Languages like C and C++ allow a programmer to write code at a very detailed level which has good execution speed (especially in the case of C). But in most applications, execution speed is not important, and in many cases one would prefer to write at a higher level. For example, for text-manipulation applications, the basic unit in C/C++ is a character, while for languages like Perl and Python the basic units are lines of text and words within lines. One can work with lines and words in C/C++, but one must go to greater effort to accomplish the same thing.
The term scripting language has never been formally defined, but here are the typical characteristics:
The first really popular scripting language was Perl. It is still in wide usage today, but the languages with momentum are Python and the Python-like Ruby. Many people, including me, greatly prefer Python to Perl, as it is much cleaner and more elegant. Python is very popular among the developers at Google.
Advocates of Python, often called pythonistas, say that Python is so clear and so enjoyable to write in that one should use Python for all of one's programming work, not just for scripting work. They believe it is superior to C or C++.1 Personally, I believe that C++ is bloated and its pieces don't fit together well; Java is nicer, but its strongly-typed nature is in my view a nuisance and an obstacle to clear programming. I was pleased to see that Eric Raymond, the prominent promoter of the open source movement, has also expressed the same views as mine regarding C++, Java and Python.
Anyone with even a bit of programming experience should find the material through Section 8 to be quite accessible.
The material beginning with Section 10 will feel quite comfortable to anyone with background in an object-oriented programming (OOP) language such as C++ or Java. If you lack this background, you will still be able to read these sections, but will probably need to go through them more slowly than those who do know OOP; just focus on the examples, not the terminology.
There will be a couple of places in which we describe things briefly in a Unix context, so some Unix knowledge would be helpful, but it certainly is not required. Python is used on Windows and Macintosh platforms too, not just Unix.
Our approach here is different from that of most Python books, or even most Python Web tutorials. The usual approach is to painfully go over all details from the beginning. For example, the usual approach would be to state all possible forms that a Python integer can take on, and for that matter how many different command-line options one can launch Python with.
I avoid this here. Again, the aim is to enable the reader to quickly acquire a Python foundation. He/she should then be able to delve directly into some special topic if and when the need arises.
I would suggest that you first read through Section 8, and then give Python a bit of a try yourself. First experiment a bit in Python's interactive mode (Section 3.4). Then try writing a few short programs yourself. These can be entirely new programs, or merely modifications of the example programs presented below.2
This will give you a much more concrete feel of the language. If your main use of Python will be to write short scripts and you won't be using the Python library, this will probably be enough for you. However, most readers will need to go further, acquiring a basic knowledge of Python's OOP features and Python modules/packages. So you should next read through Section 17.
That would be a very solid foundation for you from which to make good use of Python. Eventually, you may start to notice that many Python programmers make use of Python's functional programming features, and you may wish to understand what the others are doing or maybe use these features yourself. If so, Section 19 will get you started.
Don't forget the appendices! The key ones are Sections A and B.
I also have a number of tutorials on Python special programming, e.g. network programming, iterators/generators, etc. See http://heather.cs.ucdavis.edu/~matloff/python.html.
Here is a simple, quick example. Suppose I wish to find the value of
for x = 0.1, 0.2, ..., 0.9. I could find these numbers by placing the following code,
for i in range(10): x = 0.1*i print x print x/(1-x*x)
in a file, say fme.py, and then running the program by typing
python fme.py
at the command-line prompt. The output will look like this:
0.0 0.0 0.1 0.10101010101 0.2 0.208333333333 0.3 0.32967032967 0.4 0.47619047619 0.5 0.666666666667 0.6 0.9375 0.7 1.37254901961 0.8 2.22222222222 0.9 4.73684210526
How does the program work? First, Python's range() function is an example of the use of lists, i.e. Python arrays,3 even though not quite explicitly. Lists are absolutely fundamental to Python, so watch out in what follows for instances of the word ``list''; resist the temptation to treat it as the English word ``list,'' instead always thinking about the Python construct list.
Python's range() function returns a list of consecutive integers, in this case the list [0,1,2,3,4,5,6,7,8,9]. Note that this is official Python notation for lists--a sequence of objects (these could be all kinds of things, not necessarily numbers), separated by commas and enclosed by brackets.
So, the for statement above is equivalent to:
for i in [0,1,2,3,4,5,6,7,8,9]:
As you can guess, this will result in 10 iterations of the loop, with i first being 0, then 1, etc.
The code
for i in [2,3,6]:
would give us three iterations, with i taking on the values 2, 3 and 6.
Python has a while construct too (though not an until). There is also a break statement like that of C/C++, used to leave loops ``prematurely.'' For example:
x = 5
while 1:
x += 1
if x == 8:
print x
break
Now focus your attention on that inoccuous-looking colon at the end of the for line, which defines the start of a block. Unlike languages like C/C++ or even Perl, which use braces to define blocks, Python uses a combination of a colon and indenting to define a block. I am using the colon to say to the Python interpreter,
Hi, Python interpreter, how are you? I just wanted to let you know, by inserting this colon, that a block begins on the next line. I've indented that line, and the two lines following it, further right than the current line, in order to tell you those three lines form a block.
I chose 3-space indenting, but the amount wouldn't matter as long as I am consistent. If for example I were to write4
for i in range(10):
print 0.1*i
print g(0.1*i)
the Python interpreter would give me an error message, telling me that I have a syntax error.5 I am only allowed to indent further-right within a given block if I have a sub-block within that block, e.g.
for i in range(10):
if i%2 == 1:
print 0.1*i
print g(0.1*i)
Here I am printing out only the cases in which the variable i is an odd number; % is the ``mod'' operator as in C/C++.6 Again, note the colon at the end of the if line, and the fact that the two print lines are indented further right than the if line.
Note also that, again unlike C/C++/Perl, there are no semicolons at the end of Python source code statements. A new line means a new statement. If you need a very long line, you can use the backslash character for continuation, e.g.
x = y + \
z
A really nice feature of Python is its ability to run in interactive mode. You usually won't do this, but it's a great way to do a quick tryout of some feature, to really see how it works. Whenever you're not sure whether something works, your motto should be, ``When in doubt, try it out!'', and interactive mode makes this quick and easy.
We'll also be doing a lot of this in this tutorial, with interactive mode being an easy way to do a quick illustration of a feature.
Instead of executing this program from the command line in batch mode as we did above, we could enter and run the code in interactive mode:
% python >>> for i in range(10): ... x = 0.1*i ... print x ... print x/(1-x*x) ... 0.0 0.0 0.1 0.10101010101 0.2 0.208333333333 0.3 0.32967032967 0.4 0.47619047619 0.5 0.666666666667 0.6 0.9375 0.7 1.37254901961 0.8 2.22222222222 0.9 4.73684210526 >>>
Here I started Python, and it gave me its 
interactive prompt.
Then I just started typing in the code, line by line. Whenever I was
inside a block, it gave me a special prompt, ``...'', for that purpose.
When I typed a blank line at the end of my code, the Python interpreter
realized I was done, and ran the code.7
While in interactive mode, one can go up and down the command history by using the arrow keys, thus saving typing.
To exit interactive Python, hit ctrl-d.
Automatic printing: By the way, in interactive mode, just referencing or producing an object, or even an expression, without assigning it, will cause its value to print out, even without a print statement. For example:
>>> for i in range(4): ... 3*i ... 0 3 6 9
Again, this is true for general objects, not just expressions, e.g.:
>>> open('x')
<open file 'x', mode 'r' at 0xb7eaf3c8>
Here we opened the file x, which produces a file object. Since we did not assign to a variable, say f, for reference later in the code, i.e. we did not do the more typical
f = open('x')
the object was printed out. We'd get that same information this way:
>>> f = open('x')
>>> f
<open file 'x', mode 'r' at 0xb7f2a3c8>
Among other things, this means you can use Python as a quick calculator (which I do a lot). If for example I needed to know what 5% above $88.88 is, I could type
% python >>> 1.05*88.88 93.323999999999998
Among other things, one can do quick conversions between decimal and hex:
>>> 0x12 18 >>> hex(18) '0x12'
If I need math functions, I must import the Python math library first. This is analogous to what we do in C/C++, where we must have a #include line for the library in our source code and must link in the machine code for the library. Then we must refer to the functions in the context of the math library. For example, the functions sqrt() and sin() must be prefixed by math:8
>>> import math >>> math.sqrt(88) 9.3808315196468595 >>> math.sin(2.5) 0.59847214410395655
This program reads a text file, specified on the command line, and prints out the number of lines and words in the file:
[fontsize=\relsize{-2},numbers=left]
# reads in the text file whose name is specified on the command line,
# and reports the number of lines and words
import sys
def checkline():
global l
global wordcount
w = l.split()
wordcount += len(w)
wordcount = 0
f = open(sys.argv[1])
flines = f.readlines()
linecount = len(flines)
for l in flines:
checkline()
print linecount, wordcount
Say for example the program is in the file tme.py, and we have a text file x with contents
This is an example of a text file.
(There are five lines in all, the first and last of which are blank.)
If we run this program on this file, the result is:
python tme.py x 5 8
On the surface, the layout of the code here looks like that of a C/C++ program: First an import statement, analogous to #include (with the corresponding linking at compile time) as stated above; second the definition of a function; and then the ``main'' program. This is basically a good way to look at it, but keep in mind that the Python interpreter will execute everything in order, starting at the top. In executing the import statement, for instance, that might actually result in some code being executed, if the module being imported has some free-standing code. More on this later. Execution of the def statement won't execute any code for now, but the act of defining the function is considered execution.
