fix re debug segvs in global destruction, and a tweak to Benchmark to prevent infinit...

[p5sagit/p5-mst-13.2.git] / lib / version.pod

=head1 NAME

version - Perl extension for Version Objects

=head1 SYNOPSIS

  use version;
  $version = version->new("12.2.1"); # must be quoted for Perl < 5.8.1
  print $version;               # v12.2.1
  print $version->numify;       # 12.002001
  if ( $version gt "12.2" )     # true

  $alphaver = version->new("1.02_03"); # must be quoted!
  print $alphaver;              # 1.02_0300
  print $alphaver->is_alpha();  # true
  
  $ver = qv("1.2.0");           # v1.2.0

  $perlver = version->new(5.005_03); # must not be quoted!
  print $perlver;               # 5.005030

=head1 DESCRIPTION

Overloaded version objects for all versions of Perl.  This module
implements all of the features of version objects which will be part
of Perl 5.10.0.

=head2 BEST PRACTICES

If you intend for your module to be used by different releases of Perl,
and/or for your $VERSION scalar to mean what you think it means, there 
are a few simple rules to follow:

=over 4

=item * Be consistent

Whichever of the two types of version objects that you choose to employ, 
you should stick to either L<Numeric Versions> or L<Extended Versions>
and not mix them together.  While this is I<possible>, it is very 
confusing to the average user.

If you intend to use L<Extended Versions>, you are strongly encouraged 
to use the L<qv()> operator with a quoted term, e.g.:

  use version; our $VERSION = qv("1.2.3");

on a single line as above. 

At the very least, decide on which of the several ways to initialize 
your version objects you prefer and stick with it.  It is also best to 
be explicit about what value you intend to assign your version object 
and to not rely on hidden behavior of the parser. 

=item * Be careful

If you are using Module::Build or ExtUtils::MakeMaker, so that you can
release your module to CPAN, you have to recognize that neither of those
programs completely handles version objects natively (yet).  If you use
version objects with Module::Build, you should add an explicit dependency
to the release of version.pm in your Build.PL:

  my $builder = Module::Build->new(
     ...
     requires => {
         ... ,
         'version'    => 0.50,
         ...,
     },
     ...
  );

and it should Just Work(TM).  Module::Build will [hopefully soon] 
include full support for version objects; there are no current plans 
to patch ExtUtils::MakeMaker to support version objects.

=head2 What IS a version

For the purposes of this module, a version "number" is a sequence of
positive integer values separated by one or more decimal points and 
optionally a single underscore.  This corresponds to what Perl itself 
uses for a version, as well as extending the "version as number" that 
is discussed in the various editions of the Camel book.

There are actually two distinct kinds of version objects:

=over 4

=item * Numeric Versions

Any initial parameter which "looks like a number", see L<Numeric
Versions>.  This also covers versions with a single decimal point and
a single embedded underscore, see L<Numeric Alpha Versions>, even though
these must be quoted to preserve the underscore formatting.

=item * Extended Versions

Any initial parameter which contains more than one decimal point
and an optional embedded underscore, see L<Extended Versions>.  This 
is what is commonly used in most open source software as the "external"
version (the one used as part of the tag or tarfile name).  The use
of the exported L<qv()> function also produces this kind of version
object.

=back

Both of these methods will produce similar version objects, in that
the default stringification will yield the version L<Normal Form> only 
if required:

  $v  = version->new(1.002);     # 1.002, but compares like 1.2.0
  $v  = version->new(1.002003);  # 1.002003
  $v2 = version->new( "1.2.3");  # v1.2.3

In specific, version numbers initialized as L<Numeric Versions> will
stringify in Numeric form.  Version numbers initialized as L<Extended Versions>
will be stringified as L<Normal Form>.

