chmod +x module1.py

% module1.py
Hello module world!

Errors. The interaction below demonstrates the sort of error messages you get when you complete this exercise. Really, you’re triggering Python exceptions; the default exception handling behavior terminates the running Python program and prints an error message and stack trace on the screen. The stack trace shows where you were in a program when the exception occurred. In Part VII, you will learn that you can catch exceptions using try statements and process them arbitrarily; you’ll also see that Python includes a full-blown source code debugger for special error detection requirements. For now, notice that Python gives meaningful messages when programming errors occur (instead of crashing silently):

% python
>>> 1 / 0
Traceback (innermost last):
  File "<stdin>", line 1, in ?
ZeroDivisionError: integer division or modulo
>>>
>>> x
Traceback (innermost last):
  File "<stdin>", line 1, in ?
NameError: x

Breaks. When you type this code:

L = [1, 2]
L.append(L)

you create a cyclic data structure in Python. In Python releases before Version 1.5.1, the Python printer wasn’t smart enough to detect cycles in objects, and it would print an unending stream of [1, 2, [1, 2, [1, 2, [1, 2,—and so on, until you hit the break key combination on your machine (which, technically, raises a keyboard-interrupt exception that prints a default message). Beginning with Python Version 1.5.1, the printer is clever enough to detect cycles and prints [[...]] instead.

The reason for the cycle is subtle and requires information you will gain in Part II. But in short, assignment in Python always generates references to objects (which you can think of as implicitly followed pointers). When you run the first assignment above, the name L becomes a named reference to a two-item list object. Python lists are really arrays of object references, with an append method that changes the array in place by tacking on another object reference. Here, the append call adds a reference to the front of L at the end of L, which leads to the cycle illustrated in Figure B-1. Believe it or not, cyclic data structures can sometimes be useful (but not when printed!).

The basics. Here are the sort of results you should get, along with a few comments about their meaning. Note that ; is used in a few of these to squeeze more than one statement on a single line; the ; is a statement separator.

Numbers

>>> 2 ** 16              # 2 raised to the power 16
65536
>>> 2 / 5, 2 / 5.0       # Integer / truncates, float / doesn't
(0, 0.40000000000000002)

Strings

>>> "spam" + "eggs"      # Concatenation
'spameggs'
>>> S = "ham"
>>> "eggs " + S
'eggs ham'
>>> S * 5                # Repetition
'hamhamhamhamham'
>>> S[:0]                # An empty slice at the front--[0:0]
''
>>> "green %s and %s" % ("eggs", S)  # Formatting
'green eggs and ham'

Tuples

>>> ('x',)[0]                        # Indexing a single-item tuple
'x'
>>> ('x', 'y')[1]                    # Indexing a 2-item tuple
'y'

Lists

>>> L = [1,2,3] + [4,5,6]            # List operations
>>> L, L[:], L[:0], L[-2], L[-2:]
([1, 2, 3, 4, 5, 6], [1, 2, 3, 4, 5, 6], [  ], 5, [5, 6])
>>> ([1,2,3]+[4,5,6])[2:4]
[3, 4]
>>> [L[2], L[3]]                       # Fetch from offsets; store in a list
[3, 4]
>>> L.reverse(  ); L                   # Method: reverse list in-place
[6, 5, 4, 3, 2, 1]
>>> L.sort(  ); L                      # Method: sort list in-place
[1, 2, 3, 4, 5, 6]
>>> L.index(4)                         # Method: offset of first 4 (search)
3

Dictionaries

>>> {'a':1, 'b':2}['b']              # Index a dictionary by key.
2
>>> D = {'x':1, 'y':2, 'z':3}
>>> D['w'] = 0                       # Create a new entry.
>>> D['x'] + D['w']
1
>>> D[(1,2,3)] = 4                   # A tuple used as a key 
                     (immutable)
>>> D
{'w': 0, 'z': 3, 'y': 2, (1, 2, 3): 4, 'x': 1}
>>> D.keys(  ), D.values(  ), D.has_key((1,2,3))          # Methods
(['w', 'z', 'y', (1, 2, 3), 'x'], [0, 3, 2, 4, 1], 1)

Empties

>>> [[  ]], ["",[  ],(  ),{  },None]         # Lots of nothings: empty objects
([[  ]], ['', [  ], (  ), {  }, None])

Indexing and slicing. Indexing out-of-bounds (e.g., L[4]) raises an error; Python always checks to make sure that all offsets are within the bounds of a sequence.

On the other hand, slicing out of bounds (e.g., L[-1000:100]) works, because Python scales out-of-bounds slices so that they always fit (they’re set to zero and the sequence length, if required).

Extracting a sequence in reverse—with the lower bound greater than the higher bound (e.g., L[3:1])—doesn’t really work. You get back an empty slice ([ ]), because Python scales the slice limits to make sure that the lower bound is always less than or equal to the upper bound (e.g., L[3:1] is scaled to L[3:3], the empty insertion point at offset 3). Python slices are always extracted from left to right, even if you use negative indexes (they are first converted to positive indexes by adding the length). Note that Python 2.3 three-limit slices modify this behavior somewhat: L[3:1:-1] does extract from right to left.