Here are some features in this program which were not in the first example:
I will discuss these features in the next few sections.
First, let's explain sys.argv. Python includes a module (i.e. library) named sys, one of whose member variables is argv. The latter is a Python list, analogous to argv in C/C++.9 Element 0 of the list is the script name, in this case tme.py, and so on, just as in C/C++. In our example here, in which we run our program on the file x, sys.argv[1] will be the string 'x' (strings in Python are generally specified with single quote marks). Since sys is not loaded automatically, we needed the import line.
Both in C/C++ and Python, those command-line arguments are of course strings. If those strings are supposed to represent numbers, we could convert them. If we had, say, an integer argument, in C/C++ we would do the conversion using atoi(); in Python, we'd use int().10 For floating-point, in Python we'd use float().11
The function open() is similar to the one in C/C++. Our line
f = open(sys.argv[1])
created an object of file class, and assigned it to f .
The readlines() function of the file class returns a list (keep in mind, ``list'' is an official Python term) consisting of the lines in the file. Each line is a string, and that string is one element of the list. Since the file here consisted of five lines, the value returned by calling readlines() is the five-element list
['','This is an','example of a','text file','']
(Though not visible here, there is an end-of-line character in each string.)
Variables are not declared in Python. A variable is created when the first assignment to it is executed. For example, in the program tme.py above, the variable flines does not exist until the statement
flines = f.readlines()
is executed.
By the way, a variable which has not been assigned a value yet has the value None (and this can be assigned to a variable, tested for in an if statement, etc.).
Python does not really have global variables in the sense of C/C++, in which the scope of a variable is an entire program. We will discuss this further in Section 14.1.5, but for now assume our source code consists of just a single .py file; in that case, Python does have global variables pretty much like in C/C++.
Python tries to infer the scope of a variable from its position in the code. If a function includes any code which assigns to a variable, then that variable is assumed to be local. So, in the code for checkline(), Python would assume that l and wordcount are local to checkline() if we don't inform it otherwise. We do the latter with the global keyword.
Use of global variables simplifies the presentation here, and I personally believe that the unctuous criticism of global variables is unwarranted. (See http://heather.cs.ucdavis.edu/~matloff/globals.html.) In fact, in one of the major types of programming, threads, use of globals is basically mandatory.
You may wish, however, to at least group together all your globals into a class, as I do. See Appendix D.
The function len() returns the number of elements in a list, in this case, the number of lines in the file (since readlines() returned a list in which each element consisted of one line of the file).
The method split() is a member of the string class.12 It splits a string into a list of words, for example.13 So, for instance, in checkline() when l is 'This is an' then the list w will be equal to ['This','is','an']. (In the case of the first line, which is blank, w will be equal to the empty list, .)
As is typical in scripting languages, type in the sense of C/C++ int or float is not declared in Python. However, the Python interpreter does internally keep track of the type of all objects. Thus Python variables don't have types, but their values do. In other words, a variable X might be bound to an integer at one point in your program and then be rebound to a class instance at another point. In other words, Python uses dynamic typing.
Python's types include notions of scalars, sequences (lists or tuples) and dictionaries (associative arrays, discussed in Sec. 7.3), classes, function, etc.
Unlike Perl, Python does distinguish between numbers and their string representations. The functions eval() and str() can be used to convert back and forth. For example:
>>> 2 + '1.5'
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: unsupported operand type(s) for +: 'int' and 'str'
>>> 2 + eval('1.5')
3.5
>>> str(2 + eval('1.5'))
'3.5'
There are also int() to convert from strings to integers, and float(), to convert from strings to floating-point values:
>>> n = int('32')
>>> n
32
>>> x = float('5.28')
>>> x
5.2800000000000002
See also Section 17.4.
Lists are actually special cases of sequences, which are all array-like but with some differences. Note though, the commonalities; all of the following (some to be explained below) apply to any sequence type:
As stated earlier, lists are denoted by brackets and commas. For instance, the statement
x = [4,5,12]
would set x to the specified 3-element array.
Arrays may grow dynamically, using the list class' append() or extend() functions. For example, if after the abovfe statement we were to execute
x.append(-2)
x would now be equal to [4,5,12,-2].
A number of other operations are available for lists, a few of which are illustrated in the following code:
[fontsize=\relsize{-2},numbers=left]
>>> x = [5,12,13,200]
>>> x
[5, 12, 13, 200]
>>> x.append(-2)
>>> x
[5, 12, 13, 200, -2]
>>> del x[2]
>>> x
[5, 12, 200, -2]
>>> z = x[1:3] # array "slicing": elements 1 through 3-1 = 2
>>> z
[12, 200]
>>> yy = [3,4,5,12,13]
>>> yy[3:] # all elements starting with index 3
[12, 13]
>>> yy[:3] # all elements up to but excluding index 3
[3, 4, 5]
>>> yy[-1] # means "1 item from the right end"
13
>>> x.insert(2,28) # insert 28 at position 2
>>> x
[5, 12, 28, 200, -2]
>>> 28 in x # tests for membership; 1 for true, 0 for false
1
>>> 13 in x
0
>>> x.index(28) # finds the index within the list of the given value
2
>>> x.remove(200) # different from "delete," since it's indexed by value
>>> x
[5, 12, 28, -2]
>>> w = x + [1,"ghi"] # concatenation of two or more lists
>>> w
[5, 12, 28, -2, 1, 'ghi']
>>> qz = 3*[1,2,3] # list replication
>>> qz
[1, 2, 3, 1, 2, 3, 1, 2, 3]
>>> x = [1,2,3]
>>> x.extend([4,5])
>>> x
[1, 2, 3, 4, 5]
>>> y = x.pop(0) # deletes and returns 0th element
>>> y
1
>>> x
[2, 3, 4, 5]
We also saw the in operator in an earlier example, used in a for loop.
A list could include mixed elements of different types, including other lists themselves.
The Python idiom includes a number of common ``Python tricks'' involving sequences, e.g. the following quick, elegant way to swap two variables x and y:
>>> x = 5 >>> y = 12 >>> [x,y] = [y,x] >>> x 12 >>> y 5
Multidimensional lists can be implemented as lists of lists. For example:
>>> x = [] >>> x.append([1,2]) >>> x [[1, 2]] >>> x.append([3,4]) >>> x [[1, 2], [3, 4]] >>> x[1][1] 4
But be careful! Look what can go wrong:
>>> x = 4*[0] >>> y = 4*[x] >>> y [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]] >>> y[0][2] 0 >>> y[0][2] = 1 >>> y [[0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 1, 0]]
The problem is that that assignment to y was really a list of four references to the same thing (x). When the object pointed to by x changed, then all four rows of y changed.
The Python Wikibook (http://en.wikibooks.org/wiki/Python_Programming/Lists) suggests a solution, in the form of list comprehensions, which we cover in Section 19.4:
>>> z = [[0]*4 for i in range(5)] >>> z [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]] >>> z[0][2] = 1 >>> z [[0, 0, 1, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]
Tuples are like lists, but are immutable, i.e. unchangeable. They are enclosed by parentheses or nothing at all, rather than brackets. The parentheses are mandatory if there is an ambiguity without them, e.g. in function arguments. A comma must be used in the case of empty or single tuple, e.g. (,) and (5,).
The same operations can be used, except those which would change the tuple. So for example
x = (1,2,'abc') print x[1] # prints 2 print len(x) # prints 3 x.pop() # illegal, due to immutability
A nice function is zip(), which strings together corresponding components of several lists, producing tuples, e.g.
>>> zip([1,2],['a','b'],[168,168]) [(1, 'a', 168), (2, 'b', 168)]
Strings are essentially tuples of character elements. But they are quoted instead of surrounded by parentheses, and have more flexibility than tuples of character elements would have. For example:
[fontsize=\relsize{-2},numbers=left]
>>> x = 'abcde'
>>> x[2]
'c'
>>> x[2] = 'q' # illegal, since strings are immmutable
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: object doesn't support item assignment
>>> x = x[0:2] + 'q' + x[3:5]
>>> x
'abqde'
(You may wonder why that last assignment
>>> x = x[0:2] + 'q' + x[3:5]
does not violate immmutability. The reason is that x is really a pointer, and we are simply pointing it to a new string created from old ones. See Section 11.)
As noted, strings are more than simply tuples of characters:
>>> x.index('d') # as expected
3
>>> 'd' in x # as expected
1
>>> x.index('de') # pleasant surprise
3
As can be seen, the index() function from the str class has been overloaded, making it more flexible.
There are many other handy functions in the str class. For example, we saw the split() function earlier. The opposite of this function is join(). One applies it to a string, with a sequence of strings as an argument. The result is the concatenation of the strings in the sequence, with the original string between each of them:15
>>> '---'.join(['abc','de','xyz'])
'abc---de---xyz'
>>> q = '\n'.join(('abc','de','xyz'))
>>> q
'abc\nde\nxyz'
>>> print q
abc
de
xyz
Here are some more:
>>> x = 'abc'
>>> x.upper()
'ABC'
>>> 'abc'.upper()
'ABC'
>>> 'abc'.center(5) # center the string within a 5-character set
' abc '
>>> 'abc de f'.replace(' ','+')
'abc+de+f'
A very rich set of functions for string manipulation is also available in the re (``regular expression'') module.
