Quality matters. Programs have bugs. Programs need maintenance and expansion. Programs have multiple programmers.
To program well, we must find the balance between getting the job done now and making our lives easier in the future. We must balance time, resources, and quality. The craft of programming demands that we do that to the best of our ability.
To write Perl well, we must understand the language. We must also cultivate a sense of good taste for the language and the design of programs. The only way to do so is to practice maintaining code and reading and writing great code. This path has no shortcuts, but it does have guideposts.
Maintainability is the nebulous measurement of how easy it is to understand and modify a program. Write some code. Come back to it in six months. How long does it take you to fix a bug or add a feature? That's maintainability.
Maintainability doesn't measure whether you have to look up the syntax for a builtin or a library function. It doesn't measure how someone who has never programmed before will or won't read your code. It's more interesting to ask whether a competent programmer who understands the problem you're trying to solve will find it easy or difficult to modify the program. What problems get in the way of fixing a bug or adding an enhancement correctly?
To write maintainable software, you need experience solving real problems, an understanding of the idioms and techniques and style of your programming language, and good taste. You can develop all of these by concentrating on a few principles:
Well-designed systems have little duplication. They use functions, modules, objects, and roles to extract duplicate code into distinct components which accurately model the domain of the problem. The best designs allow you to add features by removing code.
Some problems require clever solutions. When this happens, encapsulate this code behind a simple interface and document your cleverness.
Sometimes you need powerful, robust code. Sometimes you need a one-liner. Simplicity means knowing the difference and building only what you need. This is no excuse to avoid error checking or modularity or validation or security. Simple code can use advanced features. Simple code can use CPAN modules, and many of them. Simple code may require work to understand. Yet simple code solves problems effectively, without unnecessary work.
Perl borrows liberally from other languages. Perl lets you write the code you want to write. C programmers often write C-style Perl, just as Java programmers write Java-style Perl. Effective Perl programmers write Perlish Perl, embracing the language's idioms.
CPAN developers, Perl Mongers, and mailing list participants have hard-won experience solving problems in myriad ways. Talk to them. Read their code. Ask questions. Learn from them and let them learn from you.
Perl::Critic
), reformatting (Perl::Tidy
), and private distribution systems (CPAN::Mini
, Carton
, Pinto
). Take advantage of the CPAN infrastructure; follow the CPAN model of writing, documenting, packaging, testing, and distributing your code.Writing maintainable code means designing maintainable code. Good design comes from good habits:
If you find a bug, report it. Patch it, if possible. Fix a typo. Ask for a feature. Say "Thank you!" We are better together than we are separately. We are powerful and effective when we reuse code.
When you're ready—when you create something new or fix something old in a reusable way—share your code. Join us. We solve problems.
Good programmers anticipate the unexpected. Files that should exist won't. A huge disk that will never fill up will. The always-up network goes down. The unbreakable database breaks. Exceptions happen. Robust software must handle them. If you can recover, great! If you can't, log the relevant information and retry.
Perl handles exceptional conditions through exceptions: a dynamically-scoped control flow mechanism designed to raise and handle errors.
Suppose you want to write a log file. If you can't open the file, something has gone wrong. Use die
to throw an exception ... or let autodie
(The autodie Pragma) add it for you, in this case.:
sub open_log_file
{
my $name = shift;
open my $fh, '>>', $name
or die "Can't open logging file '$name': $!";
return $fh;
}
die()
sets the global variable $@
to its operand and immediately exits the current function without returning anything. This thrown exception will continue up the call stack (Controlled Execution) until something catches it. If nothing catches the exception, the program will exit with an error.
Exception handling uses the same dynamic scope (Dynamic Scope) as local
symbols.
Sometimes allowing an exception to end the program is useful. A program run as a timed process might throw an exception when the error logs have filled, causing an SMS to go out to administrators. Yet not all exceptions should be fatal. Your program might need to recover from some exceptions. Other exceptions might only give you a chance to save your user's work and exit cleanly.
Use the block form of the eval
operator to catch an exception:
# log file may not open
my $fh = eval { open_log_file( 'monkeytown.log' ) };
If the file open succeeds, $fh
will contain the filehandle. If it fails, $fh
will remain undefined, and program flow will continue.
The block argument to eval
introduces a new scope, both lexical and dynamic. If open_log_file()
called other functions and something eventually threw an exception, this eval
could catch it.
An exception handler is a blunt tool. It will catch all exceptions thrown in its dynamic scope. To check which exception you've caught (or if you've caught an exception at all), check the value of $@
. Be sure to local
ize $@
before you attempt to catch an exception, as $@
is a global variable:
local $@;
# log file may not open
my $fh = eval { open_log_file( 'monkeytown.log' ) };
# caught exception
if (my $exception = $@) { ... }
Copy $@
to a lexical variable immediately to avoid the possibility of subsequent code clobbering the global variable $@
. You never know what else has used an eval
block elsewhere and reset $@
.
