=for Pod::Coverage empty
=head1 NAME
File::Globstar - Globbing With Double Asterisk Expansion
=head1 SYNOPSIS
use File::Globstar qw(globstar fnmatchstar);
@files = globstar '**/*.css';
@files = globstar 'css/**/*.css';
@files = globstar 'scss/**';
print "Match!\n" if fnmatchstar '*.pl', 'hello.pl';
print "Case-insensitive match!\n"
if fnmatchstar '*.pl', 'Makefile.PL', ignoreCase => 1;
$re = File::Globstar::translatestar('**/*.css');
$safe_pattern = quotestar $config_srcdir;
=head1 DESCRIPTION
Shortcut: If you want to implement file inclusion or exclusion in the style
of L<.gitignore|https://git-scm.com/docs/gitignore>, have a look at
L<File::Globstar::ListMatch>.
Many globbing implementations have been recently extended to accept the
pattern F<**> in place of a path element. The double asterisk matches
all files and directories in the current directory and all of its
descendants including the current directory itself (in other words: the
directory F<.> is included in the match, the parent directory F<..> is
not).
The convention is especially popular
in the Node.js ecosystem and is also used by L<git|https://git-scm.com/>
when evaluating ignore patterns (see L<https://git-scm.com/docs/gitignore>).
=head2 FUNCTIONS
=over 4
=item B<globstar PATTERN[, DIRECTORY]>
Return all files and directories matching B<PATTERN> in B<DIRECTORY>.
B<DIRECTORY> defaults to the current directory if empty or
undefined.
An invalid B<PATTERN> matches nothing and B<globstar()> returns an empty
list for it.
The function I<only> expand the sequence "**" (at the appropriate) positions.
All other heavy lifting is done by L<File::Glob::bsd_glob()> (which is also
the backend for the standard Perl B<glob> operator <HANDLE>.
Currently, only the (forward!) slash is accepted as a path separator!
=item B<fnmatchstar PATTERN, STRING[, OPTIONS]>
Returns 1 if B<STRING> matches B<PATTERN>, false otherwise. If a Perl truthy
value is passed as the optional third argument, case is ignored for the
match.
B<OPTIONS> is an optional hash of named arguments. The only
supported option is "ignoreCase" at the moment.
Invalid B<PATTERN>s never match.
Unlike B<globstar()>, the function does not rely on L<File::Glob> and is
implemented entirely in Perl. The semantics of B<PATTERN> are as follows:
=over 8
=item B<*>
A single asterisk stands for zero or more arbitrary characters except for the
slash C</>.
=item B<?>
The question mark stands for exactly one character except for the slash C</>.
=item B<**>
A double asterisk stands for an arbitrary sequence of 0 or more characters.
It is only allowed when preceded by either the beginning of the string or
a slash. Likewise it must be followed by a slash or the end of the pattern.
=item B<[RANGE]>, B<[!RANGE]>
A character range or a negated character range that is preceded by an
exclamation mark. A range cannot be empty.
The following features are supported:
=over 12
=item B<CHARACTER>
By default, all B<CHARACTER>s stand for themselves.
=item B<CHARACTER1-CHARACTER2>
All characters from B<CHARACTER1> to B<CHARACTER2> according to the collation
valid for the currently selected locale.
=item B<[:CLASS:]>
A character class like "[:print:]", "[:upper:]", "[:lower:]". Example:
[0-9[:lower:]]
This pattern stands for exactly one character, either one of the ASCII
digits 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9 or any lowercase character
(according to the current locale).
=item B<\CHARACTER>
Any character with a special meaning can be escaped with a backslash.
You can include a hyphen in a range by using it as the last character. You
can include a closing square bracket by using it as the first character.
This is implied by the above rules.
Example:
[]-]
This pattern matches either a closing square bracket "]" or a hyphen "-".
=back
Note that POSIX collation classes (for example "[.ch.]") and POSIX equivalence
classes (for example "[=a=]") are not supported. Rationale: B<bsd_glob()>
does not support them and in Perl regular expression character classes they
are currently not supported and their usage throws an exception.
=back
=item B<pnmatchstar PATTERN, STRING[, OPTIONS]>
Returns 1 if B<STRING> matches B<PATTERN>, false otherwise.
Similar to B<fnmatchstar()> but B<PATTERN> and B<STRING> are subject to some
pre-processing and matching is eventually done "ascendingly". B<pnmatchstar()>
is the function underneath L<File::Globstar::ListMatch>. You can think of it
as a matching engine for one pattern (line) in a
L<.gitignore|https://git-scm.com/docs/gitignore> or similar file.
B<PATTERN> is either a (scalar) string that will be preprocessed and then
compiled with B<translatestar()> or a compiled regular expression. In the
latter case, the compilation step is skipped. You should pass a regular
expression that was returned by a B<translatestar()> call with the option
B<pathMode>.
The pre-processing of B<PATTERN> goes through the steps outlined below. The
intermediate results for the input B<STRING> "!/src/**/.deps/" is added in
parentheses.