>>> L = [1, 2, 3, 4]
>>> L[4]
Traceback (innermost last):
  File "<stdin>", line 1, in ?
IndexError: list index out of range
>>> L[-1000:100]
[1, 2, 3, 4]
>>> L[3:1]
[  ]
>>> L
[1, 2, 3, 4]
>>> L[3:1] = ['?']
>>> L
[1, 2, 3, '?', 4]

Indexing, slicing, and del. Your interaction with the interpreter should look something like the following code. Note that assigning an empty list to an offset stores an empty list object there, but assigning an empty list to a slice deletes the slice. Slice assignment expects another sequence, or you’ll get a type error; it inserts items inside the sequence assigned, not the sequence itself:

>>> L = [1,2,3,4]
>>> L[2] = [  ]
>>> L
[1, 2, [  ], 4]
>>> L[2:3] = [  ]
>>> L
[1, 2, 4]
>>> del L[0]
>>> L
[2, 4]
>>> del L[1:]
>>> L
[2]
>>> L[1:2] = 1
Traceback (innermost last):
  File "<stdin>", line 1, in ?
TypeError: illegal argument type for built-in operation

Tuple assignment. The values of X and Y are swapped. When tuples appear on the left and right of an assignment symbol (=), Python assigns objects on the right to targets on the left, according to their positions. This is probably easiest to understand by noting that targets on the left aren’t a real tuple, even though they look like one; they are simply a set of independent assignment targets. The items on the right are a tuple, which get unpacked during the assignment (the tuple provides the temporary assignment needed to achieve the swap effect).

>>> X = 'spam'
>>> Y = 'eggs'
>>> X, Y = Y, X
>>> X
'eggs'
>>> Y
'spam'

Dictionary keys. Any immutable object can be used as a dictionary key—integers, tuples, strings, and so on. This really is a dictionary, even though some of its keys look like integer offsets. Mixed type keys work fine too.

>>> D = {  }
>>> D[1] = 'a'
>>> D[2] = 'b'
>>> D[(1, 2, 3)] = 'c'
>>> D
{1: 'a', 2: 'b', (1, 2, 3): 'c'}

Dictionary indexing. Indexing a nonexistent key (D['d']) raises an error; assigning to a nonexistent key (D['d']='spam') creates a new dictionary entry. On the other hand, out-of-bounds indexing for lists raises an error too, but so do out-of-bounds assignments. Variable names work like dictionary keys; they must have already been assigned when referenced, but are created when first assigned. In fact, variable names can be processed as dictionary keys if you wish (they’re made visible in module namespace or stack-frame dictionaries).

>>> D = {'a':1, 'b':2, 'c':3}
>>> D['a']
1
>>> D['d']
Traceback (innermost last):
  File "<stdin>", line 1, in ?
KeyError: d
>>> D['d'] = 4
>>> D
{'b': 2, 'd': 4, 'a': 1, 'c': 3}
>>>
>>> L = [0,1]
>>> L[2]
Traceback (innermost last):
  File "<stdin>", line 1, in ?
IndexError: list index out of range
>>> L[2] = 3
Traceback (innermost last):
  File "<stdin>", line 1, in ?
IndexError: list assignment index out of range

Generic operations. Question answers:

String indexing. Since strings are collections of one-character strings, every time you index a string, you get back a string, which can be indexed again. S[0][0][0][0][0] just keeps indexing the first character over and over. This generally doesn’t work for lists (lists can hold arbitrary objects), unless the list contains strings.

>>> S = "spam"
>>> S[0][0][0][0][0]
's'
>>> L = ['s', 'p']
>>> L[0][0][0]
's'

Immutable types. Either of the solutions below work. Index assignment doesn’t, because strings are immutable.

>>> S = "spam"
>>> S = S[0] + 'l' + S[2:]
>>> S
'slam'
>>> S = S[0] + 'l' + S[2] + S[3]
>>> S
'slam'

Nesting. Here is a sample:

>>> me = {'name':('mark', 'e', 'lutz'), 'age':'?', 'job':'engineer'}
>>> me['job']
'engineer'
>>> me['name'][2]
'lutz'

Files. Here’s one way to create and read back a text file in Python (ls is a Unix command; use dir on Windows):

#File: maker.py
file = open('myfile.txt', 'w')
file.write('Hello file world!
')          # Or: open(  ).write(  )
file.close(  )                             # close not always needed

#File: reader.py
file = open('myfile.txt', 'r')
print file.read(  )                        # Or print open(  ).read(  )

% python maker.py
% python reader.py
Hello file world!

% ls -l myfile.txt
-rwxrwxrwa   1 0        0             19 Apr 13 16:33 myfile.txt

The dir function revisited: Here’s what you get for lists; dictionaries do the same (but with different method names). Note that the dir result expanded in Python 2.2—you’ll see a large set of additional underscore names that implement expression operators, and support the subclassing in Part VI. The __methods__ attribute disappeared in 2.2 as well, because it wasn’t consistently implemented—use dir to to fetch attribute lists today instead:

>>> [  ].__methods__
['append', 'count', 'index', 'insert', 'remove', 'reverse', 'sort',...]
>>> dir([  ])
['append', 'count', 'index', 'insert', 'remove', 'reverse', 'sort',...]

