What's wrong with Perl

By: Lars Marius Garshol

Belorussian translation, by Bohdan Zograf.

Estonian translation by Adrian Pantilimonu

Contents

Before we start

Just a note before you continue: this is based on my personal experience with Perl. I know other people have other opinions of this language and they are welcome to them. I just want to present mine, because I see lots of messages on Usenet from people who seem to be about to learn Perl and I keep wanting to tell them that perhaps it's not a good idea. So, I wrote this article to get this off my chest once and for all.

If you think that anything in this article is objectively wrong, then please email me about it. I'd like this article to be as factually correct as possible. If you just disagree with me you can tell me that too.

I should perhaps explain why I refer to Perl as "the Camel": the Bible on all things Perl is Larry Walls "Programming Perl", which is published by O'Reilly. O'Reilly usually puts a nice 19th century engraving of an animal on the cover of their books. "Programming Perl" got a camel and has been known as "the camel book" ever since. Larry Wall also often refers to Perl as the camel.

The article should be up to date as of Python 1.5.2 and Perl 5.005. If anything has changed since those versions, feel free to tell me about it.

Getting to know the Camel

I learned Perl 5 in early '97. I downloaded Patrick M. Ryans excellent introduction to Perl and found that a lot of the things I'd been doing the hard way with C, Pascal and Java were much much easier in Perl. For text processing and access to system functions Perl looked like a real God-send. Great, I thought, and bought myself "Learning Perl", by Randal Schwartz. (Also known as the "llama book".)

I read it pretty quickly and sucked up all these wonderful new features. The first program I made read a web server log and counted the number of times each page had been accessed. I wrote it in half an hour and it worked immediately! Not only did it work, but Perl also seemed to be able to overlook unimportant errors instead of crashing or aborting like C/Pascal/Java programs would.

(That program was released in many versions, and had quite a few users. So Perl isn't useless, just inconvenient.)

Getting to dislike the Camel

I was really enchanted with this language and started using it more and more. However, as I kept using it I kept discovering new things I didn't like. After a while it added up to a pretty sizable list.

Illegible syntax

Though I'll admit readability suffers slightly...

--Larry Wall in <2969@jato.Jpl.Nasa.Gov>

One of the first things I discovered I didn't like was the syntax. It's very complex and there are lots of operators and special syntaxes. This means that you get short, but complex code with many syntax errors that take some time to sort out.

It also means that reading someone elses code is difficult. You can't easily understand someone elses scripts and adapt them to your own needs, nor can you easily take over the maintenance of someone elses program.

I write pretty clean Perl code, because I stay away from most of the obfuscating features, but even so it gets pretty hard to read. This is one ordinary example:

foreach $Key (@SearchEngines) {
    if ($fields[11] =~ /$Key/i) {
	  $HitFrom[4]++;           #Yes, search engine
	  $SEReferrals{$SENames{$Key}}++;
	  $PageRefs{$fields[6]}{$SENames{$Key}}++;
	  $found=1;
	  last;
    }
}

It's not even a particularly bad example. Here's a function from the Perl distribution that produces a Soundex value from a name or word:

sub soundex
{
  local (@s, $f, $fc, $_) = @_;

  push @s, '' unless @s;	# handle no args as a single empty string

  foreach (@s)
  {
    tr/a-z/A-Z/;
    tr/A-Z//cd;

    if ($_ eq '')
    {
      $_ = $soundex_nocode;
    }
    else
    {
      ($f) = /^(.)/;
      tr/AEHIOUWYBFPVCGJKQSXZDTLMNR/00000000111122222222334556/;
      ($fc) = /^(.)/;
      s/^$fc+//;
      tr///cs;
      tr/0//d;
      $_ = $f . $_ . '000';
      s/^(.{4}).*/$1/;
    }
  }

  wantarray ? @s : shift @s;
}

However, even this isn't really bad. The Perl Journal has conducted an Obfuscated Perl contest. The winners are here . Be warned, though. These programs gives the word unreadable entirely new and previously unimagined meanings. (And no, this isn't an argument for Perl being unreadable, but mainly included as a funny and curious item.)

Programmers and readability

Some people reading this have complained that 'But anyone can write unreadable code in any language!' and this is certainly true. However, some languages seem to encourage hard-to-read code, while others seem to discourage it.

From what I have seen of my own and other people's code Perl really encourages hard-to-read programs. The Soundex example above comes from the Perl distribution and was found by just randomly looking through 2 or 3 files. Looking through it again now I see many other examples (like lib/pod/text.pm and lib/file/copy.pm), even though most scripts are too short to be hard to read.

