NAME Scope::Upper - Act on upper scopes. VERSION Version 0.16 SYNOPSIS "reap", "localize", "localize_elem", "localize_delete" and "WORDS" : package Scope; use Scope::Upper qw; sub new { my ($class, $name) = @_; localize '$tag' => bless({ name => $name }, $class) => UP; reap { print Scope->tag->name, ": end\n" } UP; } # Get the tag stored in the caller namespace sub tag { my $l = 0; my $pkg = __PACKAGE__; $pkg = caller $l++ while $pkg eq __PACKAGE__; no strict 'refs'; ${$pkg . '::tag'}; } sub name { shift->{name} } # Locally capture warnings and reprint them with the name prefixed sub catch { localize_elem '%SIG', '__WARN__' => sub { print Scope->tag->name, ': ', @_; } => UP; } # Locally clear @INC sub private { for (reverse 0 .. $#INC) { # First UP is the for loop, second is the sub boundary localize_delete '@INC', $_ => UP UP; } } ... package UserLand; { Scope->new("top"); # initializes $UserLand::tag { Scope->catch; my $one = 1 + undef; # prints "top: Use of uninitialized value..." { Scope->private; eval { require Cwd }; print $@; # prints "Can't locate Cwd.pm in @INC (@INC contains:) at..." } require Cwd; # loads Cwd.pm } } # prints "top: done" "unwind" and "want_at" : package Try; use Scope::Upper qw; sub try (&) { my @result = shift->(); my $cx = SUB UP; # Point to the sub above this one unwind +(want_at($cx) ? @result : scalar @result) => $cx; } ... sub zap { try { my @things = qw; return @things; # returns to try() and then outside zap() # not reached }; # not reached } my @stuff = zap(); # @stuff contains qw my $stuff = zap(); # $stuff contains 3 "uplevel" : package Uplevel; use Scope::Upper qw; sub target { faker(@_); } sub faker { uplevel { my $sub = (caller 0)[3]; print "$_[0] from $sub()"; } @_ => CALLER(1); } target('hello'); # "hello from Uplevel::target()" DESCRIPTION This module lets you defer actions *at run-time* that will take place when the control flow returns into an upper scope. Currently, you can: * hook an upper scope end with "reap" ; * localize variables, array/hash values or deletions of elements in higher contexts with respectively "localize", "localize_elem" and "localize_delete" ; * return values immediately to an upper level with "unwind", and know which context was in use then with "want_at" ; * execute a subroutine in the context of an upper subroutine stack frame with "uplevel". FUNCTIONS In all those functions, $context refers to the target scope. You have to use one or a combination of "WORDS" to build the $context passed to these functions. This is needed in order to ensure that the module still works when your program is ran in the debugger. The only thing you can assume is that it is an *absolute* indicator of the frame, which means that you can safely store it at some point and use it when needed, and it will still denote the original scope. "reap $callback, $context" Adds a destructor that calls $callback (in void context) when the upper scope represented by $context ends. "localize $what, $value, $context" Introduces a "local" delayed to the time of first return into the upper scope denoted by $context. $what can be : * A glob, in which case $value can either be a glob or a reference. "localize" follows then the same syntax as "local *x = $value". For example, if $value is a scalar reference, then the "SCALAR" slot of the glob will be set to $$value - just like "local *x = \1" sets $x to 1. * A string beginning with a sigil, representing the symbol to localize and to assign to. If the sigil is '$', "localize" follows the same syntax as "local $x = $value", i.e. $value isn't dereferenced. For example, localize '$x', \'foo' => HERE; will set $x to a reference to the string 'foo'. Other sigils ('@', '%', '&' and '*') require $value to be a reference of the corresponding type. When the symbol is given by a string, it is resolved when the actual localization takes place and not when "localize" is called. Thus, if the symbol name is not qualified, it will refer to the variable in the package where the localization actually takes place and not in the one where the "localize" call was compiled. For example, { package Scope; sub new { localize '$tag', $_[0] => UP } } { package Tool; { Scope->new; ... } } will localize $Tool::tag and not $Scope::tag. If you want the other behaviour, you just have to specify $what as a glob or a qualified name. Note that if $what is a string denoting a variable that wasn't declared beforehand, the relevant slot will be vivified as needed and won't be deleted from the glob when the localization ends. This situation never arises with "local" because it only compiles when the