Coding basic loops. As you work through this exercise, you’ll wind up with code that looks like the following:

>>> S = 'spam'
>>> for c in S:
...     print ord(c)
...
115
112
97
109

>>> x = 0
>>> for c in S: x = x + ord(c)        # Or: x += ord(c)
...
>>> x
433

>>> x = [  ]
>>> for c in S: x.append(ord(c))
...
>>> x
[115, 112, 97, 109]

>>> map(ord, S)
[115, 112, 97, 109]

Backslash characters. The example prints the bell character (a) 50 times; assuming your machine can handle it, and when run outside of IDLE, you may get a series of beeps (or one long tone, if your machine is fast enough). Hey—we warned you.

Sorting dictionaries. Here’s one way to work through this exercise (see Chapter 6 if this doesn’t make sense). Remember, you really do have to split the keys and sort calls up like this, because sort returns None. In Python 2.2, you can iterate through dictionary keys directly without calling keys (e.g., for key in D:), but the keys list will not be sorted like it is by this code:

>>> D = {'a':1, 'b':2, 'c':3, 'd':4, 'e':5, 'f':6, 'g':7}
>>> D
{'f': 6, 'c': 3, 'a': 1, 'g': 7, 'e': 5, 'd': 4, 'b': 2}
>>>
>>> keys = D.keys(  )
>>> keys.sort(  )
>>> for key in keys:
...     print key, '=>', D[key]
...
a => 1
b => 2
c => 3
d => 4
e => 5
f => 6
g => 7

Program logic alternatives. Here’s sample code for the solutions. Your results may vary a bit; this exercise is mostly designed to get you playing with code alternatives, so anything reasonable gets full credit:

L = [1, 2, 4, 8, 16, 32, 64]
X = 5

i = 0
while i < len(L):
    if 2 ** X == L[i]:
        print 'at index', i
        break
    i = i+1
else:
    print X, 'not found'
        

L = [1, 2, 4, 8, 16, 32, 64]
X = 5

for p in L:
    if (2 ** X) == p:
        print (2 ** X), 'was found at', L.index(p)
        break
else:
    print X, 'not found'


L = [1, 2, 4, 8, 16, 32, 64]
X = 5

if (2 ** X) in L:
    print (2 ** X), 'was found at', L.index(2 ** X)
else:
    print X, 'not found'
        

X = 5
L = [  ]
for i in range(7): L.append(2 ** i)
print L

if (2 ** X) in L:
    print (2 ** X), 'was found at', L.index(2 ** X)
else:
    print X, 'not found'
        

X = 5
L = map(lambda x: 2**x, range(7))
print L

if (2 ** X) in L:
    print (2 ** X), 'was found at', L.index(2 ** X)
else:
    print X, 'not found'

The basics. There’s not much to this one, but notice that using print (and hence your function) is technically a polymorphic operation, which does the right thing for each type of object:

% python
>>> def func(x): print x
...
>>> func("spam")
spam
>>> func(42)
42
>>> func([1, 2, 3])
[1, 2, 3]
>>> func({'food': 'spam'})
{'food': 'spam'}

Arguments. Here’s a sample solution. Remember that you have to use print to see results in the test calls, because a file isn’t the same as code typed interactively; Python doesn’t normally echo the results of expression statements in files.

def adder(x, y):
    return x + y

print adder(2, 3)
print adder('spam', 'eggs')
print adder(['a', 'b'], ['c', 'd'])

% python mod.py
5
spameggs
['a', 'b', 'c', 'd']

varargs. Two alternative adder functions are shown in the following file, adders.py. The hard part here is figuring out how to initialize an accumulator to an empty value of whatever type is passed in. The first solution, uses manual type testing to look for an integer and an empty slice of the first argument (assumed to be a sequence) otherwise. The second solution, uses the first argument to initialize and scan items 2 and beyond, much like one of the min function variants shown in Chapter 13.

The second solution is better. Both of these assume all arguments are the same type and neither works on dictionaries; as we saw in Part II, + doesn’t work on mixed types or dictionaries. We could add a type test and special code to add dictionaries too, but that’s extra credit.

def adder1(*args):
    print 'adder1',
    if type(args[0]) == type(0):    # Integer?
         sum = 0                    # Init to zero.
    else:                           # else sequence:
         sum = args[0][:0]          # Use empty slice of arg1.
    for arg in args:
        sum = sum + arg
    return sum

def adder2(*args):
    print 'adder2',
    sum = args[0]               # Init to arg1.
    for next in args[1:]:
        sum = sum + next        # Add items 2..N.
    return sum

for func in (adder1, adder2):
    print func(2, 3, 4)
    print func('spam', 'eggs', 'toast')
    print func(['a', 'b'], ['c', 'd'], ['e', 'f'])

% python adders.py
adder1 9
adder1 spameggstoast
adder1 ['a', 'b', 'c', 'd', 'e', 'f']
adder2 9
adder2 spameggstoast
adder2 ['a', 'b', 'c', 'd', 'e', 'f']

