1. Live Fast,
Die Young,
Leave A Good Looking Corpse

2. Code Fast,
die() Early,
Throw Structured Exceptions

3. Throw Structured Exceptions
John SJ Anderson / @genehack / dc.pm
YAPC::NA 2012 ~ Madison WI

4. “Classic” Perl exceptions

5. “Classic” Perl exceptions
• Throw an exception with die()

6. “Classic” Perl exceptions
• Throw an exception with die()
• Or Carp::croak(), Carp::confess(), etc.

7. “Classic” Perl exceptions
• Throw an exception with die()
• Or Carp::croak(), Carp::confess(), etc.
• TIMTOWTDI!

8. “Classic” Perl exceptions
• Throw an exception with die()
• Or Carp::croak(), Carp::confess(), etc.
• TIMTOWTDI!
• Catch an exception with eval {}

9. “Classic” Perl exceptions
• Throw an exception with die()
• Or Carp::croak(), Carp::confess(), etc.
• TIMTOWTDI!
• Catch an exception with eval {}
• Handle an exception by looking at $@

10. “Classic” Perl exception throwing
1 #! /usr/bin/perl
2
3 use strict;
4 use warnings;
5
6 eval { my $result = this_might_fail() };
7
8 if( $@ ) {
9 # handle the error here
10 }

11. Problems with “classic” Perl exceptions

12. Problems with “classic” Perl exceptions
• $@ can get clobbered

13. Problems with “classic” Perl exceptions
• $@ can get clobbered
• $@ can get clobbered by code you don’t own

14. Problems with “classic” Perl exceptions
• $@ can get clobbered
• $@ can get clobbered by code you don’t own
• $@ might be a false value

15. Problems with “classic” Perl exceptions
• $@ can get clobbered
• $@ can get clobbered by code you don’t own
• $@ might be a false value
• If $@ is a string, you’re depending on duplicated
information, which will break.

16. Use Try::Tiny for
“semi-modern” Perl exceptions

17. Use Try::Tiny for
“semi-modern” Perl exceptions
• Provides try{}/catch{}/finally{} blocks

18. Use Try::Tiny for
“semi-modern” Perl exceptions
• Provides try{}/catch{}/finally{} blocks
• Handles details of properly dealing with complexities
around $@

19. Use Try::Tiny for
“semi-modern” Perl exceptions
• Provides try{}/catch{}/finally{} blocks
• Handles details of properly dealing with complexities
around $@
• Lightweight and generally Just Works(tm).

20. Use Try::Tiny for
“semi-modern” Perl exceptions
• Provides try{}/catch{}/finally{} blocks
• Handles details of properly dealing with complexities
around $@
• Lightweight and generally Just Works(tm).
• N.b.: you have to end try{}/catch{} with a
semicolon. Don’t forget this!

21. Use Try::Tiny for
“semi-modern” Perl exceptions
1 #! /usr/bin/perl
2
3 use strict;
4 use warnings;
5
6 use Try::Tiny;
7
8 try {
9 my $result = this_might_fail();
10 }
11 catch {
12 # handle the error here
13 };

22. Problems with
“semi-modern” Perl exceptions
• $@ can get clobbered
• $@ can get clobbered by code you don’t own
• $@ might be a false value
• If $@ is a string, you’re depending on duplicated
information, which will break.

23. Problems with
“semi-modern” Perl exceptions
• If $@ is a string, you’re depending on duplicated
information, which will break.

24. Wait, where’s the duplicated information?
1 my $answer;
2 try {
3 # any of these might throw an exception
4 my $this = this_might_fail();
5 my $that = something_else_might_fail();
6 $answer = combine_them( $this , $that );
7 }
8 catch {
9 # our error is in $_
10 if( $_ =~ /some error/ ) {
11 # handle some error
12 }
13 elsif( $_ =~ /another error/ ) {
14 # handle another error
15 }
16 else { # not sure what the problem is, just give up
17 confess( $_ );
18 }
19 };