The str class is built-in for newer versions of Python. With an older version, you will need a statement
import string
That latter class does still exist, and the newer str class does not quite duplicate it.
The Python function sort() can be applied to any sequence. For
nonscalars, one provides a ``compare'' function, which returns a
negative, zero or positive value, signfiying 
, 
or 
. As an
illustration, let's sort an array of arrays, using the second elements
as keys:
>>> x = [[1,4],[5,2]] >>> x [[1, 4], [5, 2]] >>> x.sort() >>> x [[1, 4], [5, 2]] >>> def g(u,v): ... return u[1]-v[1] ... >>> x.sort(g) >>> x [[5, 2], [1, 4]]
(This would be more easily done using ``lambda'' functions. See Section 19.1.)
There is a Python library module, bisect, which does binary search and related sorting.
Dictionaries are associative arrays. The technical meaning of this will be discussed below, but from a pure programming point of view, this means that one can set up arrays with non-integer indices. The statement
x = {'abc':12,'sailing':'away'}
sets x to what amounts to a 2-element array with x['abc'] being 12 and x['sailing'] equal to 'away'. We say that 'abc' and 'sailing' are keys, and 12 and 'away' are values. Keys can be any immmutable object, i.e. numbers, tuples or strings.16 Use of tuples as keys is quite common in Python applications, and you should keep in mind that this valuable tool is available.
Internally, x here would be stored as a 4-element array, and the execution of a statement like
w = x['sailing']
would require the Python interpreter to search through that array for the key 'sailing'. A linear search would be slow, so internal storage is organized as a hash table. This is why Perl's analog of Python's dictionary concept is actually called a hash.
Here are examples of usage of some of the member functions of the dictionary class:
[fontsize=\relsize{-2},numbers=left]
>>> x = {'abc':12,'sailing':'away'}
>>> x['abc']
12
>>> y = x.keys()
>>> y
['abc', 'sailing']
>>> z = x.values()
>>> z
[12, 'away']
x['uv'] = 2
>>> x
{'abc': 12, 'uv': 2, 'sailing': 'away'}
Note how we added a new element to x near the end.
The keys need not be tuples. For example:
>>> x
{'abc': 12, 'uv': 2, 'sailing': 'away'}
>>> f = open('z')
>>> x[f] = 88
>>> x
{<open file 'z', mode 'r' at 0xb7e6f338>: 88, 'abc': 12, 'uv': 2, 'sailing': 'away'}
Deletion of an element from a dictionary can be done via pop(), e.g.
>>> x.pop('abc')
12
>>> x
{<open file 'x', mode 'r' at 0xb7e6f338>: 88, 'uv': 2, 'sailing': 'away'}
The in operator works on dictionary keys, e.g.
>>> 'uv' in x True >>> 2 in x False
Obviously the keyword def is used to define a function. Note once again that the colon and indenting are used to define a block which serves as the function body. A function can return a value, using the return keyword, e.g.
return 8888
However, the function does not have a type even if it does return something, and the object returned could be anything--an integer, a list, or whatever.
Functions are first-class objects, i.e. can be assigned just like variables. Function names are variables; we just temporarily assign a set of code to a name. Consider:
>>> def square(x): # define code, and point the variable square to it ... return x*x ... >>> square(3) 9 >>> gy = square # now gy points to that code too >>> gy(3) 9 >>> def cube(x): ... return x**3 ... >>> cube(3) 27 >>> square = cube # point the variable square to the cubing code >>> square(3) 27 >>> square = 8.8 >>> square 8.8000000000000007 # don't be shocked by the 7 >>> gy(3) # gy still points to the squaring code 9
The raw_input() function will display a prompt and read in what is typed. For example,
name = raw_input('enter a name: ')
would display ``enter a name:'', then read in a response, then store that response in name. Note that the user input is returned in string form, and needs to be converted if the input consists of numbers.
If you don't want the prompt, don't specify one:
>>> y = raw_input() 3 >>> y '3'
Alternatively, you can directly specify stdin:
>>> import sys >>> z = sys.stdin.readlines() abc de f >>> z ['abc\n', 'de\n', 'f\n']
After typing `f', I hit ctrl-d to close the stdin file.)
In some cases, it is important to know whether a module is being executed on its own, or via import. This can be determined through Python's built-in variable __name__, as follows.
Whatever the Python interpreter is running is called the top-level program. If for instance you type
% python x.py
then the code in x.py is the top-level program. If you are running Python interactively, then the code you type in is the top-level program.
The top-level program is known to the interpreter as __main__, and the module currently being run is referred to as __name__. So, to test whether a given module is running on its own, versus having been imported by other code, we check whether __name__ is __main__. If the answer is yes, you are in the top level, and your code was not imported; otherwise it was.
For example, let's add a statement
print __name__
to our very first code example, from Section 3.1, in the file fme.py:
print __name__ for i in range(10): x = 0.1*i print x print x/(1-x*x)
Let's run the program twice. First, we run it on its own:
% python fme.py __main__ 0.0 0.0 0.1 0.10101010101 0.2 0.208333333333 0.3 0.32967032967 ... [remainder of output not shown]
Now look what happens if we run it from within Python's interactive interpreter:
>>> __name__ '__main__' >>> import fme fme 0.0 0.0 0.1 0.10101010101 0.2 0.208333333333 0.3 0.32967032967 ... [remainder of output not shown]
Our module's statement
print __name__
printed out __main__ the first time, but printed out fme the second time.
It is customary to collect one's ``main program'' (in the C sense) into a function, typically named main(). So, let's change our example above to fme2.py:
def main():
for i in range(10):
x = 0.1*i
print x
print x/(1-x*x)
if __name__ == '__main__':
main()
The advantage of this is that when we import this module, the code won't be executed right away. Instead, fme2.main() must be called, either by the importing module or by the interactive Python interpreter. Here is an example of the latter:
>>> import fme2 >>> fme2.main() 0.0 0.0 0.1 0.10101010101 0.2 0.208333333333 0.3 0.32967032967 0.4 0.47619047619 ...
Among other things, this will be a vital point in using debugging tools (Section A). So get in the habit of always setting up access to main() in this manner in your programs.
In contrast to Perl, Python has been object-oriented from the beginning, and thus has a much nicer, cleaner, clearer interface for OOP.
As an illustration, we will develop a class which deals with text files. Here are the contents of the file tfe.py:
[fontsize=\relsize{-2},numbers=left]
class textfile:
ntfiles = 0 # count of number of textfile objects
def __init__(self,fname):
textfile.ntfiles += 1
self.name = fname # name
self.fh = open(fname) # handle for the file
self.lines = self.fh.readlines()
self.nlines = len(self.lines) # number of lines
self.nwords = 0 # number of words
self.wordcount()
def wordcount(self):
"finds the number of words in the file"
for l in self.lines:
w = l.split()
self.nwords += len(w)
def grep(self,target):
"prints out all lines containing target"
for l in self.lines:
if l.find(target) >= 0:
print l
a = textfile('x')
b = textfile('y')
print "the number of text files open is", textfile.ntfiles
print "here is some information about them (name, lines, words):"
for f in [a,b]:
print f.name,f.nlines,f.nwords
a.grep('example')
In addition to the file x I used in Section 4 above, I had the 2-line file y. Here is what happened when I ran the program:
% python tfe.py the number of text files opened is 2 here is some information about them (name, lines, words): x 5 8 y 2 5 example of a
Let's take a look at the class textfile. The first thing to note is the prevalence of the keyword self, meaning the current instance of the class, analogous to this in C++ and Java.17 So, self is a pointer to the current instance of the class.
In general OOP terminology, an instance variable x of a class is a member variable for which each instance of the class has a separate value of that variable, In the C++ or Java world, you know this as a variable which is not declared static. The term instance variable is the generic OOP term, non-language specific.
To see how these work in Python, recall first that a variable in Python is created when something is assigned to it. So, an instance variable in an instance of a Python class does not exist until it is assigned to.
For example, when
self.name = fname
is executed, the member variable name for the current instance of the class is created, and is assigned the indicated value.
A class variable say v, has a common value in all instances of the class. Again in the C++ or Java world, you know this as a static variable. It is designated as such by having some reference to v in code which is in the class but not in any method of the class. An example is the code
ntfiles = 0 # count of number of textfile objects
above.18
Note that a class variable v of a class u is referred to as u.v within methods of the class and in code outside the class. For code inside the class but not within a method, it is referred to as simply v. Take a moment now to go through our example program above, and see examples of this with our ntfiles variable.
The constructor for a class must be named __init()__. The argument self is mandatory, and you can add others, as I've done in this case with a filename.
The destructor is __del()__. Note that it is only invoked when garbage collection is done, i.e. when all variables pointing to the object are gone.
The method wordcount() is an instance method, i.e. it applies specifically to the given object of this class. Again, in C++/Java terminology, this is a non-static method. Unlike C++ and Java, where this is an implicit argument to instance methods, Python wisely makes the relation explicit; the argument self is required.