Keywords. Here is our solution to the first part of this exercise (file mod.py). To iterate over keyword arguments, use a **args form in the function header and use a loop like: for x in args.keys( ): use args[x].

def adder(good=1, bad=2, ugly=3):
    return good + bad + ugly

print adder(  )
print adder(5)
print adder(5, 6)
print adder(5, 6, 7)
print adder(ugly=7, good=6, bad=5)

% python mod.py
6
10
14
18
18

and 6. Here are our solutions to exercises 5 and 6 (file dicts.py). These are just coding exercises, though, because Python 1.5 added dictionary methods, to do things like copying and adding (merging) dictionaries: D.copy( ), and D1.update(D2). (See Python’s library manual or the Python Pocket Reference for more details). X[:] doesn’t work for dictionaries, since they’re not sequences (see Chapter 6 for details). Also remember that if we assign (e = d) rather than copy, we generate a reference to a shared dictionary object; changing d changes e, too.

def copyDict(old):
    new = {  }
    for key in old.keys(  ):
        new[key] = old[key]
    return new

def addDict(d1, d2):
    new = {  }
    for key in d1.keys(  ):
        new[key] = d1[key]
    for key in d2.keys(  ):
        new[key] = d2[key]
    return new

% python
>>> from dicts import *
>>> d = {1:1, 2:2}
>>> e = copyDict(d)
>>> d[2] = '?'
>>> d
{1: 1, 2: '?'}
>>> e
{1: 1, 2: 2}

>>> x = {1:1}
>>> y = {2:2}
>>> z = addDict(x, y)
>>> z
{1: 1, 2: 2}

More argument matching examples. Here is the sort of interaction you should get, along with comments that explain the matching that goes on:

def f1(a, b): print a, b             # Normal args

def f2(a, *b): print a, b            # Positional varargs

def f3(a, **b): print a, b           # Keyword varargs

def f4(a, *b, **c): print a, b, c    # Mixed modes

def f5(a, b=2, c=3): print a, b, c   # Defaults

def f6(a, b=2, *c): print a, b, c    # Defaults and positional varargs


% python
>>> f1(1, 2)                  # Matched by position (order matters)
1 2
>>> f1(b=2, a=1)              # Matched by name (order doesn't matter)
1 2

>>> f2(1, 2, 3)               # Extra positionals collected in a tuple
1 (2, 3)

>>> f3(1, x=2, y=3)           # Extra keywords collected in a dictionary
1 {'x': 2, 'y': 3}

>>> f4(1, 2, 3, x=2, y=3)     # Extra of both kinds
1 (2, 3) {'x': 2, 'y': 3}

>>> f5(1)                     # Both defaults kick in.
1 2 3
>>> f5(1, 4)                  # Only one default used
1 4 3

>>> f6(1)                     # One argument: matches "a"
1 2 (  )
>>> f6(1, 3, 4)               # Extra positional collected
1 3 (4,)

Primes revisited. Below is the primes example wrapped up in a function and module (file primes.py) so it can be run multiple times. We added an if test to trap negatives, 0, and 1. We also changed / to // to make this immune from the Python 3.0 / “true” division changes we studied in Chapter 4, and support floating-point numbers. The // operator works in both the current and future division scheme, but the future / operator fails (uncomment the from and change // to / to see the differences in 2.2 and 3.0).

#from __future__ import division

def prime(y):
    if y <= 1:                              # For some y > 1
        print y, 'not prime'
    else:
        x = y // 2                          # Future / fails
        while x > 1:
            if y % x == 0:                  # No remainder?
                print y, 'has factor', x
                break                       # Skip else.
            x -= 1
        else:
            print y, 'is prime'

prime(13); prime(13.0)
prime(15); prime(15.0)
prime(3);  prime(2)
prime(1);  prime(-3)

Here is the module in action; the // operator allows it to works for floating-point numbers too, even though it perhaps should not:

% python primes.py
13 is prime
13.0 is prime
15 has factor 5
15.0 has factor 5.0
3 is prime
2 is prime
1 not prime
-3 not prime

This function still isn’t very reusable yet—it could return values instead of printing—but it’s enough to run experiments. It’s also still not a strict mathematical prime (floating-points work), and is still inefficient. Improvements are left as exercises for more mathematically-minded readers. Hint: a for loop over range(y, 1, -1) may be a bit quicker than the while (in fact, it’s roughly twice as fast in 2.2), but the algorithm is the real bottleneck here. To time alternatives, use the built-in time module, and coding patterns like those used in this general function-call timer (see the library manual for details):

def timer(reps, func, *args):
    import time
    start = time.clock(  )
    for i in xrange(reps):
        apply(func, args)
    return time.clock(  ) - start

List comprehensions. Here is the sort of code you should write; we may have a preference, but we’re not telling:

>>> values = [2, 4, 9, 16, 25]
>>> import math

>>> res = [  ]
>>> for x in values: res.append(math.sqrt(x))
...
>>> res
[1.4142135623730951, 2.0, 3.0, 4.0, 5.0]

>>> map(math.sqrt, values)
[1.4142135623730951, 2.0, 3.0, 4.0, 5.0]

>>> [math.sqrt(x) for x in values]
[1.4142135623730951, 2.0, 3.0, 4.0, 5.0]

