Revision 19 as of 2004-12-28 02:58:26

Clear message

This page follows a [http://groups.google.ca/groups?hl=en&lr=&selm=mailman.8378.1103832808.5135.python-list%40python.org&prev=/groups%3Fq%3Ddubious%2BFernando%26hl%3Den%26lr%3D%26group%3Dcomp.lang.python.*%26selm%3Dmailman.8378.1103832808.5135.python-list%2540python.org%26rnum%3D2 suggestion] of FernadoPerez on comp.lang.python. His suggestion was that there should somewhere be a collection of bad python practices together with an explanation of the badness and a preferred alternative. This seems a good job for a wiki-page, so ...

Premature Optimization

Premature Optimization is spending effort on execution efficiency before determining which parts of the code are actually significant to the program efficiency.

Dubious Way

This occurs in a variety of contexts, all of which involve spending extra time making code run faster before first writing a simple, concise implementation that produces the correct results.

Example of over-optimizing attribute accesses:

   1 app = lst.append
   2 for item in iterable:
   3     app(func(item))

The Problem

While a correctly applied optimization can indeed speed up code, optimizing code that is only seldom used can waste significant development time, and can make code harder to read. Optimizations should only be sought when a programmer has isolated (using a profiler, etc.) a significant bottleneck in program efficiency. Write correct code first, then make it fast (if necessary).

Preferred Way

Don't optimize until necessary.

Example of non-optimized but more readable attribute access:

   1 for item in iterable:
   2     lst.append(func(item))

The above examles look equally readable to me. In fact, the binding of a function to an object

app = lst.append

enhances readiblity and maintainability if used more than once:

app("a")
app("b")

is preferred over

lst.append("a")
lst.append("b")

since the binding of lst and append happens twice - copy and paste.

String Concatenation

String concatenation is building up a relatively lengthy string from a collection of strings.

Dubious Way

Newcomers to Python often try to build strings up like this:

   1 >>> string_list = ['one', 'big', 'string', 'in', 'pieces']
   2 >>> new_string = "" 
   3 >>> for s in string_list: 
   4         new_string = new_string + s
   5 >>> print new_string 
   6 onebigstringinpieces 
   7 >>>

The Problem

This is slow and resource heavy. Each time through the for loop, a new string is built and the old one is discarded. That might not matter so much for such a small case, but as the number of elements to be joined creeps up, so too does the inefficiency.

The above code is perfectly fine. Its readible, and it works. Worrying about speed/performance when it is not an issue is one of the worst programming practices. "Premature Optimization is the Root of All Evil" -- see PrematureOptimization

Preferred Alternatives

String Formatting

For short cases where the number of strings to be joined is known, you can use string formatting as follows:

   1 >>> print '%s%s%s%s%s' % tuple(string_list)
   2 onebigstringinpieces

This is much more efficient, but is also rather more limited in the range of circumstances to which it applies. It could be made more general by constructing the formatting string as a function of len(string_list), but this would be a bit dubious, too, when we have. Its also less readible, less maintainible.

The join method of strings

The join method of the string type lets you perform the concatenation as follows:

   1 >>> print "".join(string_list)
   2 onebigstringinpieces

This is quite efficient and perfectly general as it applies to any arbitrary list of strings. (You don't need to know the list length in advance.)

The major thing to puzzle the newcomer here is why "".join(some_list) rather than some_list.join(). The way to think of this is that you are using the string "" to join the elements of some_list. Hence,

   1 >>> print 'JOINT'.join(string_list)
   2 oneJOINTbigJOINTstringJOINTinJOINTpieces

That said, some do consider this aspect of the join method of strings odd enough to count as a PythonWart.

If the ''.join(some_list) syntax really bothers you, one option is to bind the method to a different name, e.g.

   1 >>> join = ''.join
   2 >>> print join(string_list)
   3 onebigstringinpieces
   4 >>> 
   5 >>> underscore_join = '_'.join
   6 >>> print underscore_join(string_list)
   7 one_big_string_in_pieces

Boolean Redundancy

(Please name me better!)

Among the most common tasks in programming is to test if a condition obtains and act accordingly. It is common for newcomers to Python to adopt an all-together overly verbose idiom for this.

Dubious Way

   1 if (count > 10) == True:
   2     # continue process

and

   1 def count_tester(count):
   2     if count > 10:
   3         return True
   4     else:
   5         return False

The Problem

There is a (very slight) speed of execution inefficiency in these examples. But much more important is the speed of entry and understanding inefficiency. All other things being equal, extra typing is evil. And, unless some gain in clarity is purchased by the extra characters, the more characters in the code, the longer that code will take to understand. (The programming time you save could well be your own!)

Preferred Alternatives

   1 if count > 10:
   2     # continue process
   3 
   4 def count_tester(count):
   5     return count > 10

Overuse of lambda

Lambda forms allow anonymous functions to be created and used as part of an expression. However, when a function is already named, wrapping this function in a lambda can decrease readability and affect program efficiency.

Dubious Way

   1 dict(a=lambda x: str(x),
   2      b=lambda x: some_dict.get(x))

The Problem

Using a lambda when a function is already named incurs the extra overhead of one function call, which is generally undesirable as function calls are relatively expensive in Python. Even setting execution efficiency aside, the lambda-less versions are preferred because they are generally more concise and easier to read.

Preferred Alternatives

   1 dict(a=str,
   2      b=some_dict.get)

Counting Items without Enumerate

It is often useful to keep track of the index of an item in an iterable, for example, for reporting the line number of a string in a file. As of Python 2.3 the preferred way to do this is using the builtin enumerate instead of a manually updated count variable. In versions of Python before 2.3, this is actually not all that dubious... ;-)

Dubious Way

   1 count = 0
   2 for item in iterable:
   3     try:
   4         do_something(item)
   5     except Exception:
   6         raise Exception('error on item %r' % count)
   7     count += 1

The Problem

While timings are comparable, manual update is more verbose and has a greater risk of programming error if the programmer forgets to update the count variable.

Preferred Alternatives

UsingEnumerate:

   1 for count, item in enumerate(iterable):
   2     try:
   3         do_something(item)
   4     except Exception:
   5         raise Exception('error on item %r' % count)

Overuse of Regular Expressions

Regular Expressions provde a powerful tool for doing complicated string searches. However, for simple string searches, regular expressions are often overkill.

(Note that such overuse of regular expressions is often the result of converting Perl code to Python code.)

Dubious Way

   1 matcher = re.compile(r'defg')
   2 if matcher.search(s):
   3     do_something()

   1 matcher = re.compile(r'(\S+)')
   2 words = matcher.findall(s)

The Problem

For simple tasks, using a regular expression can add unnecessary overhead from compiling the regular expression and using the match object to search. When applicable, using string methods can often be faster and more concise.

Preferred Alternatives

   1 if 'defg' in s:
   2     do_something()

Note that in pre-2.3 Pythons 'in' only worked with single character strings. In Python 2.3 and above, 'in' works with multi-character substrings as above.

   1 words = s.split()

Unable to edit the page? See the FrontPage for instructions.