=head2 Numeric Versions

These correspond to historical versions of Perl itself prior to 5.6.0,
as well as all other modules which follow the Camel rules for the
$VERSION scalar.  A numeric version is initialized with what looks like
a floating point number.  Leading zeros B<are> significant and trailing
zeros are implied so that a minimum of three places is maintained
between subversions.  What this means is that any subversion (digits
to the right of the decimal place) that contains less than three digits
will have trailing zeros added to make up the difference, but only for
purposes of comparison with other version objects.  For example:

                                   # Prints     Equivalent to  
  $v = version->new(      1.2);    # 1.200      v1.200.0
  $v = version->new(     1.02);    # 1.020      v1.20.0
  $v = version->new(    1.002);    # 1.002      v1.2.0
  $v = version->new(   1.0023);    # 1.002300   v1.2.300
  $v = version->new(  1.00203);    # 1.002030   v1.2.30
  $v = version->new( 1.002003);    # 1.002003   v1.2.3

All of the preceding examples are true whether or not the input value is 
quoted.  The important feature is that the input value contains only a 
single decimal.  See also L<Alpha Versions> for how to handle 

IMPORTANT NOTE: As shown above, if your numeric version contains more 
than 3 significant digits after the decimal place, it will be split on 
each multiple of 3, so 1.0003 is equivalent to v1.0.300, due to the need 
to remain compatible with Perl's own 5.005_03 == 5.5.30 interpretation.  
Any trailing zeros are ignored for mathematical comparison purposes.

=head2 Extended Versions

These are the newest form of versions, and correspond to Perl's own
version style beginning with 5.6.0.  Starting with Perl 5.10.0,
and most likely Perl 6, this is likely to be the preferred form.  This
method normally requires that the input parameter be quoted, although 
Perl's after 5.8.1 can use v-strings as a special form of quoting, but
this is highly discouraged.

Unlike L<Numeric Versions>, Extended Versions have more than
a single decimal point, e.g.:

                                   # Prints
  $v = version->new( "v1.200");    # v1.200.0
  $v = version->new("v1.20.0");    # v1.20.0
  $v = qv("v1.2.3);                # v1.2.3
  $v = qv("1.2.3");                # v1.2.3
  $v = qv("1.20");                 # v1.20.0

In general, Extended Versions permit the greatest amount of freedom
to specify a version, whereas Numeric Versions enforce a certain
uniformity.  See also L<New Operator> for an additional method of
initializing version objects.

Just like L<Numeric Versions>, Extended Versions can be used as 
L<Alpha Versions>.

=head2 Numeric Alpha Versions

The one time that a numeric version must be quoted is when a alpha form is
used with an otherwise numeric version (i.e. a single decimal point).  This
is commonly used for CPAN releases, where CPAN or CPANPLUS will ignore alpha
versions for automatic updating purposes.  Since some developers have used
only two significant decimal places for their non-alpha releases, the
version object will automatically take that into account if the initializer
is quoted.  For example Module::Example was released to CPAN with the
following sequence of $VERSION's:

  # $VERSION    Stringified
  0.01          0.010
  0.02          0.020
  0.02_01       0.02_0100
  0.02_02       0.02_0200
  0.03          0.030
  etc.

As you can see, the version object created from the values in the first
column may contain a trailing 0, but will otherwise be both mathematically
equivalent and sorts alpha-numerically as would be expected.

=head2 Object Methods

Overloading has been used with version objects to provide a natural
interface for their use.  All mathematical operations are forbidden,
since they don't make any sense for base version objects.

=over 4

=item * New Operator

Like all OO interfaces, the new() operator is used to initialize
version objects.  One way to increment versions when programming is to
use the CVS variable $Revision, which is automatically incremented by
CVS every time the file is committed to the repository.

In order to facilitate this feature, the following
code can be employed:

  $VERSION = version->new(qw$Revision: 2.7 $);

and the version object will be created as if the following code
were used:

  $VERSION = version->new("v2.7");

In other words, the version will be automatically parsed out of the
string, and it will be quoted to preserve the meaning CVS normally
carries for versions.  The CVS $Revision$ increments differently from
numeric versions (i.e. 1.10 follows 1.9), so it must be handled as if
it were a L<Extended Version>.

