# Copyright 2008, 2009, 2010, 2011, 2012, 2015 Kevin Ryde # This file is part of constant-defer. # # constant-defer is free software; you can redistribute it and/or modify it # under the terms of the GNU General Public License as published by the Free # Software Foundation; either version 3, or (at your option) any later # version. # # constant-defer is distributed in the hope that it will be useful, but # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License # for more details. # # You should have received a copy of the GNU General Public License along # with constant-defer. If not, see . package constant::defer; use strict; use vars '$VERSION'; $VERSION = 6; sub import { my $class = shift; $class->_create_for_package (scalar(caller), @_); } sub _create_for_package { my $class = shift; my $target_package = shift; while (@_) { my $name = shift; if (ref $name eq 'HASH') { unshift @_, %$name; next; } unless (@_) { require Carp; Carp::croak ("Missing value sub for $name"); } my $subr = shift; ### $constant::defer::DEBUG_LAST_SUBR = $subr; my ($fullname, $basename); if ($name =~ /::([^:]*)$/s) { $fullname = $name; $basename = $1; } else { $basename = $name; $fullname = "${target_package}::$name"; } ## print "constant::defer $arg -- $fullname $basename $old\n"; $class->_validate_name ($basename); $class->_create_fullname ($fullname, $subr); } } sub _create_fullname { my ($class, $fullname, $subr) = @_; my $run = sub { unshift @_, $fullname, $subr; goto &_run; }; my $func = sub () { unshift @_, \$run; goto $run; }; no strict 'refs'; *$fullname = $func; ### $constant::defer::DEBUG_LAST_RUNNER = $run; } sub _run { my $fullname = shift; my $subr = shift; my $run_ref = shift; ### print "_run() $fullname $subr\n"; my @ret = &$subr(@_); if (@ret == 1) { # constant.pm has an optimization to make a constant by storing a scalar # value directly into the %{Foo::Bar::} hash if there's no typeglob for # the name yet. But that doesn't apply here, there's always a glob from # having converted a function. # # The function created only has name __ANON__ in its coderef GV (as # fetched by Sub::Identify for instance). This is the same as most # function creating modules, including Memoize.pm. Plain constant.pm # likewise, except when it uses the scalar ref in symbol table # optimization, in that case a later upgrade to a function gets a name. # my $value = $ret[0]; $subr = sub () { $value }; } elsif (@ret == 0) { $subr = \&_nothing; } else { $subr = sub () { @ret }; } $$run_ref = $subr; { no strict 'refs'; local $^W = 0; # no warnings 'redefine'; eval { *$fullname = $subr } or die $@; } goto $subr; } # not as strict as constant.pm sub _validate_name { my ($class, $name) = @_; if ($name =~ m{[()] # no parens like CODE(0x1234) if miscounted args |^[0-9] # no starting with a number |^$ # not empty }x) { require Carp; Carp::croak ("Constant name '$name' is invalid"); } } sub _nothing () { } ## no critic (ProhibitSubroutinePrototypes) 1; __END__ =for stopwords bareword stringizing inline there'd fakery subclassing Ryde multi-value inlined coderef subrs subr =head1 NAME constant::defer -- constant subs with deferred value calculation =for test_synopsis my ($some,$thing,$an,$other); =head1 SYNOPSIS use constant::defer FOO => sub { return $some + $thing; }, BAR => sub { return $an * $other; }; use constant::defer MYOBJ => sub { require My::Class; return My::Class->new_thing; } =head1 DESCRIPTION C creates a subroutine which on the first call runs given code to calculate its value, and on any subsequent calls just returns that value, like a constant. The value code is discarded once run, allowing it to be garbage collected. Deferring a calculation is good if it might take a lot of work or produce a big result but is only needed sometimes or only well into a program run. If it's never needed then the value code never runs. A deferred constant is generally not inlined or folded (see L) since it's not a single scalar value. In the current implementation a deferred constant becomes a plain constant after the first use, so may inline etc in code compiled after that (see L below). See F in the constant-defer source code for a complete sample program. =head2 Uses Here are some typical uses. =over 4 =item * A big value or slow calculation only sometimes needed, use constant::defer SLOWVALUE => sub { long calculation ...; return $result; }; if ($option) { print "s=", SLOWVALUE, "\n"; } =item * A shared object instance created when needed then re-used, use constant::defer FORMATTER => sub { return My::Formatter->new }; if ($something) { FORMATTER()->format ... } =item * The value code might load requisite modules too, again deferring that until actually needed, use constant::defer big => sub { require Some::Big::Module; return Some::Big::Module->create_something(...); }; =item * Once-only setup code can be created with no return value. The code is garbage collected after the first run and becomes a do-nothing. Remember to have an empty or C return value so as not to keep the last expression result alive forever. use constant::defer MY_INIT => sub { many lines of setup code ...; return; }; sub new { MY_INIT(); ... } =back =head1 IMPORTS There are no functions as such, everything works through the C import. =over 4 =item C<< use constant::defer NAME1=>SUB1, NAME2=>SUB2, ...; >> The parameters are name/subroutine pairs. For each one a sub called C is created, running the given C the first time its value is needed. C defaults to the caller's package, or a fully qualified name can be given. Remember that bareword stringizing of C<=E> doesn't act on a fully qualified name, so add quotes in that case. use constant::defer 'Other::Package::BAR' => sub { ... }; For compatibility with the C module a hash of name/sub arguments is accepted too. But C doesn't need this style since there's only