3 # ABSTRACT: minimal try/catch with proper preservation of $@
8 use Exporter 5.57 'import';
9 our @EXPORT = our @EXPORT_OK = qw(try catch finally);
12 $Carp::Internal{+__PACKAGE__}++;
14 BEGIN { eval "use Sub::Name; 1" or *{subname} = sub {1} }
16 # Need to prototype as @ not $$ because of the way Perl evaluates the prototype.
17 # Keeping it at $$ means you only ever get 1 sub because we need to eval in a list
18 # context & not a scalar one
21 my ( $try, @code_refs ) = @_;
23 # we need to save this here, the eval block will be in scalar context due
25 my $wantarray = wantarray;
27 # work around perl bug by explicitly initializing these, due to the likelyhood
28 # this will be used in global destruction (perl rt#119311)
29 my ( $catch, @finally ) = ();
31 # find labeled blocks in the argument list.
32 # catch and finally tag the blocks by blessing a scalar reference to them.
33 foreach my $code_ref (@code_refs) {
35 if ( ref($code_ref) eq 'Try::Tiny::Catch' ) {
36 croak 'A try() may not be followed by multiple catch() blocks'
38 $catch = ${$code_ref};
39 } elsif ( ref($code_ref) eq 'Try::Tiny::Finally' ) {
40 push @finally, ${$code_ref};
43 'try() encountered an unexpected argument ('
44 . ( defined $code_ref ? $code_ref : 'undef' )
45 . ') - perhaps a missing semi-colon before or'
50 # FIXME consider using local $SIG{__DIE__} to accumulate all errors. It's
51 # not perfect, but we could provide a list of additional errors for
54 # name the blocks if we have Sub::Name installed
56 subname("${caller}::try {...} " => $try);
57 subname("${caller}::catch {...} " => $catch) if $catch;
58 subname("${caller}::finally {...} " => $_) foreach @finally;
60 # save the value of $@ so we can set $@ back to it in the beginning of the eval
61 # and restore $@ after the eval finishes
66 # failed will be true if the eval dies, because 1 will not be returned
68 my $failed = not eval {
71 # evaluate the try block in the correct context
74 } elsif ( defined $wantarray ) {
80 return 1; # properly set $failed to false
83 # preserve the current error and reset the original value of $@
87 # set up a scope guard to invoke the finally block at the end
89 map { Try::Tiny::ScopeGuard->_new($_, $failed ? $error : ()) }
92 # at this point $failed contains a true value if the eval died, even if some
93 # destructor overwrote $@ as the eval was unwinding.
95 # if we got an error, invoke the catch block.
97 # This works like given($error), but is backwards compatible and
98 # sets $_ in the dynamic scope for the body of C<$catch>
100 return $catch->($error);
103 # in case when() was used without an explicit return, the C<for>
104 # loop will be aborted and there's no useful return value
109 # no failure, $@ is back to what it was, everything is fine
110 return $wantarray ? @ret : $ret[0];
115 my ( $block, @rest ) = @_;
117 croak 'Useless bare catch()' unless wantarray;
120 bless(\$block, 'Try::Tiny::Catch'),
126 my ( $block, @rest ) = @_;
128 croak 'Useless bare finally()' unless wantarray;
131 bless(\$block, 'Try::Tiny::Finally'),
137 package # hide from PAUSE
138 Try::Tiny::ScopeGuard;
140 use constant UNSTABLE_DOLLARAT => ($] < '5.013002') ? 1 : 0;
148 my ($code, @args) = @{ $_[0] };
150 local $@ if UNSTABLE_DOLLARAT;
156 "Execution of finally() block $code resulted in an exception, which "
157 . '*CAN NOT BE PROPAGATED* due to fundamental limitations of Perl. '
158 . 'Your program will continue as if this event never took place. '
159 . "Original exception text follows:\n\n"
160 . (defined $@ ? $@ : '$@ left undefined...')
175 You can use Try::Tiny's C<try> and C<catch> to expect and handle exceptional
176 conditions, avoiding quirks in Perl and common mistakes:
178 # handle errors with a catch handler
182 warn "caught error: $_"; # not $@
185 You can also use it like a standalone C<eval> to catch and ignore any error
186 conditions. Obviously, this is an extreme measure not to be undertaken
189 # just silence errors
196 This module provides bare bones C<try>/C<catch>/C<finally> statements that are designed to
197 minimize common mistakes with eval blocks, and NOTHING else.
199 This is unlike L<TryCatch> which provides a nice syntax and avoids adding
200 another call stack layer, and supports calling C<return> from the C<try> block to
201 return from the parent subroutine. These extra features come at a cost of a few
202 dependencies, namely L<Devel::Declare> and L<Scope::Upper> which are
203 occasionally problematic, and the additional catch filtering uses L<Moose>
204 type constraints which may not be desirable either.
206 The main focus of this module is to provide simple and reliable error handling
207 for those having a hard time installing L<TryCatch>, but who still want to
208 write correct C<eval> blocks without 5 lines of boilerplate each time.