A new version object can be created as a copy of an existing version
object, either as a class method:

  $v1 = version->new(12.3);
  $v2 = version->new($v1);

or as an object method:

  $v1 = version->new(12.3);
  $v2 = $v1->new();

and in each case, $v1 and $v2 will be identical.

=back

=over 4

=item * qv()

An alternate way to create a new version object is through the exported
qv() sub.  This is not strictly like other q? operators (like qq, qw),
in that the only delimiters supported are parentheses (or spaces).  It is
the best way to initialize a short version without triggering the floating
point interpretation.  For example:

  $v1 = qv(1.2);         # 1.2.0
  $v2 = qv("1.2");       # also 1.2.0

As you can see, either a bare number or a quoted string can usually 
be used interchangably, except in the case of a trailing zero, which
must be quoted to be converted properly.  For this reason, it is strongly
recommended that all initializers to qv() be quoted strings instead of
bare numbers.

To prevent the C<qv()> function from being exported to the caller's namespace,
either use version with a null parameter:

  use version ();

or just require version, like this:

  require version;

Both methods will prevent the import() method from firing and exporting the
C<qv()> sub.  This is true of subclasses of version as well, see
L<SUBCLASSING> for details.

=back

For the subsequent examples, the following three objects will be used:

  $ver   = version->new("1.2.3.4"); # see "Quoting" below
  $alpha = version->new("1.2.3_4"); # see "Alpha versions" below
  $nver  = version->new(1.002);     # see "Numeric Versions" above

=over 4

=item * Normal Form

For any version object which is initialized with multiple decimal
places (either quoted or if possible v-string), or initialized using
the L<qv()> operator, the stringified representation is returned in
a normalized or reduced form (no extraneous zeros), and with a leading 'v':

  print $ver->normal;         # prints as v1.2.3.4
  print $ver->stringify;      # ditto
  print $ver;                 # ditto
  print $nver->normal;        # prints as v1.2.0
  print $nver->stringify;     # prints as 1.002, see "Stringification" 

In order to preserve the meaning of the processed version, the 
normalized representation will always contain at least three sub terms.
In other words, the following is guaranteed to always be true:

  my $newver = version->new($ver->stringify);
  if ($newver eq $ver ) # always true
    {...}

=back

=over 4

=item * Numification

Although all mathematical operations on version objects are forbidden
by default, it is possible to retrieve a number which corresponds 
to the version object through the use of the $obj->numify
method.  For formatting purposes, when displaying a number which
corresponds a version object, all sub versions are assumed to have
three decimal places.  So for example:

  print $ver->numify;         # prints 1.002003004
  print $nver->numify;        # prints 1.002

Unlike the stringification operator, there is never any need to append
trailing zeros to preserve the correct version value.

=back

=over 4

=item * Stringification

In order to mirror as much as possible the existing behavior of ordinary
$VERSION scalars, the stringification operation will display differently,
depending on whether the version was initialized as a L<Numeric Version>
or L<Extended Version>.

What this means in practice is that if the normal CPAN and Camel rules are
followed ($VERSION is a floating point number with no more than 3 decimal
points), the stringified output will be exactly the same as the numified
output.  There will be no visible difference, although the internal 
representation will be different, and the L<Comparison operators> will 
function using the internal coding.

If a version object is initialized using a L<Extended Version> form, then
the stringified form will be the L<Normal Form>.  The $obj->normal
operation can always be used to produce the L<Normal Form>, even if the
version was originally a L<Numeric Version>.

  print $ver->stringify;    # prints v1.2.3.4
  print $nver->stringify;   # prints 1.002

=back

=over 4

=item * Comparison operators

Both C<cmp> and C<E<lt>=E<gt>> operators perform the same comparison between
terms (upgrading to a version object automatically).  Perl automatically
generates all of the other comparison operators based on those two.
In addition to the obvious equalities listed below, appending a single
trailing 0 term does not change the value of a version for comparison
purposes.  In other words "v1.2" and "1.2.0" will compare as identical.