Basics, import. This one is simpler than you may think. When you’re done, your file and interaction should look close to the following code (file mymod.py); remember that Python can read a whole file into a string or lines list, and the len built-in returns the length of strings and lists:

def countLines(name):
    file = open(name, 'r')
    return len(file.readlines(  ))

def countChars(name):
    return len(open(name, 'r').read(  ))

def test(name):                                  # Or pass file object
    return countLines(name), countChars(name)    # Or return a dictionary

% python
>>> import mymod
>>> mymod.test('mymod.py')
(10, 291)

On Unix, you can verify your output with a wc command; on Windows, right-click on your file to views its properties. But note that your script may report fewer characters than Windows does—for portability, Python converts Windows line-end markers to , thereby dropping one byte (character) per line. To match byte counts with Windows exactly, you have to open in binary mode (rb) or add back the number of lines.

Incidentally, to do the “ambitious” part (passing in a file object, so you only open the file once), you’ll probably need to use the seek method of the built-in file object. We didn’t cover it in the text, but it works just like C’s fseek call (and calls it behind the scenes): seek resets the current position in the file to an offset passed in. After a seek, future input/output operations are relative to the new position. To rewind to the start of a file without closing and reopening, call file.seek(0); the file read methods all pick up at the current position in the file, so you need to rewind to reread. Here’s what this tweak would look like:

def countLines(file):
    file.seek(0)                      # Rewind to start of file.
    return len(file.readlines(  ))

def countChars(file): 
    file.seek(0)                      # Ditto (rewind if needed)
    return len(file.read(  ))

def test(name):
    file = open(name, 'r')                       # Pass file object.
    return countLines(file), countChars(file)    # Open file only once.

>>> import mymod2
>>> mymod2.test("mymod2.py")
(11, 392)

from/from*. Here’s the from* part. Replace * with countChars to do the rest.

% python
>>> from mymod import *
>>> countChars("mymod.py")
291

__main__. If you code it properly, it works in either mode (program run or module import):

def countLines(name):
    file = open(name, 'r')
    return len(file.readlines(  ))

def countChars(name):
    return len(open(name, 'r').read(  ))

def test(name):                                  # Or pass file object
    return countLines(name), countChars(name)    # Or return a dictionary

if __name__ == '__main__':
    print test('mymod.py')

% python mymod.py
(13, 346)

Nested imports. Here is our solution (file myclient.py):

from mymod import countLines, countChars
print countLines('mymod.py'), countChars('mymod.py')

% python myclient.py
13 346

As for the rest of this one: mymod’s functions are accessible (that is, importable) from the top level of myclient, since from simply assigns to names in the importer (it works almost as though mymod’s defs appeared in myclient). For example, another file can say this:

import myclient
myclient.countLines(...)

from myclient import countChars
countChars(...)

If myclient used import instead of from, you’d need to use a path to get to the functions in mymod through myclient:

import myclient
myclient.mymod.countLines(...)

from myclient import mymod
mymod.countChars(...)

In general, you can define collector modules that import all the names from other modules, so they’re available in a single convenience module. Using the following code, you wind up with three different copies of name somename: mod1.somename, collector.somename, and __main__.somename; all three share the same integer object initially, and only the name somename exists at the interactive prompt as is:

#File: mod1.py
somename = 42

#File: collector.py
from mod1 import *       # Collect lots of names here.
from mod2 import *       # from assigns to my names.
from mod3 import *

>>> from collector import somename

Package imports. For this, we put the mymod.py solution file listed for exercise 3 into a directory package. The following is what we did to set up the directory and its required __init__.py file in a Windows console interface; you’ll need to interpolate for other platforms (e.g., use mv and vi instead of move and edit). This works in any directory (we just happened to run our commands in Python’s install directory), and you can do some of this from a file explorer GUI, too.

When we were done, we had a mypkg subdirectory, which contained files __init__.py and mymod.py. You need an __init__.py in the mypkg directory, but not in its parent; mypkg is located in the home directory component of the module search path. Notice how a print statement coded in the directory’s initialization file only fires the first time it is imported, not the second:

C:python22> mkdir mypkg
C:Python22> move mymod.py mypkgmymod.py
C:Python22> edit mypkg\__init__.py
...coded a print statement...

C:Python22> python
>> import mypkg.mymod
initializing mypkg
>>> mypkg.mymod.countLines('mypkgmymod.py')
13
>>> from mypkg.mymod import countChars
>>> countChars('mypkgmymod.py')
346

Reload. This exercise just asks you to experiment with changing the changer.py example in the book, so there’s nothing to show here.

Circular imports. The short story is that importing recur2 first works, because the recursive import then happens at the import in recur1, not at a from in recur2.

The long story goes like this: importing recur2 first works, because the recursive import from recur1 to recur2 fetches recur2 as a whole, instead of getting specific names. recur2 is incomplete when imported from recur1, but because it uses import instead of from, you’re safe: Python finds and returns the already created recur2 module object and continues to run the rest of recur1 without a glitch. When the recur2 import resumes, the second from finds name Y in recur1 (it’s been run completely), so no error is reported. Running a file as a script is not the same as importing it as a module; these cases are the same as running the first import or from in the script interactively. For instance, running recur1 as a script is the same as importing recur2 interactively, since recur2 is the first module imported in recur1.