210 It's designed to work as correctly as possible in light of the various
211 pathological edge cases (see L</BACKGROUND>) and to be compatible with any style
212 of error values (simple strings, references, objects, overloaded objects, etc).
214 If the C<try> block dies, it returns the value of the last statement executed in
215 the C<catch> block, if there is one. Otherwise, it returns C<undef> in scalar
216 context or the empty list in list context. The following examples all
217 assign C<"bar"> to C<$x>:
219 my $x = try { die "foo" } catch { "bar" };
220 my $x = try { die "foo" } || "bar";
221 my $x = (try { die "foo" }) // "bar";
223 my $x = eval { die "foo" } || "bar";
225 You can add C<finally> blocks, yielding the following:
228 try { die 'foo' } finally { $x = 'bar' };
229 try { die 'foo' } catch { warn "Got a die: $_" } finally { $x = 'bar' };
231 C<finally> blocks are always executed making them suitable for cleanup code
232 which cannot be handled using local. You can add as many C<finally> blocks to a
233 given C<try> block as you like.
235 Note that adding a C<finally> block without a preceding C<catch> block
236 suppresses any errors. This behaviour is consistent with using a standalone
237 C<eval>, but it is not consistent with C<try>/C<finally> patterns found in
238 other programming languages, such as Java, Python, Javascript or C#. If you
239 learnt the C<try>/C<finally> pattern from one of these languages, watch out for
244 All functions are exported by default using L<Exporter>.
246 If you need to rename the C<try>, C<catch> or C<finally> keyword consider using
247 L<Sub::Import> to get L<Sub::Exporter>'s flexibility.
253 Takes one mandatory C<try> subroutine, an optional C<catch> subroutine and C<finally>
256 The mandatory subroutine is evaluated in the context of an C<eval> block.
258 If no error occurred the value from the first block is returned, preserving
261 If there was an error and the second subroutine was given it will be invoked
262 with the error in C<$_> (localized) and as that block's first and only
265 C<$@> does B<not> contain the error. Inside the C<catch> block it has the same
266 value it had before the C<try> block was executed.
268 Note that the error may be false, but if that happens the C<catch> block will
271 Once all execution is finished then the C<finally> block, if given, will execute.
275 Intended to be used in the second argument position of C<try>.
277 Returns a reference to the subroutine it was given but blessed as
278 C<Try::Tiny::Catch> which allows try to decode correctly what to do
279 with this code reference.
283 Inside the C<catch> block the caught error is stored in C<$_>, while previous
284 value of C<$@> is still available for use. This value may or may not be
285 meaningful depending on what happened before the C<try>, but it might be a good
286 idea to preserve it in an error stack.
288 For code that captures C<$@> when throwing new errors (i.e.
289 L<Class::Throwable>), you'll need to do:
310 Intended to be the second or third element of C<try>. C<finally> blocks are always
311 executed in the event of a successful C<try> or if C<catch> is run. This allows
312 you to locate cleanup code which cannot be done via C<local()> e.g. closing a file
315 When invoked, the C<finally> block is passed the error that was caught. If no
316 error was caught, it is passed nothing. (Note that the C<finally> block does not
317 localize C<$_> with the error, since unlike in a C<catch> block, there is no way
318 to know if C<$_ == undef> implies that there were no errors.) In other words,
319 the following code does just what you would expect:
324 # ...code run in case of error
327 print "The try block died with: @_\n";
329 print "The try block ran without error.\n";
333 B<You must always do your own error handling in the C<finally> block>. C<Try::Tiny> will
334 not do anything about handling possible errors coming from code located in these
337 Furthermore B<exceptions in C<finally> blocks are not trappable and are unable
338 to influence the execution of your program>. This is due to limitation of
339 C<DESTROY>-based scope guards, which C<finally> is implemented on top of. This
340 may change in a future version of Try::Tiny.
342 In the same way C<catch()> blesses the code reference this subroutine does the same
343 except it bless them as C<Try::Tiny::Finally>.
349 There are a number of issues with C<eval>.
353 When you run an C<eval> block and it succeeds, C<$@> will be cleared, potentially
354 clobbering an error that is currently being caught.
356 This causes action at a distance, clearing previous errors your caller may have
359 C<$@> must be properly localized before invoking C<eval> in order to avoid this
362 More specifically, C<$@> is clobbered at the beginning of the C<eval>, which
363 also makes it impossible to capture the previous error before you die (for
364 instance when making exception objects with error stacks).
366 For this reason C<try> will actually set C<$@> to its previous value (the one
367 available before entering the C<try> block) in the beginning of the C<eval>
370 =head2 Localizing $@ silently masks errors
372 Inside an C<eval> block, C<die> behaves sort of like:
376 return_undef_from_eval();
379 This means that if you were polite and localized C<$@> you can't die in that
380 scope, or your error will be discarded (printing "Something's wrong" instead).
382 The workaround is very ugly:
393 =head2 $@ might not be a true value
401 because due to the previous caveats it may have been unset.