For example, the following relations hold:

  As Number        As String           Truth Value
  -------------    ----------------    -----------
  $ver >  1.0      $ver gt "1.0"       true
  $ver <  2.5      $ver lt             true
  $ver != 1.3      $ver ne "1.3"       true
  $ver == 1.2      $ver eq "1.2"       false
  $ver == 1.2.3.4  $ver eq "1.2.3.4"   see discussion below

It is probably best to chose either the numeric notation or the string
notation and stick with it, to reduce confusion.  Perl6 version objects
B<may> only support numeric comparisons.  See also L<Quoting>.

WARNING: Comparing version with unequal numbers of decimal points (whether
explicitly or implicitly initialized), may yield unexpected results at
first glance.  For example, the following inequalities hold:

  version->new(0.96)     > version->new(0.95); # 0.960.0 > 0.950.0
  version->new("0.96.1") < version->new(0.95); # 0.096.1 < 0.950.0

For this reason, it is best to use either exclusively L<Numeric Versions> or
L<Extended Versions> with multiple decimal points.

=back

=over 4

=item * Logical Operators 

If you need to test whether a version object
has been initialized, you can simply test it directly:

  $vobj = version->new($something);
  if ( $vobj )   # true only if $something was non-blank

You can also test whether a version object is an L<Alpha version>, for
example to prevent the use of some feature not present in the main
release:

  $vobj = version->new("1.2_3"); # MUST QUOTE
  ...later...
  if ( $vobj->is_alpha )       # True

=back

=head2 Quoting

Because of the nature of the Perl parsing and tokenizing routines,
certain initialization values B<must> be quoted in order to correctly
parse as the intended version, especially when using the L<qv()> operator.
In all cases, a floating point number passed to version->new() will be
identically converted whether or not the value itself is quoted.  This is
not true for L<qv()>, however, when trailing zeros would be stripped on
an unquoted input, which would result in a very different version object.

In addition, in order to be compatible with earlier Perl version styles,
any use of versions of the form 5.006001 will be translated as v5.6.1.  
In other words, a version with a single decimal point will be parsed as
implicitly having three digits between subversions, but only for internal
comparison purposes.

The complicating factor is that in bare numbers (i.e. unquoted), the
underscore is a legal numeric character and is automatically stripped
by the Perl tokenizer before the version code is called.  However, if
a number containing one or more decimals and an underscore is quoted, i.e.
not bare, that is considered a L<Alpha Version> and the underscore is
significant.

If you use a mathematic formula that resolves to a floating point number,
you are dependent on Perl's conversion routines to yield the version you
expect.  You are pretty safe by dividing by a power of 10, for example,
but other operations are not likely to be what you intend.  For example:

  $VERSION = version->new((qw$Revision: 1.4)[1]/10);
  print $VERSION;          # yields 0.14
  $V2 = version->new(100/9); # Integer overflow in decimal number
  print $V2;               # yields something like 11.111.111.100

Perl 5.8.1 and beyond will be able to automatically quote v-strings but
that is not possible in earlier versions of Perl.  In other words:

  $version = version->new("v2.5.4");  # legal in all versions of Perl
  $newvers = version->new(v2.5.4);    # legal only in Perl >= 5.8.1

=head2 What about v-strings?

Beginning with Perl 5.6.0, an alternate method to code arbitrary strings
of bytes was introduced, called v-strings.  They were intended to be an
easy way to enter, for example, Unicode strings (which contain two bytes
per character).  Some programs have used them to encode printer control
characters (e.g. CRLF).  They were also intended to be used for $VERSION,
but their use as such has been problematic from the start.