Before Version 2.2, Python had no formal provision for class methods, i.e. methods which do not apply to specific objects of the class. Now Python has two (slightly differing) ways to do this, using the functions staticmethod() and classmethod(). I will present use of the former, in the following enhancement to the code in Section 10.1 within the class textfile:
class textfile:
...
def totfiles():
print "the total number of text files is", textfile.ntfiles
totfiles = staticmethod(totfiles)
...
# here we are in "main"
...
textfile.totfiles()
...
Note that class methods do not have the self argument. (Nor should they, right?)
Note also that this method could be called even if there are not yet any instances of the class textfile. In the example here, 0 would be printed out, since no files had yet been counted.19
Inheritance is very much a part of the Python philosophy. A statement like
class b(a):
starts the definition of a subclass b of a class a. Multiple inheritance, etc. can also be done.
Note that when the constructor for a derived class is called, the constructor for the base class is not automatically called. If you wish the latter constructor to be invoked, you must invoke it yourself, e.g.
class b(a):
def __init__(self,xinit): # constructor for class b
self.x = xinit # define and initialize an instance variable x
a.__init__(self) # call base class constructor
The official Python tutorial notes, ``[In the C++ sense] all methods in Python are effectively virtual.'' If you wish to extend, rather than override a method in the base class, you can refer to the latter by prepending the base name, as in our example a.__init__(self) above.
A Python class instance is implemented internally as a dictionary. For example, in our program tfe.py above, the object b is implemented as a dictionary.
Among other things, this means that you can add member variables to an instance of a class ``on the fly,'' long after the instance is created. We are simply adding another key and value to the dictionary. In our ``main'' program, for example, we could have a statement like
b.name = 'zzz'
A variable which has been assigned a mutable value is actually a pointer to the given object. For example, consider this code:
>>> x = [1,2,3] # x is mutable >>> y = x # x and y now both point to [1,2,3] >>> x[2] = 5 # the mutable object pointed to by x now "mutes" >>> y[2] # this means y[2] changes to 5 too! 5 >>> x = [1,2] >>> y = x >>> y [1, 2] >>> x = [3,4] >>> y [1, 2]
In the first few lines, x and y are references to a list, a mutable object. The statement
x[2] = 5
then changes one aspect of that object, but x still points to that object. On the other hand, the code
x = [3,4]
now changes x itself, having it point to a different object, while y is still pointing to the first object.
If in the above example we wished to simply copy the list referenced by x to y, we could use slicing, e.g.
y = x[:]
Then y and x would point to different objects; x would point to the same object as before, but the statement for y would create a new object, which y would point to. Even though those two objects have the same values for the time being, if the object pointed to by x changes, y's object won't change.
As you can imagine, this gets delicate when we have complex objects. See Python's copy module for functions that will do object copying to various depths.
An important similar issue arises with arguments in function calls. Any argument which is a variable which points to a mutable object can change the value of that object from within the function, e.g.:
>>> def f(a): ... a = 2*a # numbers are immutable ... >>> x = 5 >>> f(x) >>> x 5 >>> def g(a): ... a[0] = 2*a[0] # lists are mutable ... >>> y = [5] >>> g(y) >>> y [10]
Function names are references to objects too. What we think of as the name of the function is actually just a pointer--a mutable one--to the code for that function. For example,
>>> def f(): ... print 1 ... >>> def g(): ... print 2 ... >>> f() 1 >>> g() 2 >>> [f,g] = [g,f] >>> f() 2 >>> g() 1
For some more practice with these notions, see Section C.
Objects can be deleted from Python's memory by using del, e.g.
>>> del x
NOTE CAREFULLY THAT THIS IS DIFFERENT FROM DELETION FROM A LIST OR DICTIONARY. If you use remove() or pop(), for instance, you are simply removing the pointer to the object from the given data structure, but as long as there is at least one reference, i.e. a pointer, to an object, that object still takes up space in memory.
This can be a major issue in long-running programs. If you are not careful to delete objects, or if they are not simply garbage-collected when their scope disappears, you can accumulate more and more of them, and have a very serious memory problem. If you see your machine running ever more slowly while a program is running, you should immediately suspect this.
One can use the operator to compare sequences, e.g.
if x < y:
for lists x and y. The comparison is lexicographic.
For example,
>>> [12,'tuv'] < [12,'xyz'] True >>> [5,'xyz'] > [12,'tuv'] False
Of course, since strings are sequences, we can compare them too:
>>> 'abc' < 'tuv' True >>> 'xyz' < 'tuv' False >>> 'xyz' != 'tuv' True
Note the effects of this on, for example, the max() function:
>>> max([[1, 2], [0], [12, 15], [3, 4, 5], [8, 72]]) [12, 15] >>> max([8,72]) 72
We can set up comparisons for non-sequence objects, e.g. class instances, by defining a __cmp()__ function in the class. The definition starts with
def __cmp__(self,other):
It must be defined to return a negative, zero or positive value, depending on whether self is less than, equal to or greater than other.
Very sophisticated sorting can be done if one combines Python's sort() function with a specialized cmp() function.
You've often heard that it is good software engineering practice to write your code in ``modular'' fashion, i.e. to break it up into components, top-down style, and to make your code ``reusable,'' i.e. to write it in such generality that you or someone else might make use of it in some other programs. Unlike a lot of follow-like-sheep software engineering shiboleths, this one is actually correct! :-)
A module is a set of classes, library functions and so on, all in one file. Unlike Perl, there are no special actions to be taken to make a file a module. Any file whose name has a .py suffix is a module!20
As our illustration, let's take the textfile class from our example above. We could place it in a separate file tf.py, with contents
[fontsize=\relsize{-2},numbers=left]
# file tf.py
class textfile:
ntfiles = 0 # count of number of textfile objects
def __init__(self,fname):
textfile.ntfiles += 1
self.name = fname # name
self.fh = open(fname) # handle for the file
self.lines = self.fh.readlines()
self.nlines = len(self.lines) # number of lines
self.nwords = 0 # number of words
self.wordcount()
def wordcount(self):
"finds the number of words in the file"
for l in self.lines:
w = l.split()
self.nwords += len(w)
def grep(self,target):
"prints out all lines containing target"
for l in self.lines:
if l.find(target) >= 0:
print l
Note that even though our module here consists of just a single class, we could have several classes, plus global variables,21 executable code not part of any function, etc.)
Our test program file, tftest.py, might now look like this:
[fontsize=\relsize{-2},numbers=left]
# file tftest.py
import tf
a = tf.textfile('x')
b = tf.textfile('y')
print "the number of text files open is", tf.textfile.ntfiles
print "here is some information about them (name, lines, words):"
for f in [a,b]:
print f.name,f.nlines,f.nwords
a.grep('example')
The Python interpreter, upon seeing the statement import tf, would load the contents of the file tf.py.22 Any executable code in tf.py is then executed, in this case
ntfiles = 0 # count of number of textfile objects
(The module's executable code might not only be within classes. See what happens when we do import fme2 in an example in Section 9 below.)
Later, when the interpreter sees the reference to tf.textfile, it would look for an item named textfile within the module tf, i.e. within the file tf.py, and then proceed accordingly.
An alternative approach would be:
[fontsize=\relsize{-2},numbers=left]
from tf import textfile
a = textfile('x')
b = textfile('y')
print "the number of text files open is", textfile.ntfiles
print "here is some information about them (name, lines, words):"
for f in [a,b]:
print f.name,f.nlines,f.nwords
a.grep('example')
This saves typing, since we type only ``textfile'' instead of ``tf.textfile,'' making for less cluttered code. But arguably it is less safe (what if tftest.py were to have some other item named textfile?) and less clear (textfile's origin in tf might serve to clarify things in large programs).
The statement
from tf import *
would import everything in tf.py in this manner.
In any event, by separating out the textfile class, we have helped to modularize our code, and possibly set it up for reuse.
Like the case of Java, the Python interpreter compiles any code it executes to byte code for the Python virtual machine. If the code is imported, then the compiled code is saved in a file with suffix .pyc, so it won't have to be recompiled again later.
Since modules are objects, the names of the variables, functions, classes etc. of a module are attributes of that module. Thus they are retained in the .pyc file, and will be visible, for instance, when you run the dir() function on that module (Section B.1).
A module's (free-standing, i.e. not part of a function) code executes immediately when the module is imported.
Modules are objects. They can be used as arguments to functions, return values from functions, etc.
The list sys.modules shows all modules ever imported into the currently running program.
Python does not truly allow global variables in the sense that C/C++ do. An imported Python module will not have direct access to the globals in the module which imports it, nor vice versa.
For instance, consider these two files, x.py,
# x.py import y def f(): global x x = 6 def main(): global x x = 3 f() y.g() if __name__ == '__main__': main()
and y.py:
# y.py def g(): global x x += 1
The variable x in x.py is visible throughout the module x.py, but not in y.py. In fact, execution of the line
x += 1
in the latter will cause an error message to appear, ``global name 'x' is not defined.''
Indeed, a global variable in a module is merely an attribute (i.e. a member entity) of that module, similar to a class variable's role within a class. When module B is imported by module A, B's namespace is copied to A's. If module B has a global variable X, then module A will create a variable of that name, whose initial value is whatever module B had for its variable of that name at the time of importing. But changes to X in one of the modules will NOT be reflected in the other.