403 C<$@> could also be an overloaded error object that evaluates to false, but
404 that's asking for trouble anyway.
406 The classic failure mode is:
408 sub Object::DESTROY {
413 my $obj = Object->new;
422 In this case since C<Object::DESTROY> is not localizing C<$@> but still uses
423 C<eval>, it will set C<$@> to C<"">.
425 The destructor is called when the stack is unwound, after C<die> sets C<$@> to
426 C<"foo at Foo.pm line 42\n">, so by the time C<if ( $@ )> is evaluated it has
427 been cleared by C<eval> in the destructor.
429 The workaround for this is even uglier than the previous ones. Even though we
430 can't save the value of C<$@> from code that doesn't localize, we can at least
431 be sure the C<eval> was aborted due to an error:
433 my $failed = not eval {
439 This is because an C<eval> that caught a C<die> will always return a false
444 Using Perl 5.10 you can use L<perlsyn/"Switch statements">.
446 The C<catch> block is invoked in a topicalizer context (like a C<given> block),
447 but note that you can't return a useful value from C<catch> using the C<when>
448 blocks without an explicit C<return>.
450 This is somewhat similar to Perl 6's C<CATCH> blocks. You can use it to
451 concisely match errors:
456 when (/^Can't locate .*?\.pm in \@INC/) { } # ignore
466 C<@_> is not available within the C<try> block, so you need to copy your
467 arglist. In case you want to work with argument values directly via C<@_>
468 aliasing (i.e. allow C<$_[1] = "foo">), you need to pass C<@_> by reference:
471 my ( $self, @args ) = @_;
472 try { $self->bar(@args) }
480 try { $_ = $self->bar($_) for @$args }
485 C<return> returns from the C<try> block, not from the parent sub (note that
486 this is also how C<eval> works, but not how L<TryCatch> works):
496 say "this text WILL be displayed, even though an exception is thrown";
499 Instead, you should capture the return value:
506 return unless $success;
508 say "This text WILL NEVER appear!";
511 sub parent_sub_with_catch {
517 # do something with $_
518 return undef; #see note
520 return unless $success;
522 say "This text WILL NEVER appear!";
525 Note that if you have a C<catch> block, it must return C<undef> for this to work,
526 since if a C<catch> block exists, its return value is returned in place of C<undef>
527 when an exception is thrown.
531 C<try> introduces another caller stack frame. L<Sub::Uplevel> is not used. L<Carp>
532 will not report this when using full stack traces, though, because
533 C<%Carp::Internal> is used. This lack of magic is considered a feature.
537 The value of C<$_> in the C<catch> block is not guaranteed to be the value of
538 the exception thrown (C<$@>) in the C<try> block. There is no safe way to
539 ensure this, since C<eval> may be used unhygenically in destructors. The only
540 guarantee is that the C<catch> will be called if an exception is thrown.
544 The return value of the C<catch> block is not ignored, so if testing the result
545 of the expression for truth on success, be sure to return a false value from
553 return; # avoid returning a true value;
560 C<$SIG{__DIE__}> is still in effect.
562 Though it can be argued that C<$SIG{__DIE__}> should be disabled inside of
563 C<eval> blocks, since it isn't people have grown to rely on it. Therefore in
564 the interests of compatibility, C<try> does not disable C<$SIG{__DIE__}> for
565 the scope of the error throwing code.
569 Lexical C<$_> may override the one set by C<catch>.
571 For example Perl 5.10's C<given> form uses a lexical C<$_>, creating some
579 warn $_; # will print $foo, not the error
580 warn $_[0]; # instead, get the error like this
585 Note that this behavior was changed once again in L<Perl5 version 18
586 |https://metacpan.org/module/perldelta#given-now-aliases-the-global-_>.
587 However, since the entirety of lexical C<$_> is now L<considired experimental
588 |https://metacpan.org/module/perldelta#Lexical-_-is-now-experimental>, it
589 is unclear whether the new version 18 behavior is final.
599 Much more feature complete, more convenient semantics, but at the cost of
600 implementation complexity.
604 Automatic error throwing for builtin functions and more. Also designed to
605 work well with C<given>/C<when>.
609 A lightweight role for rolling your own exception classes.
613 Exception object implementation with a C<try> statement. Does not localize
616 =item L<Exception::Class::TryCatch>
618 Provides a C<catch> statement, but properly calling C<eval> is your
621 The C<try> keyword pushes C<$@> onto an error stack, avoiding some of the
622 issues with C<$@>, but you still need to localize to prevent clobbering.
626 =head1 LIGHTNING TALK
628 I gave a lightning talk about this module, you can see the slides (Firefox
631 L<http://web.archive.org/web/20100628040134/http://nothingmuch.woobling.org/talks/takahashi.xul>
635 L<http://web.archive.org/web/20100305133605/http://nothingmuch.woobling.org/talks/yapc_asia_2009/try_tiny.yml>
637 =head1 VERSION CONTROL
639 L<http://github.com/doy/try-tiny/>