=over 8
=item B<*> If B<PATTERN> begins with an exclamation mark, it is discarded,
and the B<PATTERN> is marked as negated (current "/src/**/.deps/").
=item B<*> If B<PATTERN> ends with a slash, it is discarded,
and the B<PATTERN> is marked as a "directory match" (current "/src/**/.deps").
=item B<*> If B<PATTERN> begins with a slash, it is discarded,
and the B<PATTERN> is marked as a "full path match" (current "src/**/.deps").
=item B<*> If B<PATTERN> is not already marked as a "full path match", every
remaining slash will activate "full path match" (no modification in this
last step).
=back
Note that the pre-processing step is skipped, if you pass a compiled regular
expression instead of a (scalar) string for B<PATTERN>. And to be honest,
B<pnmatchstar()> does not really do the pre-processing. It is in fact done by
B<translatestar()> when called with the option B<pathMode>. See
below for details.
The B<STRING> argument (for example "lib/File/Globstar/") is always
pre-processed:
=over 8
=item B<*> A possible trailing slash is discarded, and the B<STRING> is then
marked as representing a "directory". The example string would now be
"lib/File/Globstar" without the trailing slash.
=item B<*> If full path mode is B<NOT> active (either because of a leading
or innner slash in B<PATTERN>), a possible leading "directory portion" of
B<STRING> is discarded. The "directory portion" is everything until and
including the last slash in B<STRING>. The example string would have been
truncated to just "Globstar" by now.
=back
The resulting B<STRING> is then compared to B<PATTERN>. If it matches, and
B<PATTERN> was not negated, 1 is returned. Likewise, if it doesn't match and
B<PATTERN> was negated.
In "directory mode" (B<PATTERN> ended with a slash), a match also requires
that B<STRING> is considered a directory. It either needs a trailing slash
or you must pass the option B<isDirectory> with a truthy value.
In full path mode, the comparison continues and "ascends": The "base name"
of B<STRING>, that is the last slash and everything following it is discarded,
and B<pnmatchstar()> is called recursively, but always with the option
B<isDirectory> set to 1 until a match is found. If it never matches,
the function ultimately returns false.
You can pass the following optional named arguments to B<pnmatchstar()>:
=over 8
=item B<ignoreCase TRUE|FALSE>
Compile the regular expression with the "i" modifier so that it matches in a
case-insensitive manner.
The option is ignored, when a compiled regular expression was passed as the
first argument instead of a scalar B<PATTERN>.
=item B<isDirectory TRUE|FALSE>
Mark B<STRING> as a directory, if B<TRUE>.
The option is ignored, when a compiled regular expression was passed
as the first argument instead of a scalar B<PATTERN>. It is also ignored
(forced to B<TRUE>), when a trailing slash was removed from B<STRING>.
Rationale: The name of a non-directory can never contain a slash.
=back
The function was introduced in version 0.2.
=item B<translatestar PATTERN[, OPTIONS]>
Compiles B<PATTERN> into a Perl regular expression or throws an exception in
case of failure. This function is used internally by B<fnmatchstar()> and
B<pnmatchstar()>.
B<OPTIONS> is an optional hash of named arguments:
=over 4
=item B<ignoreCase TRUE|FALSE>
If B<TRUE>, return a regular expression with the "i" modifier set, so that
case is ignored, when matching.
=item B<pathMode TRUE|FALSE>
Remove a possible leading exclamation mark from B<PATTERN> and mark the
regular expression in that case as negated. If a trailing slash could
be removed, the regular expression will be marked as a "directory match".
If B<PATTERN> contains more slashes, the regular expression is marked as
a "full path match".
Inner slashes are preserved. A leading slash is discarded, and only triggers
"full path match" mode.
See B<pnmatchstar()> above or the section L</"EXAMPLES"> below for more
details.
The option was intrucded in version 0.2.
=back
=item B<quotestar STRING[, NEGATABLE]>
Escapes all characters special to globstar patterns. This are:
=over 8
=item B<\\> the backslash
=item B<*> the asterisk/star
=item B<[> the opening square bracket
=item B<]> the closing square bracket (actually unneeded but ...)
=back
If the optional argument B<NEGATABLE> is truthy, a leading exclamation mark
will also be backslash-escaped.
=back
=head2 EXAMPLES
We start with some examples for the double asterisk pattern.
=over 4
=item B<**/*.css>
All files matching the name F<*.css> in the current directory and all its
descendants, for example F<main.css>, F<styles/body.css>,
F<styles/body/footer.css>, F<styles/footer/whatever.css> and so on.
=item B<src/**>
The directory F<src> and all files and directories underneath F<src>, for
example F<src/a>, F<src/b>, F<src/a/x/y>, and so on.
=item B<src/**/*.c>
All F<*.c> files in the directory F<src> or one if its descendants, for
example F<src/file.c>, F<src/a/file.c>, F<src/b/file.c>, F<src/a/x/file.c>,
and so on.
