#!perl -w
# run this document through perl to check its syntax
use Pod::Checker;
podchecker(\*DATA);
__END__

=head1 NAME

App::exceptions - Programming with Exceptions

=head1 INTRODUCTION

Programming with Exceptions
is a much more reliable way to prepare for unusual conditions
than trying to handle error codes returned from functions and methods
That is why programming with exceptions is an Enterprise Programming
topic.  However, Perl has not always supported programming with
exceptions, and many Perl programmers are not familiar with this
style of programming using Perl.

This document was written to explain the support within the Perl
language for programming using exceptions in general.  It also explains
the standards for exception programming in the App-Context framework.

The short answer is that we recommend the following modules
to assist in exception programming.

  Carp
  CGI::Carp
  Fatal
  Exception::Class
  Devel::Stacktrace

However, it is important to understand what each does for you,
why, and how they all fit together.
So for the longer answer, read on.

=head1 IMPORTANCE OF EXCEPTION PROGRAMMING

Exception Programming is a technique to deal with exceptional 
conditions (i.e. errors) encountered during runtime. 
It is contrasted with the more common technique of 
Error Checking.

Error Checking is a technique where an operation is performed.
Then all possible error conditions are checked and handled.
This requires the developer to address the following challenges.

=over

=item 1. Error Completeness

The developer must consider all possible error conditions
(difficult to foresee and consider at every level).

=item 2. Error Propagation

The developer must add quite a bit of additional logic
to allow subsequent statements after the error condition
to be skipped and the error to be propagated upward in
the call stack to the place where
it can be acted on appropriately.

=back

The problems with Error Checking is that it is so difficult
to do and so rarely done completely. 
Since the program runs fine when no errors are
encountered, the developer invariably fails to put in all of
the error checks and the accompanying error propagation logic.

Exception Programming addresses these problems using the
following methods.

=over

=item 1. Error Completeness

Exceptions are grouped into an exception hierarchy.
Therefore, high level code needs only make the distinctions
between exceptions that it thinks are relevant.  If additional
exceptions are added later by low level code, they will be
handled in accordance with the upper levels of the exception
hierarchy which are already known.

=item 2. Error Propagation

Exceptions take advantage of built-in language support
(longjmp() in C) to unravel the call stack.  Additional
logic is not required by the developer to propagate the
error to a place that is prepared to handle it.

=back

Because Exception Programming addresses the issue of 
software reliability (one of the Attributes of Enterprise
Systems), it is necessarily an Enterprise Programming issue.

=head1 PERL 5 LANGUAGE SUPPORT

Perl did not always support exception programming.
Even in Perl 5.6.1, the support for exception programming
is somewhat scattered.
The following are features of Perl 5 which are relevant
to exception programming.

=head2 die

  Camel Book: Chapter 3: Functions, "die"
  Online Doc: man perlfunc (or "perldoc perlfunc")

The "die" function prints its arguments to STDERR and causes
the program to exit.  It has some extra features, like appending
__FILE__ and __LINE__ strings if the args don't end with a
newline.  The biggest feature is that it may be "caught"
if it is executed within an eval statement.

Simply put, "die" is the native Perl way of "throwing" an
exception.  The only "attributes" of the "exception" is the
message (concatenated args) of the "die" itself.

  if ($error_condition_xyz_exists) {
      die "Error XYZ occurred";
  }

The "warn" function is similar to "die", in that it prints out
its arguments to STDERR in the same way, but it does not try
to exit or throw an exception in any way.  Therefore, it's not
relevant for this discussion.

  die  "I'm dying";    # print msg to STDERR and exit

=head2 croak and confess

  Camel Book: Chapter 7: The Standard Perl Library, "Carp"
  Online Doc: man Carp (or "perldoc Carp")

The Carp.pm module is part of the Standard Perl Library
It allows you to throw exceptions (like "die"), but it reports
the line number and file name from the perspective of the 
caller of the function/method that died.

This is useful for library code, so that the programmer sees
where he invoked the function that failed rather than seeing
where within that function the failure occurred.

  use Carp;

  croak "We're outta here!";     # like "die", but line # from caller
  confess "It was my fault: $!"; # like "croak" but with stack trace

Using "croak" to throw an exception references the line # and
file name of the caller.

Using "confess" to throw an exception causes a 
stack trace to be printed also.

=head2 CGI::Carp

  Online Doc: man CGI::Carp (or "perldoc CGI::Carp")

If you are developing perl scripts which will be run
by a web server, the STDERR stream is usually redirected
the web server error log.  However, the output generated
by "die" is not formatted nicely, with a datetime stamp
and filename, as is customary with error log entries.

The CGI::Carp package replaces die, croak, and confess
with versions which work the same but format the output
a little more nicely for the web server error log.
You can even direct fatal errors to the browser.

  use CGI::Carp;

  die  "I'm dying";              # print msg to log and exit
  warn "I'm confused";           # print msg to log
  croak "We're outta here!";     # like "die", but line # from caller
  confess "It was my fault: $!"; # like "croak" but with stack trace
  carp "It was your fault!";     # like "warn", but line # from caller

  use CGI::Carp qw(fatalsToBrowser);
  die "Fatal error messages are now sent to browser";

=head2 $!

  Camel Book: Chapter 2: The Gory Details/Special Variables, $!
  Online Doc: man perlvar (or "perldoc perlvar")

If used in a string context, yields the error string
from the last system call error in the currently executing perl 
interpreter.