There are two ways to enter v-strings: a bare number with two or more
decimal points, or a bare number with one or more decimal points and a 
leading 'v' character (also bare).  For example:

  $vs1 = 1.2.3; # encoded as \1\2\3
  $vs2 = v1.2;  # encoded as \1\2 

However, the use of v-strings to initialize version objects with this 
module is only possible with Perl 5.8.1 or better (which contain special
code to enable it).  Their use is B<strongly> discouraged in all 
circumstances (especially the leading 'v' style), since the meaning will
change depending on which Perl you are running.  It is better to directly 
use L<"Extended Versions"> to ensure the proper interpretation.


=head2 Types of Versions Objects

There are two types of Version Objects:

=over 4

=item * Ordinary versions

These are the versions that normal modules will use.  Can contain as
many subversions as required.  In particular, those using RCS/CVS can
use the following:

  $VERSION = version->new(qw$Revision: 2.7 $);

and the current RCS Revision for that file will be inserted
automatically.  If the file has been moved to a branch, the Revision
will have three or more elements; otherwise, it will have only two.
This allows you to automatically increment your module version by
using the Revision number from the primary file in a distribution, see
L<ExtUtils::MakeMaker/"VERSION_FROM">.

=item * Alpha Versions

For module authors using CPAN, the convention has been to note
unstable releases with an underscore in the version string, see
L<CPAN>.  Alpha releases will test as being newer than the more recent
stable release, and less than the next stable release.  For example:

  $alphaver = version->new("12.03_01"); # must be quoted

obeys the relationship

  12.03 < $alphaver < 12.04

Alpha versions with a single decimal point will be treated exactly as if
they were L<Numeric Versions>, for parsing purposes.  The stringification for
alpha versions with a single decimal point may seem surprising, since any
trailing zeros will visible.  For example, the above $alphaver will print as

  12.03_0100

which is mathematically equivalent and ASCII sorts exactly the same as
without the trailing zeros.

Alpha versions with more than a single decimal point will be treated 
exactly as if they were L<Extended Versions>, and will display without any
trailing (or leading) zeros, in the L<Version Normal> form.  For example,

  $newver = version->new("12.3.1_1");
  print $newver; # v12.3.1_1

=head2 Replacement UNIVERSAL::VERSION

In addition to the version objects, this modules also replaces the core
UNIVERSAL::VERSION function with one that uses version objects for its
comparisons.  The return from this operator is always the numified form,
and the warning message generated includes both the numified and normal
forms (for clarity).

For example:

  package Foo;
  $VERSION = 1.2;

  package Bar;
  $VERSION = "1.3.5"; # works with all Perl's (since it is quoted)

  package main;
  use version;

  print $Foo::VERSION; # prints 1.2

  print $Bar::VERSION; # prints 1.003005

  eval "use CGI 10"; # some far future release
  print $@; # prints "CGI version 10 (10.0.0) required..."

IMPORTANT NOTE: This may mean that code which searches for a specific
string (to determine whether a given module is available) may need to be
changed.

The replacement UNIVERSAL::VERSION, when used as a function, like this:

  print $module->VERSION;

will also exclusively return the numified form.  Technically, the 
$module->VERSION function returns a string (PV) that can be converted to a 
number following the normal Perl rules, when used in a numeric context.

=head1 SUBCLASSING

This module is specifically designed and tested to be easily subclassed.
In practice, you only need to override the methods you want to change, but
you have to take some care when overriding new() (since that is where all
of the parsing takes place).  For example, this is a perfect acceptable
derived class:

  package myversion;
  use base version;
  sub new { 
      my($self,$n)=@_;
      my $obj;
      # perform any special input handling here
      $obj = $self->SUPER::new($n);
      # and/or add additional hash elements here
      return $obj;
  }

See also L<version::AlphaBeta> on CPAN for an alternate representation of
version strings.

B<NOTE:> Although the L<qv> operator is not a true class method, but rather a
function exported into the caller's namespace, a subclass of version will 
inherit an import() function which will perform the correct magic on behalf
of the subclass.

=head1 EXPORT

qv - Extended Version initialization operator

=head1 AUTHOR

John Peacock E<lt>jpeacock@cpan.orgE<gt>

=head1 SEE ALSO

L<perl>.

=cut