25. Wait, where’s the duplicated information?
1 my $answer;
2 try {
3 # any of these might throw an exception
4 my $this = this_might_fail();
5 my $that = something_else_might_fail();
6 $answer = combine_them( $this , $that );
7 }
8 catch {
9 # our error is in $_
10 if( $_ =~ /some error/ ) {
11 # handle some error
12 }
13 elsif( $_ =~ /another error/ ) {
14 # handle another error
15 }
16 else { # not sure what the problem is, just give up
17 confess( $_ );
18 }
19 };

26. Wait, where’s the duplicated information?
• As soon as somebody “fixes” the string in some die()
somewhere, you’ve potentially broken exception
handling

27. Wait, where’s the duplicated information?
• As soon as somebody “fixes” the string in some die()
somewhere, you’ve potentially broken exception
handling
• And you can’t even easily tell where, because it’s
probably in a regexp that doesn’t look at all like the
changed string

28. Wait, where’s the duplicated information?
• As soon as somebody “fixes” the string in some die()
somewhere, you’ve potentially broken exception
handling
• And you can’t even easily tell where, because it’s
probably in a regexp that doesn’t look at all like the
changed string
• Even if you have tests for the code in question, do you
really have test coverage on all your exception cases?

29. So what’s the solution?

30. So what’s the solution?
• die() can also take a reference as an argument

31. So what’s the solution?
• die() can also take a reference as an argument
• So you can die() with an object!

32. So what’s the solution?
• die() can also take a reference as an argument
• So you can die() with an object!
• Which means you can cram all sorts of useful information into your
exceptions

33. So what’s the solution?
• die() can also take a reference as an argument
• So you can die() with an object!
• Which means you can cram all sorts of useful information into your
exceptions
• And more importantly, handle them in a structured fashion that’s much less
brittle than string comparisons

34. A framework for structured exceptions:
Exception::Class
use Exception::Class (
'MyException',
'AnotherException' => { isa => 'MyException' },
'YetAnotherException' => {
isa => 'AnotherException',
description => 'These exceptions are related to IPC'
},
'ExceptionWithFields' => {
isa => 'YetAnotherException',
fields => [ 'grandiosity', 'quixotic' ],
},
);

35. A framework for structured exceptions:
Exception::Class
# try
eval { MyException->throw( error => 'I feel funny.' ) };
my $e;
# catch
if ( $e = Exception::Class->caught('MyException') ) {
warn $e->error, "n", $e->trace->as_string, "n";
warn join ' ', $e->euid, $e->egid, $e->uid, $e->gid, $e->pid;
exit;
}
elsif ( $e = Exception::Class->caught('ExceptionWithFields') ) {
$e->quixotic ? do_something_wacky() : do_something_sane();
}
else {
$e = Exception::Class->caught();
ref $e ? $e->rethrow : die $e;
}

36. Exception::Class Pros

37. Exception::Class Pros
• Nice declarative syntax

38. Exception::Class Pros
• Nice declarative syntax
• Possible to declare detailed or simple exception class
hierarchies very simply

39. Exception::Class Pros
• Nice declarative syntax
• Possible to declare detailed or simple exception class
hierarchies very simply
• Supports aliasing to make throwing particular exception
types easier

40. Exception::Class Cons

41. Exception::Class Cons
• Not really designed for use with Try::Tiny

42. Exception::Class Cons
• Not really designed for use with Try::Tiny
• Based on Class::Data::Inheritable, not Moose

43. A Moose role for structured exceptions:
Throwable
package Redirect;
use Moose;
with 'Throwable';
has url => (is => 'ro');
...then later...
Redirect->throw({ url => $url });

44. Throwable Pros

45. Throwable Pros
• Implemented as a Moose role, so your exception
classes are just normal Moose classes that consume
the role

46. Throwable Pros
• Implemented as a Moose role, so your exception
classes are just normal Moose classes that consume
the role
• So you get the usual Moose-y good stuff around
attributes and methods and such.

47. Throwable Pros
• Implemented as a Moose role, so your exception
classes are just normal Moose classes that consume
the role
• So you get the usual Moose-y good stuff around
attributes and methods and such.
• Comes with a grab-bag of typical exception behaviors
(in Throwable::X), like stack traces, printf-ish
messages, etc.