Some have argued that Perl is more like natural languages than most programming languages, and this certainly seems correct to me. And to me that is a disadvantage: natural language is extremely complex, ambiguous and full of subtle nuances and meanings. I certainly don't want my programs to be like that, but it seems that some do. I guess the reader will have to find out for him/herself which category s/he belongs in.

Too many special constructs

Many of Perl's features are built right into the language core itself, instead of being placed in separate libraries. One example of this is the handling of regular expressions, for which there is a special syntax. This has the advantage that there are some convenient ways of doing the things that are done most often, but it also means that you don't get the advantages of objects.

To take one example, Perl has a special construct called formats, which are a sort of templates you can use to generate nice textual reports. Quite handy, but built into the language. So, you can't create a list of formats, return them from functions and so on, which will in many cases be a serious inconvenience.

I think you can do these things with file handles, but since they are also handled as special cases I've never been able to figure out how. I tried using references, but never made it work.

Hard to build data structures

In the Perl documentation there is a separate manual page for how to create arrays within arrays and hashes within hashes. And it's really necessary. I had a lot of pain trying to figure out how to do this, even after reading it several times. This is something that really shook me, because in other languages this is something you just do, without thinking about it.

In Lisp, this sets the variable a to a list:

(setq a '(1 2 3 4))

Here we create list b where the first element is another list:

(setq b '((0.8 0.9 1) 2 3 4))

Here's the first list in Perl:

@a=(1,2,3,4);

and here's the second:

@b=((0.8,0.9,1),2,3,4);

(The @s before the variable names tells Perl that these are array variables.) That wasn't so bad, was it? Well, let's try to use this.

To pick out the first element of the first list in Lisp, you just write

(first a)

and Lisp gives you

1

To get the first element of the second list you write

(first b)

and Lisp gives you

(0.8 0.9 1)

Let's try this in Perl.

$a[0]

gives us

1

The $ before the variable name tells Perl that we want a single value (scalar in Perl lingo), not an array. The [0] tells Perl that we want the first value of the array. Perl, like many other languages and APIs counts from 0.

Then we do

$b[0]

and Perl happily gives us

0.8

That's right, Perl has broken into the list inside the b list and retrieved the first value of it. Or, rather, it flattened b into one list when we created it, so it's now really one consecutive list with 6 elements.

To do this right we should have written

@b=([(0.8,0.9,1)],2,3,4);

when we created the list. The []s enter a reference to the inner list as the first element of the outer list instead of flattening the inner list into the outer one.

OK. So we try again:

$b[0]

gives us

ARRAY(0xb75eb0)

So obviously we manage to find the array, but something still goes wrong along the way. The problem is that we use $b, which makes Perl think that we want a scalar and so it gives us a reference to the array instead of the array itself (which is not a scalar).

Aha! Of course! We must use

@b[0]

because @ tells Perl we want an array value. Not so. We get

ARRAY(0xb75eb0)

once again. I've never managed to understand why this is so and at this point I gave up on the entire thing.

Some weeks later I saw a helpful posting on no.perl: one should request a reference to the array, like this

@{$b[0]}

which actually gives us

(0.8 0.9 1)

So now I can write code with arrays inside arrays and hashes inside hashes.

Now, ask yourself: do you really think you should have to go through all this in order to put one list inside another?

Defining interfaces

Another major disadvantage to Perl is that of function (or subroutine, in Perl lingo) signatures, or rather, the lack of signatures. In most programming languages when you declare a function you also declare its signature, listing the names of the parameters and in some languages also their types. Perl doesn't do this.

So what in Java is

public String substring(String str, int from, int to) {

becomes

sub substring {
  local($str, $from, $to) = @_;

in Perl. In other words, you have to manually decode the parameter list. Perl has lately been extended with the notion of prototypes, which means that you can write

sub substring($, $, $) {
  local($str, $from, $to) = @_;

and have Perl check that the number of arguments is correct. This is not required, though, and there is much Perl code that does not use this syntax.

The disadvantages don't stop there, though. Many programmers don't destructure the parameter array like in the example above, which makes the code much harder to read at a glance, and this also makes it impossible to automatically generate good documentation.

And, what's more, you don't get the advantages more advanced languages have from features such as keyword arguments (without re-implementing them yourself with a hash). For example, when you want to create a hash table in Common Lisp you call the make-hash-table function, which takes the following keyword arguments:

This means that all of the following invocations will create hash tables correctly:

  (make-hash-table)
  (make-hash-table :test #'eq)
  (make-hash-table :size 1000)
  (make-hash-table :rehash-size 2.0 :rehash-threshold 0.7)

It also means that you can have function which take a large number of parameters (make-array takes 7) and still keep both readability and ease of use. You can do the same in Perl, but you are certainly not encouraged to, documentation tools won't understand it, readers may not either and it certainly isn't convenient compared to the way it is in Common Lisp:

(defun make-hash-table(&key test size rehash-size rehash-threshold)

Not really object-oriented

Although object-orientation is not as fantastic as many would like us to think, Perl does support it. However, it does so only half-heartedly, since objects were added rather late in the life of the language.