localized variable is already declared. Although I believe it shouldn't be a problem as glob slots definedness is pretty much an implementation detail, this behaviour may change in the future if proved harmful. "localize_elem $what, $key, $value, $context" Introduces a "local $what[$key] = $value" or "local $what{$key} = $value" delayed to the time of first return into the upper scope denoted by $context. Unlike "localize", $what must be a string and the type of localization is inferred from its sigil. The two only valid types are array and hash ; for anything besides those, "localize_elem" will throw an exception. $key is either an array index or a hash key, depending of which kind of variable you localize. If $what is a string pointing to an undeclared variable, the variable will be vivified as soon as the localization occurs and emptied when it ends, although it will still exist in its glob. "localize_delete $what, $key, $context" Introduces the deletion of a variable or an array/hash element delayed to the time of first return into the upper scope denoted by $context. $what can be: * A glob, in which case $key is ignored and the call is equivalent to "local *x". * A string beginning with '@' or '%', for which the call is equivalent to respectiveley "local $a[$key]; delete $a[$key]" and "local $h{$key}; delete $h{$key}". * A string beginning with '&', which more or less does "undef &func" in the upper scope. It's actually more powerful, as &func won't even "exists" anymore. $key is ignored. "unwind @values, $context" Returns @values *from* the context pointed by $context, i.e. from the subroutine, eval or format at or just above $context, and immediately restart the program flow at this point - thus effectively returning to an upper scope. The upper context isn't coerced onto @values, which is hence always evaluated in list context. This means that my $num = sub { my @a = ('a' .. 'z'); unwind @a => HERE; # not reached }->(); will set $num to 'z'. You can use "want_at" to handle these cases. "want_at $context" Like "wantarray", but for the subroutine/eval/format at or just above $context. The previous example can then be "corrected" : my $num = sub { my @a = ('a' .. 'z'); unwind +(want_at(HERE) ? @a : scalar @a) => HERE; # not reached }->(); will rightfully set $num to 26. "uplevel $code, @args, $context" Executes the code reference $code with arguments @args as if it were located at the subroutine stack frame pointed by $context, effectively fooling "caller" and "die" into believing that the call actually happened higher in the stack. The code is executed in the context of the "uplevel" call, and what it returns is returned as-is by "uplevel". sub target { faker(@_); } sub faker { uplevel { map { 1 / $_ } @_; } @_ => CALLER(1); } my @inverses = target(1, 2, 4); # @inverses contains (0, 0.5, 0.25) my $count = target(1, 2, 4); # $target is 3 Sub::Uplevel also implements a pure-Perl version of "uplevel". Both are identical, with the following caveats : * The Sub::Uplevel implementation of "uplevel" may execute a code reference in the context of any upper stack frame. The Scope::Upper version only allows to uplevel to a subroutine stack frame, and will croak if you try to target an "eval" or a format. * Exceptions thrown from the code called by this version of "uplevel" will not be caught by "eval" blocks between the target frame and the uplevel call, while they will for Sub::Uplevel's version. This means that : eval { sub { local $@; eval { sub { uplevel { die 'wut' } CALLER(2); # for Scope::Upper # uplevel(3, sub { die 'wut' }) # for Sub::Uplevel }->(); }; print "inner block: $@"; $@ and exit; }->(); }; print "outer block: $@"; will print "inner block: wut..." with Sub::Uplevel and "outer block: wut..." with Scope::Upper. * Sub::Uplevel globally overrides "CORE::GLOBAL::caller", while Scope::Upper does not. A simple wrapper lets you mimic the interface of "uplevel" in Sub::Uplevel : use Scope::Upper; sub uplevel { my $frame = shift; my $code = shift; my $cxt = Scope::Upper::CALLER($frame); &Scope::Upper::uplevel($code => @_ => $cxt); } Albeit the three exceptions listed above, it passes all the tests of Sub::Uplevel. CONSTANTS "SU_THREADSAFE" True iff the module could have been built when thread-safety features. WORDS Constants "TOP" Returns the context that currently represents the highest scope. "HERE" The context of the current scope. Getting a context from a context For any of those functions, $from is expected to be a context. When omitted, it defaults to the the current context. "UP $from" The context of the scope just above $from. "SUB $from" The context of the closest subroutine above $from. Note that $from is returned if it is already a subroutine context ; hence "SUB SUB == SUB". "EVAL $from" The context of the closest eval above $from. Note that $from is returned if it is already an eval context ; hence "EVAL EVAL == EVAL". Getting a context from a level Here, $level should denote a number of scopes above the current one. When omitted, it defaults to 0 and those functions return the same context as "HERE". "SCOPE $level" The $level-th upper context, regardless of its type. "CALLER $level" The context of the $level-th upper subroutine/eval/format. It kind of corresponds to the context represented by "caller $level", but while e.g. "caller 0" refers to the caller context, "CALLER 0" will refer to the top scope in the current context. Examples Where "reap" fires depending on the $cxt : sub { eval { sub { { reap \&cleanup => $cxt; ... } # $cxt = SCOPE(0), or HERE ... }->(); # $cxt = SCOPE(1), or UP, or SUB, or CALLER, or CALLER(0) ... }; # $cxt = SCOPE(2), or UP UP, or UP SUB, or EVAL, or CALLER(1) ... }->(); # $cxt = SCOPE(3), or SUB UP SUB, or SUB EVAL, or CALLER(2) ... Where "localize", "localize_elem" and "localize_delete" act depending on the $cxt : sub { eval { sub { { localize '$x' => 1 => $cxt; # $cxt = SCOPE(0), or HERE ... } # $cxt = SCOPE(1), or UP, or SUB, or CALLER, or CALLER(0) ... }->(); # $cxt = SCOPE(2), or UP UP, or UP SUB, or EVAL, or CALLER(1) ... }; # $cxt = SCOPE(3), or SUB UP SUB, or SUB EVAL, or CALLER(2) ... }->(); # $cxt = SCOPE(4), UP SUB UP SUB, or UP SUB EVAL, or UP CALLER(2), or TOP ... Where "unwind", "want_at" and "uplevel" point to depending on the $cxt: sub { eval { sub { { unwind @things => $cxt; # or uplevel { ... } $cxt; ... } ... }->(); # $cxt = SCOPE(0 .. 1), or HERE, or UP, or SUB, or CALLER(0) ... }; # $cxt = SCOPE(2), or UP UP, or UP SUB, or EVAL, or CALLER(1) (*) ... }->(); # $cxt = SCOPE(3), or SUB UP SUB, or SUB EVAL, or CALLER(2) ... # (*) Note that uplevel() will croak if you pass that scope frame, # because it can't target eval scopes. EXPORT The functions "reap", "localize", "localize_elem", "localize_delete", "unwind", "want_at" and "uplevel" are only exported on request, either individually or by the tags ':funcs' and ':all'. The constant "SU_THREADSAFE" is also only exported on request, individually or by the tags ':consts' and ':all'. Same goes for the words "TOP", "HERE", "UP", "SUB", "EVAL", "SCOPE" and "CALLER" that are only exported on request, individually or by the tags ':words' and ':all'. CAVEATS Be careful that local variables are restored in the reverse order in which they were localized. Consider those examples: local $x = 0; { reap sub { print $x } => HERE; local $x = 1; ... } # prints '0' ... { local $x = 1; reap sub { $x = 2 } => HERE; ... } # $x is 0 The first case is "solved" by moving the "local" before the "reap", and the second by using "localize" instead of "reap". The effects of "reap", "localize" and "localize_elem" can't cross "BEGIN" blocks, hence calling those functions in "import" is deemed to be useless. This is an hopeless case because "BEGIN" blocks are executed once while localizing constructs should do their job at each run. However, it's possible to hook the end of the current scope compilation with B::Hooks::EndOfScope. Some rare oddities may still happen when running inside the debugger. It may help to use a perl higher than 5.8.9 or 5.10.0, as they contain some context-related fixes. DEPENDENCIES XSLoader (standard since perl 5.006). SEE ALSO "local" in perlfunc, "Temporary Values via local()" in perlsub. Alias, Hook::Scope, Scope::Guard, Guard. Continuation::Escape is a thin wrapper around Scope::Upper that gives you a continuation passing style interface to "unwind". It's easier to use, but it requires you to have control over the scope where you want to return. Scope::Escape. Sub::Uplevel provides a pure-Perl implementation of "uplevel". AUTHOR Vincent Pit, "", . You can contact me by mail or on "irc.perl.org" (vincent). BUGS Please report any bugs or feature requests to "bug-scope-upper at rt.cpan.org", or through the web interface at . I will be notified, and then you'll automatically be notified of progress on your bug as I make changes. SUPPORT You can find documentation for this module with the perldoc command. perldoc Scope::Upper Tests code coverage report is available at . ACKNOWLEDGEMENTS Inspired by Ricardo Signes. Thanks to Shawn M. Moore for motivation. COPYRIGHT & LICENSE Copyright 2008,2009,2010,2011 Vincent Pit, all rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.