You shouldn't depend on $! being 
anything in particular unless you've gotten a specific error
indicating a system error.

So when you get a system call error returned from a perl function
(system calls usually relate to files, networking,
processes, or interprocess communication), you can check $!.

If you want to use good exception programming techniques, every
perl function which can fail from a system call error should
be checked for its return value so that an exception may be
thrown.

  $file = "foo.txt";
  open(FILE, "< $file") || die "Error opening [$file]: $!";

Of course, if your program's requirements allow you to
handle the error locally without
throwing an exception, you are welcome to.
However, every possible error needs to be thought about
and accounted for.

  $file = "foo.txt";
  if (open(FILE, "< $file")) {  # if file doesn't exist, that's ok
      @data = <FILE>;
      close(FILE);
  }

=head2 eval

  Camel Book: Chapter 3: Functions, "eval"
  Online Doc: man perlfunc (or "perldoc perlfunc")

Eval serves as Perl's version of a "try {}" block.

If a "die" (an exception) occurs within an "eval", the program
is not terminated, only the "eval".  All of the code
that was eval'ed, after the exception, is skipped.

Eval has two syntaxes, "eval EXPR" and "eval BLOCK".

The "eval EXPR" syntax causes the code contained in the 
EXPR expression to be compiled every time the 
statement is executed during runtime.

  eval "&do_big_function();";
  $code = "&do_another_big_operation();";
  eval $code;

The "eval BLOCK" syntax causes the code contained in the
BLOCK to be compiled only once (at script compile time),
so it is much more efficient and appropriate for 
exception programming.
(Please note the semi-colon that must follow the BLOCK.)

  eval {
      &do_big_function();
  };

If any exception ("die") is thrown within the do_big_function(),
the program will not terminate.  Control will simply be returned
to the end of the eval block.

=head2 $@

  Camel Book: Chapter 2: The Gory Details/Special Variables, $@
  Online Doc: man perlvar (or "perldoc perlvar")

The is the Perl syntax error message from the last "eval" command.
Alternatively, it is the message of an exception that was thrown
(using "die").
If null, the last eval was parsed and executed correctly, with
no exceptions.

Thus, perl exception programming can be done with the following.

  eval {
      &do_big_function();
  };
  if ($@) {
      # handle the exception ($@ is the exception message)
  }

  sub do_big_function {
      ...
      if ($error_condition_xyz_exists) {
          die "Error XYZ occurred";
      }
      ...
  }

=head2 $SIG{__DIE__}

  Camel Book: Chapter 2: The Gory Details/Special Variables, %SIG
  Online Doc: man perlvar (or "perldoc perlvar")

This handler is not necessary to know about to do exception 
programming in Perl.  However, it may be used by modules that assist
in exception handling tasks.

The %SIG hash contains references to subroutines which are called
in response to signals received by the process.  "__DIE__" is a
special internal hook which is not really an external signal.

So if you desired to do some additional processing between the
time that the "die" exception was thrown and the time it was
handled, you could replace the $SIG{__DIE__} handler.

  $SIG{__DIE__} = sub {
      # do something (@_ are the args of the "die")
  };

=head2 caller

  Camel Book: Chapter 3: Functions, "caller"
  Online Doc: man perlfunc (or "perldoc perlfunc")

The "caller" function allows you to examine the call stack
in order (for example) to print stack traces.

=head2 Weaknesses in the Native Perl Exception Support

The eval/die pair provides the critical language support
necessary for Error Propagation. 

However, the single
attribute of an exception, the message, leaves the
developer without an exception hierarchy which would
allow him to ensure Error Completeness in his
exception handling.

Furthermore, the Perl functions themselves must be
checked for errors rather than throwing exceptions.

=head1 EXCEPTION PROGRAMMING USING Fatal.pm

Causes all of the perl functions to throw exceptions
rather than simply return error codes or set error
flags.

=head1 EXCEPTION PROGRAMMING USING Error.pm

Introduces the ability to have an Exception Class
(not just a message) and to use "try/catch"
syntax.

It does this be defining try() and catch() subroutines
with appropriate prototypes, using the "{}" blocks
as closures (or anonymous subroutines).  See the following
references.

  Camel Book: Chapter 2: The Gory Details, Subroutines, Prototypes
  Camel Book: Chapter 4: References and Nested Data Structures, Using Hard References, Closures

However, you must be very careful in the placement of
your semi-colons using the try/catch syntax provided by
Error.pm or your
module will leak memory (very bad in a mod_perl context).

=head1 EXCEPTION PROGRAMMING USING Exception::Class

Introduces the capability to collect exception
classes easily in a class hierarchy.  This helps
address the issue of Error Completeness.

=head1 EXCEPTION PROGRAMMING USING Devel::Stacktrace

Adds the capability to print out more detailed
stack trace information than was possible with
"confess".

=head1 FURTHER RESEARCH

An excellent section on Exception Handling in Perl
exists in the mod_perl Guide.

  http://perl.apache.org/guide/perl.html#Exception_Handling_for_mod_perl

You can read up on future directions
for Perl and the evolving exception support here.

  http://www.perl.com/pub/a/2002/01/15/apo4.html?page=3
  http://dev.perl.org/rfc/88.html
  http://nntp.perl.org/group/perl.perl6.language.errors

=cut