Inheritance. Here’s the solution code for this exercise (file adder.py), along with some interactive tests. The __add__ overload has to appear only once, in the superclass, since it invokes type-specific add methods in subclasses.

class Adder:
    def add(self, x, y):
        print 'not implemented!'
    def __init__(self, start=[  ]):
        self.data = start
    def __add__(self, other):                # Or in subclasses?
        return self.add(self.data, other)    # Or return type?

class ListAdder(Adder):
    def add(self, x, y):
        return x + y

class DictAdder(Adder):
    def add(self, x, y):
        new = {  }
        for k in x.keys(  ): new[k] = x[k]
        for k in y.keys(  ): new[k] = y[k]
        return new

% python
>>> from adder import *
>>> x = Adder(  )
>>> x.add(1, 2)
not implemented!
>>> x = ListAdder(  )
>>> x.add([1], [2])
[1, 2]
>>> x = DictAdder(  )
>>> x.add({1:1}, {2:2})
{1: 1, 2: 2}

>>> x = Adder([1])
>>> x + [2]
not implemented!
>>>
>>> x = ListAdder([1])
>>> x + [2]
[1, 2]
>>> [2] + x
Traceback (innermost last):
  File "<stdin>", line 1, in ?
TypeError: __add__ nor __radd__ defined for these operands

Notice in the last test that you get an error for expressions where a class instance appears on the right of a +; if you want to fix this, use __radd__ methods as described in Section 21.4 in Chapter 21.

If you are saving a value in the instance anyhow, you might as well rewrite the add method to take just one argument, in the spirit of other examples in Part VI:

class Adder:
    def __init__(self, start=[  ]):
        self.data = start
    def __add__(self, other):        # Pass a single argument.
        return self.add(other)           # The left side is in self.
    def add(self, y):
        print 'not implemented!'

class ListAdder(Adder):
    def add(self, y):
        return self.data + y

class DictAdder(Adder):
    def add(self, y):
        pass  # Change me to use self.data instead of x.

x = ListAdder([1,2,3])
y = x + [4,5,6]
print y               # Prints [1, 2, 3, 4, 5, 6]

Because values are attached to objects rather than passed around, this version is arguably more object-oriented. And once you’ve gotten to this point, you’ll probably see that you could get rid of add altogether, and simply define type-specific __add__ methods in the two subclasses.

Operator overloading. The solution code (file mylist.py) uses a few operator overload methods we didn’t say much about, but they should be straightforward to understand. Copying the initial value in the constructor is important, because it may be mutable; you don’t want to change or have a reference to an object that’s possibly shared somewhere outside the class. The __getattr__ method routes calls to the wrapped list. For hints on an easier way to code this as of Python 2.2, see Section 23.1.2 in Chapter 23.

class MyList:
    def __init__(self, start):
        #self.wrapped = start[:]           # Copy start: no side effects
        self.wrapped = [  ]                  # Make sure it's a list here.
        for x in start: self.wrapped.append(x)
    def __add__(self, other):
        return MyList(self.wrapped + other)
    def __mul__(self, time):
        return MyList(self.wrapped * time)
    def __getitem__(self, offset):
        return self.wrapped[offset]
    def __len__(self):
        return len(self.wrapped)
    def __getslice__(self, low, high):
        return MyList(self.wrapped[low:high])
    def append(self, node):
        self.wrapped.append(node)
    def __getattr__(self, name):       # Other members: sort/reverse/etc
        return getattr(self.wrapped, name)
    def __repr__(self):
        return `self.wrapped`

if __name__ == '__main__':
    x = MyList('spam')
    print x
    print x[2]
    print x[1:]
    print x + ['eggs']
    print x * 3
    x.append('a')
    x.sort(  )
    for c in x: print c,

% python mylist.py
['s', 'p', 'a', 'm']
a
['p', 'a', 'm']
['s', 'p', 'a', 'm', 'eggs']
['s', 'p', 'a', 'm', 's', 'p', 'a', 'm', 's', 'p', 'a', 'm']
a a m p s

Note that it’s important to copy the start value by appending instead of slicing here, because the result may other wise not be a true list, and so would not respond to expected list methods such as append (e.g., slicing a string returns another string, not a list). You would be able to copy a MyList start value by slicing, because its class overloads the slicing operation and provides the expected list interface. You need to avoid sliced-based copying for things such as strings, however.

Subclassing. Our solution (mysub.py) appears below. Your solution should be similar.

from mylist import MyList

class MyListSub(MyList):
    calls = 0                                 # Shared by instances

    def __init__(self, start):
        self.adds = 0                         # Varies in each instance
        MyList.__init__(self, start)

    def __add__(self, other):
        MyListSub.calls = MyListSub.calls + 1   # Class-wide counter
        self.adds = self.adds + 1               # Per instance counts
        return MyList.__add__(self, other)

    def stats(self):
        return self.calls, self.adds                  # All adds, my adds

if __name__ == '__main__':
    x = MyListSub('spam')
    y = MyListSub('foo')
    print x[2]
    print x[1:]
    print x + ['eggs']
    print x + ['toast']
    print y + ['bar']
    print x.stats(  )

% python mysub.py
a
['p', 'a', 'm']
['s', 'p', 'a', 'm', 'eggs']
['s', 'p', 'a', 'm', 'toast']
['f', 'o', 'o', 'bar']
(3, 2)

Metaclass methods. We worked through this exercise as follows. Notice that operators try to fetch attributes through __getattr__ too; you need to return a value to make them work.