48. Throwable Cons

49. Throwable Cons
?

50. Throwable Cons
?
So far, I haven’t really found any.

51. Throwable Cons
?
So far, I haven’t really found any.
(Of course, that doesn’t mean there aren’t any...)

52. Error Handling
Patterns & Anti-Patterns
• DO use exceptions instead of error flags
• DO throw exceptions as early as possible
• DON’T catch exceptions unless you’re going to handle
them – just let them propagate upwards
• DO design your web application-level error actions to
handle your business logic-level exceptions

53. Use exceptions instead of error flags
1 sub some_catalyst_action :Local {
2 my( $self , $c ) = @_;
3 my $session = get_huge_session_object( $c->session );
4
5 my $validated_params = validate_request_params( $c->request->params )
6 or $c->detach( 'error' );
7
8 my $step_one_result = $c->model('BusinessLogic')->do_step_one( $session , $validated_params );
9 $c->detach( 'error' ) if $session->has_error();
10
11 my $step_two_result = $c->model('BusinessLogic')->do_step_two( $step_one_result, $session );
12 $c->detach( 'error' ) if $session->has_error();
13 }

54. Use exceptions instead of error flags
Please please please don’t write code like this!

55. Use exceptions instead of error flags

56. Use exceptions instead of error flags
• Forget just one of those checks and you’ve got a hard to track
down bug

57. Use exceptions instead of error flags
• Forget just one of those checks and you’ve got a hard to track
down bug
• $session is being passed just to provide access to that error
flag.

58. Use exceptions instead of error flags
• Forget just one of those checks and you’ve got a hard to track
down bug
• $session is being passed just to provide access to that error
flag.
• The error action gets no real info about what the problem was,
or it tries to pull it from $session itself (which has its own
problems)

59. Use exceptions instead of error flags
1 sub some_catalyst_action :Local {
2 my( $self , $c ) = @_;
3
4 try {
5 my $validated_params = validate_request_params( $c->request->params )
6
7 my $step_one_result = $c->model('BusinessLogic')->do_step_one( $session , $validated_params );
8
9 my $step_two_result = $c->model('BusinessLogic')->do_step_two( $step_one_result, $session );
10 }
11 catch { $c->detach( 'error' , [ $_ ] ) };
12 }

60. Throw exceptions as early as
possible

61. Throw exceptions as early as
possible
• If you’re going to throw an exception because you
didn’t get passed something, do it ASAP.

62. Throw exceptions as early as
possible
• If you’re going to throw an exception because you
didn’t get passed something, do it ASAP.
• In general, the quicker you can die(), the better –
because it reduces the amount of code that might
contain the bug.

63. Don’t catch exceptions
except to handle them

64. Don’t catch exceptions
except to handle them
• Most of the time, your business logic code is going to
throw exceptions, not catch them

65. Don’t catch exceptions
except to handle them
• Most of the time, your business logic code is going to
throw exceptions, not catch them
• If you do catch an exception, you should be trying to fix
the problem.

66. Don’t catch exceptions
except to handle them
• Most of the time, your business logic code is going to
throw exceptions, not catch them
• If you do catch an exception, you should be trying to fix
the problem.
• Don’t catch exceptions just to munge them or log them
and re-throw them. Munge them or log them before
you throw them.

67. Further reading
• Throwable::X: common behavior for thrown exceptions
(<http://rjbs.manxome.org/rubric/entry/1860>)
• Exceptionally Extensible Exceptions
(<http://advent.rjbs.manxome.org/2010/2010-12-03.html>)
• Structured Data and Knowing versus Guessing
(<http://www.modernperlbooks.com/mt/2010/10/structured-data-and-knowing-versus-guessing.html>)

68. Thanks!
Questions?

69. Further reading
• Throwable::X: common behavior for thrown exceptions
(<http://rjbs.manxome.org/rubric/entry/1860>)
• Exceptionally Extensible Exceptions
(<http://advent.rjbs.manxome.org/2010/2010-12-03.html>)
• Structured Data and Knowing versus Guessing
(<http://www.modernperlbooks.com/mt/2010/10/structured-data-and-knowing-versus-guessing.html>)