The result of this is that normal files, sockets, hashes and lists are not objects, which means that the interfaces to them are not as convenient as they could have been. Newer versions of Perl come with object-oriented modules with wrappers for these kinds of objects, which means that Perl has a protocol for such objects and you can write your own implementations of these protocols. However, it also means that you need to distinguish between ordinary file handles and file objects, which is a bit inconvenient.

Another thing is the fact that when creating objects you need to explicitly manage the internals of your objects. In Perl, object creation is manual. A class is declared as a package, and the functions in the package then become the methods of the class. To create an object, you make a hash table and then bless it (using the built-in function 'bless') to make it an object. The 'perlobj' man page, which explains the Perl object features, recommends this form of object initializer:

package MyClass;
sub new {
    my $class = shift;
    my $self = {};
    bless $self, $class
    $self->initialize();  # do initialization here
    return $self;
}

There are other ways of doing object initialization, some of which cause problems for inheritance. Personally, I find it amazing that this sort of thing should be necessary at all. The above is equivalent to this Python code:

class MyClass:
    pass

The result is that one can easily get object construction wrong (such as by not catering for inheritance), defining classes is awkward and it's hard to tell from code when a class is defined (for both human readers and software documentation tools).

Summing it up

In general, what this means is that Perl is a large and complex language, which takes a long time to learn properly. In my opinion, this complexity is unnecessary and a simpler language would have been much better. I think this also means that many non-expert Perl developers write suboptimal code.

Another thing is that I think few Perl developers (percentage-wise) write general and reusable modules, because you need to learn the language well before doing so, something that is relatively hard and takes time. Another thing is that the language itself does not encourage this.

Discovering the Python

Programming languages teach you not to want what they cannot provide. You have to think in a language to write programs in it, and it's hard to want something you can't describe. When I first started programming - in BASIC - I didn't miss recursion, because I didn't know there was such a thing. I thought in BASIC. I could only conceive iterative algorithms, so why should I miss recursion?

--Paul Graham, ANSI Common Lisp.

So, after discovering all these bad things about Perl, what did I do? I kept using it. After all, as bad as it was, it was still better than doing text processing and web programming with C, Pascal or Java, and there were no better alternatives.

Or so I thought. At the University bookstore there was this book called "Internet Programming with Python". Being both a language freak and an internet freak I thought this was interesting and picked it up. Somewhere in the beginning of the book there was an anecdote about a Python programmer who wrote all his Python programs so that when an error occurred (ie: when an exception was thrown) the error handler called his beeper.

Wow, I thought. This sounds interesting. So I went home, found the Python tutorial, printed it out and started playing with Python. That night I wrote a POP3 client library in Python. Just like that. After going to bed I had an idea: wouldn't it be a lot nicer if I cached the messages and made this invisible to the user? In the morning I added that in half an hour.

I've since used this library to delete email without downloading it, moving 150 emails from one POP account to another and many other things. (Yes, I made a small SMTP library as well.) I can even use it as an email client using the Python interpreter as a command line.

I kept using Python more and more after this. I wrote a link checker that went over my web pages checking them for errors and kept adding more protocols and features to it. After a while I thought: this program spends a lot of time waiting for server responses. Maybe I can speed it up by using multi-threading so that it can wait for several servers at the same time?

I'd never really used multithreading before, but knew the theory behind it. I added this to the link checker in an hour and that includes the digging in the library documentation and removing the few bugs I did introduce. (Multithreading is much more complex than it sounds at first because things happen simultaneously. That's not Python's fault, though.)

Having read this you may now be convinced that I'm a master programmer, rather than that Python is a great programming language for this kind of thing. Personally, I don't think this is true. (Remember, I'm the guy who can't even make a Perl subroutine return a file handle.) Also, from what I hear, many other people have had similar experiences with Python. Here's one example:

In my first 15 minutes programming Python I wrote a program which would download all the articles in a newsgroup into an mh folder for me - and comp.lang.python was my first newsgroup!

--Thomas Corbin, in private email to the author

Falling in love with the Python

The standard libraries

Did it stop there? It certainly didn't. Since then I've discovered these things in the Python libraries:

And these are things that come with the standard Python distribution! Perl also has most of this stuff, but it doesn't come with the interpreter and the quality of documentation and interfaces varies.