Say X does change in B, but we want code in A to be able to get the latest value of X in B. We can do that by including a function, say named GetX() in B. Assuming that A imported everything from B, then A will get a function GetX() which is a copy of B's function of that name, and whose sole purpose is to return the value of X. Unless B changes that function (which is possible, e.g. functions may be assigned), the functions in the two modules will always be the same, and thus A can use its function to get the value of X in B.
Python has no strong form of data hiding comparable to the private and other such constructs in C++. It does offer a small provision of this sort, though:
If you prepend an underscore to a variable's name in a module, it will not be imported if the from form of import is used. For example, if in the module tf.py in Section 14.1.1 were to contain a variable z, then a statement
from tf import *
would mean that z is accesible as just z rather than tf.z. If on the other hand we named this variable _z, then the above statement would not make this variable accessible as _z; we would need to use tf._z. Of course, the variable would still be visible from outside the module, but by requiring the tf. prefix we would avoid confusion with similarly-named variables in the importing module.
A double underscore results in mangling, with another underscore plus the name of the module prepended.
As mentioned earlier, one might place more than one class in a given module, if the classes are closely related. A generalization of this arises when one has several modules that are related. Their contents may not be so closely related that we would simply pool them all into one giant module, but still they may have a close enough relationship that you want to group them in some other way. This is where the notion of a package comes in.
For instance, you may write some libraries dealing with some Internet software you've written. You might have one module web.py with classes you've written for programs which do Web access, and another module em.py which is for e-mail software. Instead of combining them into one big module, you could keep them as separate files put in the same directory, say net.
To make this directory a package, simply place a file __init__.py in that directory. The file can be blank, or in more sophisticated usage can be used for some startup operations.
In order to import these modules, you would use statements like
import net.web
This tells the Python interpreter to look for a file web.py within a directory net. The latter, or more precisely, the parent of the latter, must be in your Python search path. If for example the full path name for net were
/u/v/net
then the directory /u/v would need to be in your Python search path. If you are on a Unix system and using the C shell, for instance, you could type
setenv PYTHONPATH /u/v
If you have several special directories like this, string them all together, using colons as delimiters:
setenv PYTHONPATH /u/v:/aa/bb/cc
The current path is contained in sys.path. Again, it consists of a list of strings, one string for each directory, separated by colons. It can be printed out or changed by your code, just like any other variable.23
Package directories often have subdirectories, subsubdirectories and so on. Each one must contain a __init__.py file.
By the way, Python's built-in and library functions have no C-style error return code to check to see whether they succeeded. Instead, you use Python's try/except exception-handling mechanism, e.g.
try: f = open(sys.argv[1]) except: print 'open failed:',sys.argv[1]
Here's another example:
try: i = 5 y = x[i] except: print 'no such index:', i
But the Python idiom also uses this for code which is not acting in an exception context. Say for example we want to find the index of the number 8 in the list z and if there is no such number, to first add it to the list. We could do it this way:
try: place = x.index(8) except: x.append(8) place = len(x)
There is a double-quoted string, ``finds the number of words in the file'', at the beginning of wordcount(). This is called a docstring. It serves as a kind of comment, but at runtime, so that it can be used by debuggers and the like. Also, it enables users who have only the compiled form of the method, say as a commercial product, access to a ``comment.'' Here is an example of how to access it, using tf.py from above:
>>> import tf >>> tf.textfile.wordcount.__doc__ 'finds the number of words in the file'
A docstring typically spans several lines. To create this kind of string, use triple quote marks.
The method grep() is another instance method, this one with an argument besides self.
By the way, method arguments in Python can only be pass-by-value, in the sense of C: Functions have side effects with respect to the parameter if the latter is a pointer. (Python does not have formal pointers, but it does have references; see Section 11.)
Note also that grep() makes use of one of Python's many string operations, find(). It searches for the argument string within the object string, returning the index of the first occurrence of the argument string within the object string, or returning -1 if none is found.24
Say you have a Python script x.py. So far, we have discussed running it via the command25
% python x.py
But if you state the location of the Python interpreter in the first line of x.py, e.g.
#! /usr/bin/python
and use the Unix chmod command to make x.py executable, then you can run x.py by merely typing
% x.py
This is necessary, for instance, if you are invoking the program from a Web page.
Better yet, you can have Unix search your environment for the location of Python, by putting this as your first line in x.py:
#! /usr/bin/env python
This is more portable, as different platforms may place Python in different directories.
Consider this little example:
[fontsize=\relsize{-2},numbers=left]
def f(u,v=2):
return u+v
def main():
x = 2;
y = 3;
print f(x,y) # prints 5
print f(x) # prints 4
if __name__ == '__main__': main()
Here, the argument v is called a named argument, with default value 2. The ``ordinary'' argument u is called a mandatory argument, as it must be specified while v need not be. Another term for u is positional argument, as its value is inferred by its position in the order of declaration of the function's arguments. Mandatory arguments must be declared before named arguments.
A print statement automatically prints a newline character. To suppress it, add a trailing comma. For example:
print 5, # nothing printed out yet print 12 # '5 12' now printed out, with end-of-line
The print statement automatically separates items with blanks. To suppress blanks, use the string-concatenation operator, +, and possibly the str() function, e.g.
x = 'a' y = 3 print x+str(y) # prints 'a3'
By the way, str(None) is None.
Python supports C-style ``printf()'', e.g.
print "the factors of 15 are %d and %d" % (3,5)
prints out
the factors of 15 are 3 and 5
Note the importance of writing '(3,5)' rather than '3,5'. In the latter case, the % operator would think that its operand was merely 3, whereas it needs a 2-element tuple. Recall that parentheses enclosing a tuple can be omitted as long as there is no ambiguity, but that is not the case here.
This is nice, but it is far more powerful than just for printing, but for general string manipulation. In
print "the factors of 15 are %d and %d" % (3,5)
the portion
"the factors of 15 are %d and %d" % (3,5)
is a string operation, producing a new string; the print simply prints that new string.
For example:
>>> x = "%d years old" % 12
The variable x now is the string '12 years old'.
This is another very common idiom, quite powerful.26
Below is a Python class for implementing a binary tree. The comments should make the program self-explanatory (no pun intended).27
[fontsize=\relsize{-2},numbers=left]
# bintree.py, a module for handling sorted binary trees; values to be
# stored can be general, as long as an ordering relation exists
# here, only have routines to insert and print, but could add delete,
# etc.
class treenode:
def __init__(self,v):
self.value = v;
self.left = None;
self.right = None;
def ins(self,nd): # inserts the node nd into tree rooted at self
m = nd.value
if m < self.value:
if self.left == None:
self.left = nd
else:
self.left.ins(nd)
else:
if self.right == None:
self.right = nd
else:
self.right.ins(nd)
def prnt(self): # prints the subtree rooted at self
if self.value == None: return
if self.left != None: self.left.prnt()
print self.value
if self.right != None: self.right.prnt()
class tree:
def __init__(self):
self.root = None
def insrt(self,m):
newnode = treenode(m)
if self.root == None:
self.root = newnode
return
self.root.ins(newnode)
And here is a test:
[fontsize=\relsize{-2},numbers=left]
# trybt1.py: test of bintree.py
# usage: python trybt.py numbers_to_insert
import sys
import bintree
def main():
tr = bintree.tree()
for n in sys.argv[1:]:
tr.insrt(int(n))
tr.root.prnt()
if __name__ == '__main__': main()
The good thing about Python is that we can use the same code again for nonnumerical objects, as long as they are comparable. (Recall Section 13.) So, we can do the same thing with strings, using the tree and treenode classes AS IS, NO CHANGE, e.g.
# trybt2.py: test of bintree.py
# usage: python trybt.py strings_to_insert
import sys
import bintree
def main():
tr = bintree.tree()
for s in sys.argv[1:]:
tr.insrt(s)
tr.root.prnt()
if __name__ == '__main__': main()
% python trybt2.py abc tuv 12 12 abc tuv
Or even
# trybt3.py: test of bintree.py
import bintree
def main():
tr = bintree.tree()
tr.insrt([12,'xyz'])
tr.insrt([15,'xyz'])
tr.insrt([12,'tuv'])
tr.insrt([2,'y'])
tr.insrt([20,'aaa'])
tr.root.prnt()
if __name__ == '__main__': main()
% python trybt3.py [2, 'y'] [12, 'tuv'] [12, 'xyz'] [15, 'xyz'] [20, 'aaa']
In the example in Section 10.1, it is worth calling special attention to the line
for f in [a,b]:
where a and b are objects of type textfile. This illustrates the fact that the elements within a list do not have to be scalars.28 Much more importantly, it illustrates that really effective use of Python means staying away from classic C-style loops and expressions with array elements. This is what makes for much cleaner, clearer and elegant code. It is where Python really shines.
You should almost never use C/C++ style for loops--i.e. where an
index (say j), is tested against an upper bound (say j
10), and incremented at the end of each iteration (say j++).
Indeed, you can often avoid explicit loops, and should do so whenever possible. For example, the code
self.lines = self.fh.readlines() self.nlines = len(self.lines)
in that same program is much cleaner than what we would have in, say, C. In the latter, we would need to set up a loop, which would read in the file one line at a time, incrementing a variable nlines in each iteration of the loop.29
Another great way to avoid loops is to use Python's functional programming features, described in Section 19.
Making use of Python idioms is often referred to by the pythonistas as the pythonic way to do things.