$@
usually contains a string describing the exception. Inspect its contents to see whether you can handle the exception:
if (my $exception = $@)
{
die $exception
unless $exception =~ /^Can't open logging/;
$fh = log_to_syslog();
}
Rethrow an exception by calling die()
again. Pass the existing exception or a new one as necessary.
Applying regular expressions to string exceptions can be fragile, because error messages may change over time. This includes the core exceptions that Perl itself throws. Instead of throwing an exception as a string, you may use a reference—even a blessed reference—with die
. This allows you to provide much more information in your exception: line numbers, files, and other debugging information. Retrieving information from a data structure is much easier than parsing data out of a string. Catch these exceptions as you would any other exception.
The CPAN distribution Exception::Class
makes creating and using exception objects easy:
package Zoo::Exceptions
{
use Exception::Class
'Zoo::AnimalEscaped',
'Zoo::HandlerEscaped';
}
sub cage_open
{
my $self = shift;
Zoo::AnimalEscaped->throw
unless $self->contains_animal;
...
}
sub breakroom_open
{
my $self = shift;
Zoo::HandlerEscaped->throw
unless $self->contains_handler;
...
}
Though throwing exceptions is relatively simple, catching them is less so. Using $@
correctly requires you to navigate several subtle risks:
local
ized uses further down the dynamic scope may modify $@
$@
may contain an object which returns a false value in boolean contextDIE
signal handler) may change $@
eval
and change $@
Perl 5.14 fixed some of these issues. They occur rarely, but they're often difficult to diagnose and to fix. The Try::Tiny
CPAN distribution improves the safety of exception handling and the syntax Try::Tiny
inspired improvements to Perl 5.14's exception handling..
Try::Tiny
is easy to use:
use Try::Tiny;
my $fh = try { open_log_file( 'monkeytown.log' ) }
catch { log_exception( $_ ) };
try
replaces eval
. The optional catch
block executes only when try
catches an exception. catch
receives the caught exception as the topic variable $_
.
Perl itself throws several exceptional conditions. perldoc perldiag
lists several "trappable fatal errors". Some are syntax errors that Perl produces during failed compilations, but you can catch the others during runtime. The most interesting are:
Of course you can also catch exceptions produced by autodie
(The autodie Pragma) and any lexical warnings promoted to exceptions (Registering Your Own Warnings).
Most Perl modules which provide new functions or define classes (Moose). Others, such as strict
or warnings
, influence the behavior of the language itself. This second class of module is a pragma. By convention, pragmas have lower-case names to differentiate them from other modules.
Pragmas work by exporting specific behavior or information into the lexical scopes of their callers. You've seen how declaring a lexical variable makes a symbol name available within a scope. Using a pragma makes its behavior effective within a scope as well:
{
# $lexical not visible; strict not in effect
{
use strict;
my $lexical = 'available here';
# $lexical is visible; strict is in effect
}
# $lexical again invisible; strict not in effect
}
Just as lexical declarations affect inner scopes, pragmas maintain their effects within inner scopes:
# file scope
use strict;
{
# inner scope, but strict still in effect
my $inner = 'another lexical';
}
use
a pragma as you would any other module. Pragmas take arguments, such as a minimum version number to use or a list of arguments to change the pragma's behavior:
# require variable declarations, prohibit barewords
use strict qw( subs vars );
# rely on the semantics of the 2012 book
use Modern::Perl '2012';
Sometimes you need to disable all or part of those effects within a further nested lexical scope. The no
builtin performs an unimport (Importing), which undoes the effects of a well-behaved pragma. For example, to disable the protection of strict
when you need to do something symbolic:
use Modern::Perl; # or use strict;
{
no strict 'refs';
# manipulate the symbol table here
}
Perl includes several useful core pragmas.
strict
pragma enables compiler checking of symbolic references, bareword use, and variable declaration.warnings
pragma enables optional warnings for deprecated, unintended, and awkward behaviors.utf8
pragma tells the parser to use the UTF-8 encoding to understand the source code of the current file.autodie
pragma enables automatic error checking of system calls and builtins.constant
pragma allows you to create compile-time constant values (though see the CPAN's Const::Fast
for an alternative).vars
pragma allows you to declare package global variables, such as $VERSION
or @ISA
(Blessed References).feature
pragma allows you to enable and disable newer features of Perl individually. Where use 5.14;
enables all of the Perl 5.14 features and the strict
pragma, use feature ':5.14';
does the same. This pragma is more useful to disable individual features in a lexical scope.less
pragma demonstrates how to write a pragma. As you might suspect from less
, you can write your own lexical pragmas in pure Perl. perldoc perlpragma
explains how to do so, while the explanation of $^H
in perldoc perlvar
explains how the feature works.
The CPAN has begun to gather non-core pragmas:
autovivification
disables autovivification (Autovivification)indirect
prevents the use of indirect invocation (Indirect Objects)autobox
enables object-like behavior for Perl's core types (scalars, references, arrays, and hashes).perl5i
combines and enables many experimental language extensions into a coherent whole. These tools are not widely used yet, but they have their champions. autovivification
and indirect
can help you write more correct code. autobox
and perl5i
are experiments with what Perl might one day become; they're worth playing with in small projects.