>>> class Meta:
...     def __getattr__(self, name):        
...         print 'get', name
...     def __setattr__(self, name, value):
...         print 'set', name, value
...
>>> x = Meta(  )
>>> x.append
get append
>>> x.spam = "pork"
set spam pork
>>>
>>> x + 2
get __coerce__
Traceback (innermost last):
  File "<stdin>", line 1, in ?
TypeError: call of non-function
>>>
>>> x[1]
get __getitem__
Traceback (innermost last):
  File "<stdin>", line 1, in ?
TypeError: call of non-function

>>> x[1:5]
get __len__
Traceback (innermost last):
  File "<stdin>", line 1, in ?
TypeError: call of non-function

Set objects. Here’s the sort of interaction you should get. Comments explain which methods are called.

% python
>>> from setwrapper import Set
>>> x = Set([1,2,3,4])          # Runs __init__
>>> y = Set([3,4,5])

>>> x & y                       # __and__, intersect, then __repr__
Set:[3, 4]
>>> x | y                       # __or__, union, then __repr__
Set:[1, 2, 3, 4, 5]

>>> z = Set("hello")            # __init__ removes duplicates.
>>> z[0], z[-1]                 # __getitem__ 
('h', 'o')

>>> for c in z: print c,        # __getitem__ 
...
h e l o
>>> len(z), z                   # __len__, __repr__
(4, Set:['h', 'e', 'l', 'o'])

>>> z & "mello", z | "mello"
(Set:['e', 'l', 'o'], Set:['h', 'e', 'l', 'o', 'm'])

Our solution to the multiple-operand extension subclass looks like the class below (file multiset.py). It only needs to replace two methods in the original set. The class’s documentation string explains how it works.

from setwrapper import Set

class MultiSet(Set):
    """
    inherits all Set names, but extends intersect
    and union to support multiple operands; note
    that "self" is still the first argument (stored
    in the *args argument now); also note that the
    inherited & and | operators call the new methods
    here with 2 arguments, but processing more than 
    2 requires a method call, not an expression:
    """

    def intersect(self, *others):
        res = [  ]
        for x in self:                     # Scan first sequence
            for other in others:           # for all other args.
                if x not in other: break   # Item in each one?
            else:                          # No: break out of loop
                res.append(x)              # Yes: add item to end
        return Set(res)

    def union(*args):                      # self is args[0].
        res = [  ]
        for seq in args:                   # For all args
            for x in seq:                  # For all nodes
                if not x in res:
                    res.append(x)          # Add new items to result.
        return Set(res)

Your interaction with the extension will be something along the following lines. Note that you can intersect by using & or calling intersect, but must call intersect for three or more operands; & is a binary (two-sided) operator. Also note that we could have called MutiSet simply Set to make this change more transparent if we used setwrapper.Set to refer to the original within multiset:

>>> from multiset import *
>>> x = MultiSet([1,2,3,4])
>>> y = MultiSet([3,4,5])
>>> z = MultiSet([0,1,2])

>>> x & y, x | y                               # Two operands
(Set:[3, 4], Set:[1, 2, 3, 4, 5])

>>> x.intersect(y, z)                          # Three operands
Set:[  ]
>>> x.union(y, z)
Set:[1, 2, 3, 4, 5, 0]

>>> x.intersect([1,2,3], [2,3,4], [1,2,3])     # Four operands 
Set:[2, 3]
>>> x.union(range(10))                         # non-MultiSets work too.
Set:[1, 2, 3, 4, 0, 5, 6, 7, 8, 9]

Class tree links. Below is the way we changed the Lister class, and a rerun of the test to show its format. To display inherited class attributes too, you’d need to do something like what the attrnames method currently does, but recursively, at each class reached by climbing __bases__ links. Because dir includes inherited attributes in Python 2.2, you might also simply loop through its result: say for x in dir(self) and use getattr(self,x). This won’t directly help, if you wish to represent the class tree’s structure in your display like the classtree.py example in Chapter 21.

class Lister:
    def __repr__(self):
        return ("<Instance of %s(%s), address %s:
%s>" %
                          (self.__class__.__name__,   # My class's name
                           self.supers(  ),              # My class's supers
                           id(self),                     # My address
                           self.attrnames(  )) )         # name=value list
    def attrnames(self):
        ...unchanged...
    def supers(self):
        result = ""
        first = 1
        for super in self.__class__.__bases__:   # One level up from class
            if not first:
                result = result + ", "
            first = 0
            result = result + super.__name__      # name, not repr(super) 
        return result

C:pythonexamples> python testmixin.py
<Instance of Sub(Super, Lister), address 7841200:
        name data3=42
        name data2=eggs
        name data1=spam
>