These features provide concise ways of doing things which, though certainly doable via more basic constructs, compactify your code and thus make it easier to write and read. They may also make your code run much faster. Moreover, it may help us avoid bugs, since a lot of the infracture we'd need to write ourselves, which would be bug-prone, is automatically taken care of us by the functional programming constructs.
Except for the first feature here (lambda functions), these features eliminate the need for explicit loops and explicit references to list elements. As mentioned in Section 18.1, this makes for cleaner, clearer code.
Lambda functions provide a way of defining short functions. They help you avoid cluttering up your code with a lot of ``one-liners'' which are called only once. For example:
>>> g = lambda u:u*u >>> g(4) 16
Note carefully that this is NOT a typical usage of lambda functions; it was only to illustrate the syntax. Usually a lambda functions would not be defined in a free-standing manner as above; instead, it would be defined inside other functions such as map() and filter(), as seen next.
Here is a more realistic illustration, redoing the sort example from Section 7.2.4:
>>> x = [[1,4],[5,2]] >>> x [[1, 4], [5, 2]] >>> x.sort() >>> x [[1, 4], [5, 2]] >>> x.sort(lambda u,v: u[1]-v[1]) >>> x [[5, 2], [1, 4]]
A bit of explanation is necessary. If you look at the online help for sort(), you'll find that the definition to be
sort(...)
L.sort(cmp=None, key=None, reverse=False) -- stable sort *IN PLACE*;
cmp(x, y) -> -1, 0, 1
You see that the first argument is a named argument (recall Section
17.2), cmp. That is our compare function, which we
defined above as lambda u,v: u[1]-v[1].
The general form of a lambda function is
lambda arg 1, arg 2, ...: expression
So, multiple arguments are permissible, but the function body itself must be an expression.
The map() function converts one sequence to another, by applying the same function to each element of the sequence. For example:
>>> z = map(len,["abc","clouds","rain"]) >>> z [3, 6, 4]
So, we have avoided writing an explicit for loop, resulting in code which is a little cleaner, easier to write and read.30
In the example above we used a built-in function, len(). We could also use our own functions; frequently these are conveniently expressed as lambda functions, e.g.:
>>> x = [1,2,3] >>> y = map(lambda z: z*z, x) >>> y [1, 4, 9]
The condition that a lambda function's body consist only of an expression is rather limiting, for instance not allowing if-then-else constructs. If you really wish to have the latter, you could use a workaround. For example, to implement something like
if u > 2: u = 5
we could work as follows:
>>> x = [1,2,3] >>> g = lambda u: (u > 2) * 5 + (u <= 2) * u >>> map(g,x) [1, 2, 5]
Clearly, this is not feasible except for simple situations. For more complex cases, we would use a non-lambda function. For example, here is a revised version of the program in Section 4.1:
[fontsize=\relsize{-2},numbers=left]
import sys
def checkline(l):
global wordcount
w = l.split()
wordcount += len(w)
wordcount = 0
f = open(sys.argv[1])
flines = f.readlines()
linecount = len(flines)
map(checkline,flines) # replaces the old 'for' loop
print linecount, wordcount
Note that l is now an argument to checkline().
Of course, this could be reduced even further, with the heart of the ``main'' program above being changed to
map(checkline,open(sys.argv[1]).readlines)
But this is getting pretty hard to read and debug.
The filter() function works like map(), except that it culls out the sequence elements which satisfy a certain condition. The function which filter() is applied to must be boolean-valued, i.e. return the desired true or false value. For example:
>>> x = [5,12,-2,13] >>> y = filter(lambda z: z > 0, x) >>> y [5, 12, 13]
Again, this allows us to avoid writing a for loop and an if statement.
This allows you to compactify a for loop which produces a list. For example:
>>> x = [(1,-1), (12,5), (8,16)] >>> y = [(v,u) for (u,v) in x] >>> y [(-1, 1), (5, 12), (16, 8)]
This is more compact than first initializing y to [], then having a for loop in which we call y.append().
It gets even better when done in nested form. Say for instance we have a list of lists which we want to concatenate together, ignoring the first element in each. Here's how we could do it using list comprehensions:
>>> y [[0, 2, 22], [1, 5, 12], [2, 3, 33]] >>> [a for b in y for a in b[1:]] [2, 22, 5, 12, 3, 33]
The reduce() function is used for applying the sum or other arithmetic-like operation to a list. For example,
>>> x = reduce(lambda x,y: x+y, range(5)) >>> x 10
Here range(5) is of course [0,1,2,3,4]. What reduce() does is it first adds the first two elements of [0,1,2,3,4], i.e. with 0 playing the role of x and 1, playing the role of y. That gives a sum of 1. Then that sum, 1, plays the role of x and the next element of [0,1,2,3,4], 2, plays the role of y, yielding a sum of 3, etc. Eventually reduce() finishes its work and returns a value of 10.
Once again, this allowed us to avoid a for loop, plus a statement in which we initialize x to 0 before the for loop.
Do NOT debug by simply adding and subtracting print statements. Use a debugging tool! If you are not a regular user of a debugging tool, then you are causing yourself unnecessary grief and wasted time; see my debugging slide show, at http://heather.cs.ucdavis.edu/~matloff/debug.html.
The built-in debugger for Python, PDB, is rather primitive, but it's very important to understand how it works, for two reasons:
I will present PDB below in a sequence of increasingly-useful forms:
You should be able to find PDB in the lib subdirectory of your Python package. On a Unix system, for example, that is probably in something like /usr/lib/python2.2. /usr/local/lib/python2.4, etc. To debug a script x.py, type
% /usr/lib/python2.2/pdb.py x.py
(If x.py had had command-line arguments, they would be placed after x.py on the command line.)
Of course, since you will use PDB a lot, it would be better to make an alias for it. For example, under Unix in the C-shell:
alias pdb /usr/lib/python2.2/pdb.py
so that you can write more simply
% pdb x.py
Once you are in PDB, set your first breakpoint, say at line 12:
b 12
You can make it conditional, e.g.
b 12, z > 5
Hit c (``continue''), which you will get you into x.py and then stop at the breakpoint. Then continue as usual, with the main operations being like those of GDB:
Upon entering PDB, you will get its (Pdb) prompt.
If you have a multi-file program, breakpoints can be specified in the form module_name:line_number. For instance, suppose your main module is x.py, and it imports y.py. You set a breakpoint at line 8 of the latter as follows:
(Pdb) b y:8
Note, though, that you can't do this until y has actually been imported by x.32
When you are running PDB, you are running Python in its interactive mode. Therefore, you can issue any Python command at the PDB prompt. You can set variables, call functions, etc. This can be highly useful.
For example, although PDB includes the p command for printing out the values of variables and expressions, it usually isn't necessary. To see why, recall that whenever you run Python in interactive mode, simply typing the name of a variable or expression will result in printing it out--exactly what p would have done, without typing the `p'.
So, if x.py contains a variable ww and you run PDB, instead of typing
(Pdb) p ww
you can simply type
ww
and the value of ww will be printed to the screen.33
If your program has a complicated data structure, you could write a function to print to the screen all or part of that structure. Then, since PDB allows you to issue any Python command at the PDB prompt, you could simply call this function at that prompt, thus getting more sophisticated, application-specific printing.
After your program either finishes under PDB or runs into an execution error, you can re-run it without exiting PDB--important, since you don't want to lose your breakpoints--by simply hitting c. And yes, if you've changed your source code since then, the change will be reflected in PDB.34
If you give PDB a single-step command like n when you are on a Python line which does multiple operations, you will need to issue the n command multiple times (or set a temporary breakpoint to skip over this).
For example,
for i in range(10):
does two operations. It first calls range(), and then sets i, so you would have to issue n twice.
And how about this one?
y = [(y,x) for (x,y) in x]
If x has, say, 10 elements, then you would have to issue the n command 10 times! Here you would definitely want to set a temporary breakpoint to get around it.
PDB's undeniably bare-bones nature can be remedied quite a bit by making good use of the alias command, which I strongly suggest. For example, type
alias c c;;l
This means that each time you continue, when you next stop at a breakpoint you automatically get a listing of the neighboring code. This will really do a lot to make up for PDB's lack of a GUI.
In fact, this is so important that you should put it in your PDB startup file, which in Unix is $HOME/.pdbrc.35 That way the alias is always available. You could do the same for the n and s commands:
alias c c;;l alias n n;;l alias s s;;l
There is an unalias command too, to cancel an alias.
You can write other macros which are specific to the particular program you are debugging. For example, let's again suppose you have a variable named ww in x.py, and you wish to check its value each time the debugger pauses, say at breakpoints. Then change the above alias to
alias c c;;l;;ww
In Section A.6.1 below, we'll show that if o is an object of some class, then printing o.__dict__ will print all the member variables of this object. Again, you could combine this with PDB's alias capability, e.g.
alias c c;;l;;o.__dict__
Actually, it would be simpler and more general to use
alias c c;;l;;self
This way you get information on the member variables no matter what class you are in. On the other hand, this apparently does not produce information on member variables in the parent class.