ever one thing (a sub) following each name. use constant::defer { FOO => sub { ... }, BAR => sub { ... } }; # works without the hashref too use constant::defer FOO => sub { ... }, BAR => sub { ... }; =back =head1 MULTIPLE VALUES The value sub can return multiple values to make an array style constant sub. use constant::defer NUMS => sub { return ('one', 'two') }; foreach (NUMS) { print $_,"\n"; } The value sub is always run in array context, for consistency, irrespective how the constant is used. The return from the new constant sub is an array style sub () { return @result } If the value sub was a list-style return like C shown above, then this array-style return is slightly different. In scalar context a list return means the last value (like a comma operator), but an array return in scalar context means the number of elements. A multi-value constant won't normally be used in scalar context, so the difference shouldn't arise. The array style is easier for C to implement and is the same as the plain C module does. =head1 ARGUMENTS If the constant is called with arguments then they're passed on to the value sub. This can be good for constants used as object or class methods. Passing anything to plain constants would be unusual. A cute use for a class method style is to make a "singleton" instance of the class. See F in the constant-defer source code for a complete program. package My::Class; use constant::defer INSTANCE => sub { my ($class) = @_; return $class->new }; package main; $obj = My::Class->INSTANCE; Some care might be needed if letting a subclass object become the parent C, though if a program only ever used the subclass then that might in fact be desirable. Subs created by C always have prototype C<()>, ensuring they always parse the same way. The prototype has no effect when called as a method like above, but if you want to make a plain call with arguments then use C<&> to bypass the prototype (see L). &MYCONST ('Some value'); =head1 IMPLEMENTATION Currently C creates a sub under the requested name and when called it replaces that with a new constant sub the same as C would make. This is compact and means that later loaded code might be able to inline it. It's fine to keep a reference to the initial sub and in fact that happens quite normally if importing into another module (with the usual C), or an explicit C<\&foo>, or a C<$package-Ecan('foo')>. The initial sub changes itself to jump to the new constant, it doesn't re-run the value code. The jump is currently done by a C to the new coderef, so it's a touch slower than the new constant sub directly. A spot of XS would no doubt make the difference negligible, perhaps to the point where there'd be no need for a new sub, just have the initial transform itself. If the new form looked enough like a plain constant it might inline in later loaded code. For reference, C (as of its version 0.06) considers C subrs as constants, both before and after the first call which runs the value code. C just looks for prototype S> functions, so any such subr rates as a constant. =head1 OTHER WAYS TO DO IT There's many ways to do "deferred" or "lazy" calculations. =over 4 =item * C makes a function repeat its return. Results are cached against the arguments, so it keeps the original code, whereas C discards after the first run. =item * C and its friends make a create-once Cinstance> method. C can go close with the fakery shown under L above, though without a C to query. =item * C offers some syntactic sugar for redefining the running subroutine, including to a constant. =item * C can create an instance function for a class. It's designed for objects and so doesn't allow 0 or C as the return value. =item * C mangles the code in its caller to put a value as a constant there. The effect is to run a function just once at each caller location, replacing it with the return value, though the function can also choose to constant-ize only sometimes, based on circumstances, arguments, etc. =item * A C<$foo> scalar variable can be rigged up to run code on its first access. C uses a C. C and C use C on an object. C optimizes out the object from C after the first run. C uses XS magic (removed after the first fetch) and some parsing for syntactic sugar. The advantage of a variable is that it interpolates in strings. The disadvantages are it won't inline in later loaded code; bad XS code might bypass the magic; and package variables aren't very friendly when subclassing. =item * C and C rig up an object wrapper to load and create an actual object only when a method is called, dispatching to it and replacing the caller's C<$_[0]>. The advantage is you can pass the wrapper object around, etc, deferring creation to an even later time than a sub or scalar can. =item * C, C and C help make a defer class which transforms lazy stub objects to real ones when a method call is made. A separate defer class is required for each real class. =item * C sets up a run-once code block, but with no particular return value and not discarding the code after run. =item * C and C load code on a class method call such as object creation. They're mainly to defer module loading rather than calculation of a particular value. =item * C and C makes array elements calculated on-demand from a generator function. C does a similar thing for hashes. C hides a function behind a scalar; its laziness is in the arguments which are not evaluated until used. =back =head1 SEE ALSO L, L, L L, L, L, L, L, L, L, L, L, L, L, L, L, L, L, L, L, L =head1 HOME PAGE http://user42.tuxfamily.org/constant-defer/index.html =head1 COPYRIGHT Copyright 2009, 2010, 2011, 2012, 2015 Kevin Ryde constant-defer is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. constant-defer is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with constant-defer. If not, see . =cut