Composition. Our solution is below (file lunch.py), with comments from the description mixed in with the code. This is one case where it’s probably easier to express a problem in Python than it is in English.

class Lunch:
    def __init__(self):            # Make/embed Customer and Employee.
        self.cust = Customer(  )
        self.empl = Employee(  )
    def order(self, foodName):      # Start a Customer order simulation.
        self.cust.placeOrder(foodName, self.empl)
    def result(self):               # Ask the Customer about its Food.
        self.cust.printFood(  )

class Customer:
    def __init__(self):                         # Initialize my food to None.
        self.food = None
    def placeOrder(self, foodName, employee):  # Place order with Employee.
        self.food = employee.takeOrder(foodName)
    def printFood(self):                       # Print the name of my food.
        print self.food.name

class Employee:
    def takeOrder(self, foodName):    # Return a Food, with requested name.
        return Food(foodName)

class Food:
    def __init__(self, name):          # Store food name.
        self.name = name

if __name__ == '__main__':
    x = Lunch(  )                       # Self-test code
    x.order('burritos')                 # If run, not imported
    x.result(  )
    x.order('pizza')
    x.result(  )

% python lunch.py
burritos
pizza

Zoo Animal Hierarchy. Here is the way we coded the taxonomy on Python (file zoo.py); it’s artificial, but the general coding pattern applies to many real structures—from GUIs to employee databases. Notice that the self.speak reference in Animal triggers an independent inheritance search, which finds speak in a subclass. Test this interactively per the exercise description. Try extending this hierarchy with new classes, and making instances of various classes in the tree.

class Animal:
    def reply(self):   self.speak(  )        # Back to subclass
    def speak(self):   print 'spam'          # Custom message

class Mammal(Animal):
    def speak(self):   print 'huh?'

class Cat(Mammal):
    def speak(self):   print 'meow'

class Dog(Mammal):
    def speak(self):   print 'bark'

class Primate(Mammal):
    def speak(self):   print 'Hello world!'

class Hacker(Primate): pass                # Inherit from Primate.

The Dead Parrot Sketch. Here’s how we implemented this one (file parrot.py). Notice how the line method in the Actor superclass works: by accessing self attributes twice, it sends Python back to the instance twice, and hence invokes two inheritance searches—self.name and self.says( ) find information in the specific subclasses.

class Actor:
    def line(self): print self.name + ':', `self.says(  )`

class Customer(Actor):
    name = 'customer'
    def says(self): return "that's one ex-bird!"

class Clerk(Actor):
    name = 'clerk'
    def says(self): return "no it isn't..."

class Parrot(Actor):
    name = 'parrot'
    def says(self): return None

class Scene:
    def __init__(self):
        self.clerk    = Clerk(  )       # Embed some instances.
        self.customer = Customer(  )    # Scene is a composite.
        self.subject  = Parrot(  )

    def action(self):
        self.customer.line(  )          # Delegate to embedded.
        self.clerk.line(  )
        self.subject.line(  )

try/except. Our version of the oops function (file oops.py) follows. As for the noncoding questions, changing oops to raise KeyError instead of IndexError means that the exception won’t be caught by the try handler (it “percolates” to the top level and triggers Python’s default error message). The names KeyError and IndexError come from the outermost built-in names scope. Import __builtin__ and pass it as an argument to the dir function to see for yourself.

def oops(  ):
    raise IndexError

def doomed(  ):
    try:
        oops(  )
    except IndexError:
        print 'caught an index error!'
    else:
        print 'no error caught...'

if __name__ == '__main__': doomed(  )

% python oops.py
caught an index error!

Exception objects and lists. Here’s the way we extended this module for an exception of our own (here a string, at first):

MyError = 'hello'

def oops(  ):
    raise MyError, 'world'

def doomed(  ):
    try:
        oops(  )
    except IndexError:
        print 'caught an index error!'
    except MyError, data:
        print 'caught error:', MyError, data
    else:
        print 'no error caught...'

if __name__ == '__main__':
    doomed(  )

% python oops.py
caught error: hello world

To identify the exception with a class, we just changed the first part of the file to this, and saved it as oop_oops.py:

class MyError: pass

def oops(  ):
    raise MyError(  )

...rest unchanged...

Like all class exceptions, the instance comes back as the extra data; our error message now shows both the class, and its instance (<...>).

% python oop_oops.py
caught error: __main__.MyError <__main__.MyError instance at 0x00867550>

Remember, to make this look nicer, you can define a __repr__ or __str__ method in your class to return a custom print string. See Chapter 21 for details.

Error handling. Here’s one way to solve this one (file safe2.py). We did our tests in a file, rather than interactively, but the results are about the same.

import sys, traceback

def safe(entry, *args):
    try:
        apply(entry, args)                 # catch everything else
    except:
        traceback.print_exc(  )
        print 'Got', sys.exc_type, sys.exc_value

import oops
safe(oops.oops)

% python safe2.py
Traceback (innermost last):
  File "safe2.py", line 5, in safe
    apply(entry, args)                     # catch everything else
  File "oops.py", line 4, in oops
    raise MyError, 'world'
hello: world
Got hello world