=back
Recursively find all Perl source files in F<lib>:
@sources = globstar 'lib/**/*.p[lm]';
Is F<hello.pl> in the current directory a Perl source file?
fnmatchstar '*.p[lm]', 'hello.pl';
Is F<src/simple/hello.pl> a Perl source file?
fnmatchstar '*.p[lm]', 'src/simple/hello.pl';
require Carp;
Carp::croak("bug alert!");
The above does not work because a star does not match a slash. You need a different
pattern:
fnmatchstar '**/*.p[lm]', 'src/simple/hello.pl';
fnmatchstar 'src/**/*.p[lm]', 'src/simple/hello.pl';
Both patterns work for the example but the second pattern only matches in
directory 'src' and its descendants.
With B<pnmatchstar()> you have more options:
pnmatchstar '*.p[lm]', 'src/simple/hello.pl';
Returns 1 because F<src/simple> is discarded for matching. Only the
basename portion takes part in the match.
pnmatchstar '!*.p[lm]', 'src/simple/hello.pl';
Returns false because the pattern is negated.
pnmatchstar 'src/', 'src/simple/hello.pl';
Returns 1! F<src/simple/hello.pl> does not match. F<src/simple> is tried
next but again at no avail. F<src> ultimately matches.
The following example may now be surprising:
pnmatchstar 'src/', 'src';
Does not match because B<pnmatchstar()> does not consider F<src> a
directory. Do this instead:
pnmatchstar 'src/', 'src/';
pnmatchstar 'src/', 'src', isDirectory => 1;
Both above examples work, because it is now clear that the string stands for
a directory.
Now for full path matching.
pnmatchstar 'lib/File', 'lib/File/Globstar/ListMatch.pm';
Matches because the leading part F<lib/File> matches. But remember that
this does not work:
pnmatchstar 'File/Globstar, 'lib/File/Globstar/ListMatch.pm';
pnmatchstar 'File/Globstar/ListMatch.pm, lib/File/Globstar/ListMatch.pm';
Full path mode is full path mode. A partial match is not enough.
If you use the same patterns multiple times, you can cache them:
$regex = translatestar 'src/**/*.p[lm]', pathMode => 1;
foreach $file (@files) {
print "$file matches\n" if pnmatchstar $regex, $file;
}
You should have realized by now that B<pnmatchstar()> is hard to describe
but easy to understand.
=head2 GORY DETAILS
The function B<globstar()> is a wrapper around B<bsd_glob> (see
L<File::Glob>) that extends the functionality with the double asterisk
semantics. A double asterisk is only allowed in the following cases:
=over 4
=item B<F<**/>...>
At the beginning of a pattern, when followed by a slash.
=item F<...B</**/>...>
Anywhere inside a pattern, when preceded and followed by a slash.
=item B<...F</**>>
At the end of a pattern, when preceded by a slash.
=item B<F<**>>
The pattern F<**> expands to all files and directories in the current
directory with arbitrary nesting level.
All other usages of two consecutive asterisks are considered illegal. An
illegal pattern does not match any file. The file F<foo***bar> for example
does I<not> match the pattern F<foo***bar>. It matches, however,
F<foo\*\*\*bar>.
Of course, all other features of L<File::Glob::bsd_glob()> are supported
as well. The module also suffers from the same bugs and incompatibilities.
=back
=head1 BUGS AND CAVEATS
Other than the ones currently unknown, be prepared for the following:
=over 4
=item B<COMPATIBILITY>
The module assumes that the (forward!) slash "/" is the only valid path
separator. Especially the backslash "\\" is only used for escaping and
never as a path separator. On the other hand, the underlying function
B<bsd_glob()> from B<File::Glob> I<does> support other path separators than
a slash.
This behavior may change in the future. The main reason why the backslash
is not supported as a path separator is that I have no idea how B<bsd_glob()>
behaves exactly, for example under Windows. Patches are welcome!
=item B<UNIFORM BEHAVIOR OF GLOBSTAR() AND FNMATCHSTAR()>
The functions should theoretically behave accordingly. If B<globstar()>
returns a filename for a certain pattern, then B<fnmatchstar()> should
return true for the same pattern and the filename passed as a string.
Unfortunately, the two implementations are completely independent, and this
cannot be guaranteed. Please file a bug if you find a discrepancy.
=item B<CASE-INSENSITIVE FILE SYSTEMS>
The behavior of B<globstar()> completely depends on the behavior of
the underlying L<File::Glob>. For B<fnmatchstar()> you can pass a third
argument specifying whether the match should be done case-insensitively or
not.
For real-world scenarios you should be aware that on non-Windows systems,
every directory level could behave differently: While "/media/disk1/backup"
can be case-sensitive, "/media/disk1/backup/data" could be case-preserving
and "/media/disk2/mp3" could be case-insensitive.
=back
=head1 COPYRIGHT
Copyright (C) 2016-2017 Guido Flohr <guido.flohr@cantanea.com>,
all rights reserved.
=head1 SEE ALSO
File::Glob(3pm), File::Globstar::Listmatch(3pm), glob(3), glob(7), fnmatch(3),
glob(1), perl(1)