In reading someone else's code, or even one's own, one might not be clear what type of object a variable currently references. For this, the type() function is sometimes handy. Here are some examples of its use:
>>> x = [5,12,13] >>> type(x) <type 'list'> >>> type(3) <type 'int'> >>> def f(y): return y*y ... >>> f(5) 25 >>> type(f) <type 'function'>
Emacs is a combination text editor and tools collection. Many software engineers swear by it. It is available for Windows, Macs and Unix/Linux; it is included in most Linux distributions. But even if you are not an Emacs aficionado, you may find it to be an excellent way to use PDB. You can split Emacs into two windows, one for editing your program and the other for PDB. As you step through your code in the second window, you can see yourself progress through the code in the first.
To get started, say on your file x.py, go to a command window (whatever you have under your operating system), and type either
emacs x.py
or
emacs -nw x.py
The former will create a new Emacs window, where you will have mouse operations available, while the latter will run Emacs in text-only operations in the current window. I'll call the former ``GUI mode.''
Then type M-x pdb, where for most systems ``M,'' which stands for ``meta,'' means the Escape (or Alt) key rather than the letter M. You'll be asked how to run PDB; answer in the manner you would run PDB externally to Emacs (but with a full path name), e.g.
/usr/local/lib/python2.4/pdb.py x.py 3 8
where the 3 and 8 in this example are your program's command-line arguments.
At that point Emacs will split into two windows, as described earlier. You can set breakpoints directly in the PDB window as usual, or by hitting C-x space at the desired line in your program's window; here and below, ``C-'' means hitting the control key and holding it while you type the next key.
At that point, run PDB as usual.
If you change your program and are using the GUI version of Emacs, hit
IM-Python 
Rescan to make the new version of your program known to
PDB.
In addition to coordinating PDB with your error, note that another advantage of Emacs in this context is that Emacs will be in Python mode, which gives you some extra editing commands specific to Python. I'll describe them below.
In terms of general editing commands, plug ``Emacs tutorial'' or ``Emacs commands'' into your favorite Web search engine, and you'll see tons of resources. Here I'll give you just enough to get started.
First, there is the notion of a buffer. Each file you are editing36 has its own buffer. Each other action you take produces a buffer too. For instance, if you invoke one of Emacs' online help commands, a buffer is created for it (which you can edit, save, etc. if you wish). An example relevant here is PDB. When you do M-x pdb, that produces a buffer for it. So, at any given time, you may have several buffers. You also may have several windows, though for simplicity we'll assume just two windows here.
In the following table, we show commands for both the text-only and the GUI versions of Emacs. Of course, you can use the text-based commands in the GUI too.
| action | text | GUI |
| cursor movement | arrow keys, PageUp/Down | mouse, left scrollbar |
| undo | C-x u | Edit Undo |
| cut | C-space (cursor move) C-w | select region Edit Cut |
| paste | C-y | Edit Paste |
| search for string | C-s | Edit Search |
| mark region | C-@ | select region |
| go to other window | C-x o | click window |
| enlarge window | (1 line at a time) C-x ^ | drag bar |
| repeat folowing command n times | M-x n | M-x n |
| list buffers | C-x C-b | Buffers |
| go to a buffer | C-x b | Buffers |
| exit Emacs | C-x C-c | File Exit Emacs |
In using PDB, keep in mind that the name of your PDB buffer will begin with ``gud,'' e.g. gud-x.py.
You can get a list of special Python operations in Emacs by typing C-h d and then requesting info in python-mode. One nice thing right off the bat is that Emacs' python-mode adds a special touch to auto-indenting: It will automatically indent further right after a def or class line. Here are some operations:
| action | text | GUI |
| comment-out region | C-space (cursor move) C-c # | select region
Python Comment |
| go to start of def or class | ESC C-a | ESC C-a |
| go to end of def or class | ESC C-e | ESC C-e |
| go one block outward | C-c C-u | C-c C-u |
| shift region right | mark region, C-c C-r | mark region, Python
Shift right |
| shift region left | mark region, C-c C-l | mark region, Python Shift
left |
DDD, available on many Unix systems (and freely downloadable if your system doesn't have it), is a GUI for many debuggers, such GDB (for C/C++), JDB (for Java), Perl's built-in Perl debugger, and so on. It can be used on PDB for Python, and thus make your usage of PDB more enjoyable and productive.
The first use of DDD on PDB was designed by Richard Wolff. He modified pdb.py slightly for this purpose, calling the new PDB pydb.py. It was designed for Python 1.5, but he has kindly provided an update for me. You'll need the files
http://heather.cs.ucdavis.edu/~matloff/Python/DDD/pydb.py http://heather.cs.ucdavis.edu/~matloff/Python/DDD/pydbcmd.py http://heather.cs.ucdavis.edu/~matloff/Python/DDD/pydbsupt.py
Place the files somewhere in your search path, say /usr/bin. Make sure that you give them execute permission, and that bdb.py from the Python library is in your PYTHONPATH.
To start, say for debugging fme2.py in Section 9, first make sure that main() is set up as described in that section.
When you invoke DDD, tell it to use PYDB:
ddd --debugger /usr/bin/pydb.py
Then in DDD's Console, i.e. the PDB command subwindow (near the bottom), type
(pdb) file fme2.py
Later, when you make a change to your source code, again issue the command
(pdb) file fme2.py
Your breakpoints from the last run will be retained.
Select Program Run as usual to set your command-line arguments, and
then run. (You may get a ``DDD: No Source'') popup error window, but
just click OK and ignore it.)
To set a breakpoint, right-click somewhere in blank space on the line in your source file and choose Set Breakpoint (or Set Temporary Breakpoint or Continue to Until Here, as the case may be).
To run, click on Program Run, fill in your program's command line
arguments if any in the Run with Arguments box, and click Run in that
pop-up window. You will be asked to hit Continue, which you could do by
clicking Program Run , but is more conveniently done by clicking
Continue in the little command summary window. (But don't use Run
there.)
You can then click on Next, Step, Cont etc. The marker for the current execution line is shaped like an `I', though rather faint when the mouse pointer is not in the source code section of the DDD window..
By the way, do not refer to sys.argv in freestanding code within a class. When your program is first loaded, any freestanding code will be executed, and since the command-line arguments won't have been loaded yet, so you will get an ``index out of range'' error. Avoid this by putting code involving sys.argv either inside a function in the class, or outside the class entirely.
You can inspect the value of a variable by moving the mouse pointer to any instance of the variable in the source code window.
As mentioned in Section A.1.3, if o is an object of some class, then printing o.__dict__ will print all the member variables of this object. In DDD, you can do this even more easily, as follows. Simply put that expression in a comment, e.g.
# o.__dict__
and then whenever you wish to inspect the member variables of o, simply move the mouse pointer to that expression in the comment!
Make good use of DDD's feature which allows a variable to be displayed continuously. Simply right-click on any instance of the variable, and then choose Display.
DDD, developed originally for C/C++, is not always a perfect match to Python. But since what DDD actually does is relay your clicked commands to PDB, as you can see in DDD's Console, whatever DDD can't do for you, you can type PDB commands directly into the Console.
The Winpdb debugger (www.digitalpeers.com/pythondebugger/),37 is very good. Among other things, it can be used to debug threaded code, curses-based code and so on, which many debuggers can't. Winpdb is a GUI front end to the text-based RPDB2, which is in the same package. I have a tutorial on both at http://heather.cs.ucdavis.edu/~matloff/winpdb.html.
I personally do not like integrated development environments (IDEs). They tend to be very slow to load, often do not allow me to use my favorite text editor,38 and in my view they do not add much functionality. However, if you are a fan of IDEs, here are some suggestions:
However, if you like IDEs, I do suggest Eclipse, which I have a tutorial for at http://heather.cs.ucdavis.edu/~matloff/eclipse.html. My tutorial is more complete than most, enabling you to avoid the ``gotchas'' and have smooth sailing.
There are various built-in functions in Python that you may find helpful during the debugging process.
Recall that class instances are implemented as dictionaries. If you have a class instance i, you can view the dictionary which is its implementation via i.__dict__. This will show you the values of all the member variables of the class.
Sometimes it is helpful to know the actual memory address of an object. For example, you may have two variables which you think point to the same object, but are not sure. The id() method will give you the address of the object. For example:
>>> x = [1,2,3] >>> id(x) -1084935956 >>> id(x[1]) 137809316
(Don't worry about the ``negative'' address, which just reflects the fact that the address was so high that, viewed as a 2s-complement integer, it is ``negative.'')
There is a very handy function dir() which can be used to get a quick review of what a given object or function is composed of. You should use it often.