In Python you not only get these things delivered with the interpreter, complete with documentation, they are also extremely simple to use and provide exactly the sort of things one commonly wants. Say you're writing a web robot and the robot has the URL to the current page in a string (cur_url) and a relative URL from this page to the next page in another (next_url) and you want to compute the absolute URL of the next page. This is the code:

next_url=urlparse.urljoin(cur_url,next_url)

Python also supports GUI programming via Tk on Win32, Mac and Unix. It's really easy to install, but not too well documented. There are also at many other ways to do GUI programming with Python. (Yes, there is for Perl as well.)

The SIGs

Python may become as big as Perl or Tcl. It is more "lovable" than those -- though perhaps also more controversial. It has an extremely supportive user community, and that is what will make it big.

--Guido van Rossum, creator of Python

When they want to create new libraries or "standards", the Pythoners form Special Interest Groups of volunteers, which anyone can join. Some of the results of this have so far been a common API for database modules (which means you can use the Sybase module and then exchange it with the Oracle one and only change 2 lines of code), an IDL-to-Python mapping for CORBA, a common text format for documentation strings and common tools and APIs for XML parsing are under development.

Python software

But, if this language really is so great, shouldn't people have written some pretty nifty software in it? Well, they have. Here are some examples:

Personally, my current project in Python is a fully validating XML parser .

Is it really any better than Perl?

I've included this section in case you want to know how Python handles the things I complained about in Perl.

Readability

Python is the most readable language I've ever programmed in. It took me half an hour to understand Medusa (even though it's pretty weird in concept) and another half-hour to change it so that I could map URL paths in the web server to Python functions. An hour after that I could read my Gnus mail boxes through the web server. Another hour, and I could read news through it.

Here's my own implementation of the soundex function, written in November '97, when I was still new at this:

# no_tbl is an array I've constructed that maps characters to numbers
# is_letter is written by me, but I've since discovered that Python
# has it

def soundex(string):
    """Returns the Soundex code of the string. Characters not A-Z skipped."""
    string=lower(string)   

    if not is_letter(string[0]): string=string[1:]
    last=no_tbl[ord(string[0])-97]
    res =upper(string[0]) # This is where the result will end up

    for char in string[1:]:
	if is_letter(char):
	    new=no_tbl[ord(char)-97]
	    if (new!="0" and new!=last):
		res=res+new
	    last=new

    if len(res)<4:
	return res+"0"*(4-len(res))
    else:
	return res[:4]

Documentation

There is a Python program that generates javadoc-like documentation from the documentation strings in Python programs. It can produce output in HTML, plain text and FrameMaker format.

As for the standard library documentation, you can see for yourself .

Turning off output buffering isn't blindingly obvious, but not too difficult, either. You pretty quickly learn that sys.stdout, sys.stdin and sys.stderr are file objects that represent standard out, standard in and standard error. So, since stdout is an ordinary file object, it should behave as one with respect to output buffering as well. And it does. These objects have no methods to turn off buffering, but you can flush data with the flush method:

sys.stdout.flush()

If you find that awkward there is a command-line switch for the interpreter that lets you turn off buffering. However, you have to run the interpreter with 'python -?' to find this out, so I couldn't really claim that this is too well documented.

Built-ins

Python has very few built-in features compared to Perl. Regular expressions come in a separate module and are implemented as objects. Here's a list of compiled regular expressions:

reg_list=[re.compile("foo"),re.compile("[Bb][Aa][Rr]"),re.compile("ba[z]")]

Files are also objects, as are Python modules and most other things. This means that you can pass them as parameters, stuff them in lists, subclass them and even create classes with the same methods and use them where code expects to see a file object or a module.

Data structures

This is the two list examples in Python, (=> shows what the results are):

a=[1,2,3,4]
a[0]                  => 1
b=[[0.8,0.9,1],2,3,4]
b[0]                  => [0.8,0.9,1]

Python also has hash tables, tuples (fixed-length lists) and full object-orientation.

Parting shot

What other people think

You may want to hear what other people think as well, so here are some links to other opinions on Perl and Python:

A final warning

Camels are docile when properly trained and handled but, especially in the rutting season, are liable to fits of rage. They spit when annoyed and can bite and kick dangerously.

--Encyclopedia Britannica.

If after all this, you still want to use Perl, then be my guest and go right ahead. I'm not going to stop you. You have been warned, though.

However, you may perhaps be curious about what the Encyclopedia Britannica has to say about Pythons? Here it is:

...the Monty Python Flying Circus troupe, which set a standard during the 1970s for its quirky parodies and wacky humour on television and later in films.

--Encyclopedia Britannica

That's right. Python was named after a comedy troupe, not after a nasty reptile. This is why there are so many weird names and examples in the Python tutorials and source code. :-)

Appendices

Acknowledgements

Thanks to:


Last update 2002-01-16.