To illustrate, in the example in Section 10.1 suppose we stop at the line
print "the number of text files open is", textfile.ntfiles
Then we might check a couple of things with dir(), say:
(Pdb) dir() ['a', 'b'] (Pdb) dir(textfile) ['__doc__', '__init__', '__module__', 'grep', 'wordcount', 'ntfiles']
When you first start up Python, various items are loaded. Let's see:
>>> dir() ['__builtins__', '__doc__', '__name__'] >>> dir(__builtins__) ['ArithmeticError', 'AssertionError', 'AttributeError', 'DeprecationWarning', 'EOFError', 'Ellipsis', 'EnvironmentError', 'Exception', 'False', 'FloatingPointError', 'FutureWarning', 'IOError', 'ImportError', 'IndentationError', 'IndexError', 'KeyError', 'KeyboardInterrupt', 'LookupError', 'MemoryError', 'NameError', 'None', 'NotImplemented', 'NotImplementedError', 'OSError', 'OverflowError', 'OverflowWarning', 'PendingDeprecationWarning', 'ReferenceError', 'RuntimeError', 'RuntimeWarning', 'StandardError', 'StopIteration', 'SyntaxError', 'SyntaxWarning', 'SystemError', 'SystemExit', 'TabError', 'True', 'TypeError', 'UnboundLocalError', 'UnicodeDecodeError', 'UnicodeEncodeError', 'UnicodeError', 'UnicodeTranslateError', 'UserWarning', 'ValueError', 'Warning', 'ZeroDivisionError', '_', '__debug__', '__doc__', '__import__', '__name__', 'abs', 'apply', 'basestring', 'bool', 'buffer', 'callable', 'chr', 'classmethod', 'cmp', 'coerce', 'compile', 'complex', 'copyright', 'credits', 'delattr', 'dict', 'dir', 'divmod', 'enumerate', 'eval', 'execfile', 'exit', 'file', 'filter', 'float', 'frozenset', 'getattr', 'globals', 'hasattr', 'hash', 'help', 'hex', 'id', 'input', 'int', 'intern', 'isinstance', 'issubclass', 'iter', 'len', 'license', 'list', 'locals', 'long', 'map', 'max', 'min', 'object', 'oct', 'open', 'ord', 'pow', 'property', 'quit', 'range', 'raw_input', 'reduce', 'reload', 'repr', 'reversed', 'round', 'set', 'setattr', 'slice', 'sorted', 'staticmethod', 'str', 'sum', 'super', 'tuple', 'type', 'unichr', 'unicode', 'vars', 'xrange', 'zip']
Well, there is a list of all the builtin functions and other attributes for you!
Want to know what functions and other attributes are associated with dictionaries?
>>> dir(dict) ['__class__', '__cmp__', '__contains__', '__delattr__', '__delitem__', '__doc__', '__eq__', '__ge__', '__getattribute__', '__getitem__', '__gt__', '__hash__', '__init__', '__iter__', '__le__', '__len__', '__lt__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__setitem__', '__str__', 'clear', 'copy', 'fromkeys', 'get', 'has_key', 'items', 'iteritems', 'iterkeys', 'itervalues', 'keys', 'pop', 'popitem', 'setdefault', 'update', 'values']
Suppose we want to find out what methods and attributes are associated with strings. As mentioned in Section 7.2.3, strings are now a built-in class in Python, so we can't just type
>>> dir(string)
But we can use any string object:
>>> dir('')
['__add__', '__class__', '__contains__', '__delattr__', '__doc__',
'__eq__', '__ge__', '__getattribute__', '__getitem__', '__getnewargs__',
'__getslice__', '__gt__', '__hash__', '__init__', '__le__', '__len__',
'__lt__', '__mod__', '__mul__', '__ne__', '__new__', '__reduce__',
'__reduce_ex__', '__repr__', '__rmod__', '__rmul__', '__setattr__',
'__str__', 'capitalize', 'center', 'count', 'decode', 'encode',
'endswith', 'expandtabs', 'find', 'index', 'isalnum', 'isalpha',
'isdigit', 'islower', 'isspace', 'istitle', 'isupper', 'join', 'ljust',
'lower', 'lstrip', 'replace', 'rfind', 'rindex', 'rjust', 'rsplit',
'rstrip', 'split', 'splitlines', 'startswith', 'strip', 'swapcase',
'title', 'translate', 'upper', 'zfill']
For example, let's find out about the pop() method for lists:
>>> help(list.pop)
Help on method_descriptor:
pop(...)
L.pop([index]) -> item -- remove and return item at index (default
last)
(END)
And the center() method for strings:
>>> help(''.center)
Help on function center:
center(s, width)
center(s, width) -> string
Return a center version of s, in a field of the specified
width. padded with spaces as needed. The string is never
truncated.
Hit 'q' to exit the help pager.
You can also get information by using pydoc at the Unix command line, e.g.
% pydoc string.center [...same as above]
The above methods of obtaining help were for use in Python's interactive mode. Outside of that mode, in an OS shell, you can get the same information from PyDoc. For example,
pydoc sys
will give you all the information about the sys module.
For modules outside the ordinary Python distribution, make sure they are in your Python search path, and be sure show the ``dot'' sequence, e.g.
pydoc u.v
Before Python 2.2, there was no provision for class methods. But there was simple workaround, as seen in http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/52304.
Since Python now allos class methods (two kinds, in fact), we don't need to use the workaround in our programming, but it gives us an excellent example of how object references work, so let's take a look.
In the program tfe.py in Section 10.1, we could add
[fontsize=\relsize{-2},numbers=left]
class ClassMethod:
def __init__(self,MethodName):
self.__call__ = MethodName
class textfile:
(lots of stuff omitted)
def printntfiles():
print textfile.ntfiles
tmp = ClassMethod(printntfiles)
printntfiles = tmp
textfile.printntfiles()
(I've added a line here--the one which is an assignment to tmp--for the sake of clarity, as you will see below.)
Recall from Section 11 that functions in Python are (pointers to) objects like anything else, and thus assignable. For example, look at:
>>> def f(x): ... return x*x ... >>> g = f >>> g(3) 9
In other words, f is really just a pointer to a ``function object,'' so if we assign f to g, then g points to that object too, and consequently g is the same function.
Second, it is important to know that class instances are callable. This is where the __call__() method, implicitly part of every class, comes in. One simply pretends that the class instance is a function, and then calls it. That triggers execution of the __call__() method on self (i.e. the class instance which we are ``calling'') with whatever arguments we give it. The __call__() method is empty by default, but we can supply one, and use it to our advantage.
For example, say this is the source file yyy.py:
[fontsize=\relsize{-2},numbers=left]
class u:
def __init__(self):
self.r = 8
def __call__(self,z):
return self.r*z
def main():
a = u()
print a(5)
a.r = 13
print a(5)
if __name__ == '__main__':
main()
python yyy.py 40 65
Now, look at the line from the workaround code above,
tmp = ClassMethod(printntfiles)
This creates an instance of the class ClassMethod and assigns it to tmp. The parameter was printntfiles, so the constructor line
self.__call__ = MethodName
will mean that tmp.__call__ is set to printntfiles. Forget just for a moment that both tmp.__call__ and printntfiles are functions; just focus on the fact that we have assigned one pointer (printntfiles) to another (tmp.__call__).
But of course both of those are indeed functions, and so tmp.__call__ is now a pointer to the code for the original printntfiles function, i.e. to the code
print textfile.ntfiles
Now, the next line
printntfiles = tmp
means that the variable printntfiles no longer points to its own code! Instead, it now points to an instance of the class ClassMethod.
Now, remember, printntfiles itself is a class variable in our class textfile, i.e. its full name is textfile.printntfiles. Originally, it had been a function, but remember, anything can be assigned, so now it is simply a variable, pointing to an instance of the class ClassMethod.
The point is that if we now call textfile.printntfiles--which we can do, because any Python class instance is ``callable''!--then we invoke the __call__() method of that instance. And that method, as seen above, points to the original code for our printntfiles function, i.e.
print textfile.ntfiles
Therefore that code will be executed!
So, what looks syntactically like a call to a class method,
textfile.printntfiles()
will in fact act like a class method, even though Python has no such thing.
As mentioned in Section 5, instead of using the keyword global, we may find it clearer or more organized to group all our global variables into a class. Here, in the file tmeg.py, is how we would do this to modify the example in that section, tme.py:
[fontsize=\relsize{-2},numbers=left]
# reads in the text file whose name is specified on the command line,
# and reports the number of lines and words
import sys
def checkline():
glb.linecount += 1
w = glb.l.split()
glb.wordcount += len(w)
class glb:
linecount = 0
wordcount = 0
l = []
f = open(sys.argv[1])
for glb.l in f.readlines():
checkline()
print glb.linecount, glb.wordcount
Note that when the program is first loaded, the class glb will be executed, even before main() starts.
This document was generated using the LaTeX2HTML translator Version 2002-2-1 (1.71)
Copyright © 1993, 1994, 1995, 1996,
Nikos Drakos,
Computer Based Learning Unit, University of Leeds.
Copyright © 1997, 1998, 1999,
Ross Moore,
Mathematics Department, Macquarie University, Sydney.
The command line arguments were:
latex2html PythonIntro.tex -split 0
The translation was initiated by Norm Matloff on 2008-09-17
But if you do type these examples yourself, make sure to type exactly what appears here, especially the indenting. The latter is crucial, as will be discussed later.
On the other hand, true arrays can be accessed more quickly. In C/C++,
the 
element of an array X is i words past the beginning
of the array, so we can go right to it. This is not possible with
Python lists, so the latter are slower to access. The NumPy add-on
package for Python offers true arrays.
By the way, watch out for Python statements like print a or b or c, in which the first true (i.e. nonzero) expression is printed and the others ignored; this is a common Python idiom.
ntfiles = 0
would still have been executed when we first started execution of the program. As mentioned earlier, the Python interpreter executes the file from the first line onward. When it reaches the line
class textfile:
it then executes any free-standing code in the definition of the class.
By the way, the reason your breakpoints are retained is that of course they are variables in PDB. Specifically, they are stored in member variable named breaks in the the Pdb class in pdb.py. That variable is set up as a dictionary, with the keys being names of your .py source files, and the items being the lists of breakpoints.