Enhanced glob processing
use File::Wildcard;
my $foo = File::Wildcard->new(path => "/home/me///core");
while (my $file = $foo->next) {
unlink $file;
}
When looking at how various operating systems do filename wildcard expansion (globbing), \s-1VMS\s0 has a nice syntax which allows expansion and searching of whole directory trees. It would be nice if other operating systems had something like this built in. The best Unix can manage is through the utility program \*(C`find\*(C'.
This module provides this facility to Perl. Whereas native \s-1VMS\s0 syntax uses the ellipsis \*(L"...\*(R", this will not fit in with \s-1POSIX\s0 filenames, as ... is a valid (though somewhat strange) filename. Instead, the construct \*(L"///\*(R" is used as this cannot syntactically be part of a filename, as you do not get three concurrent filename separators with nothing between (three slashes are used to avoid confusion with //node/path/name syntax).
You don't have to use this syntax, as you can do the splitting yourself and pass in an arrayref as your path.
The module also forms a regular expression for the whole of the wildcard string, and binds a series of back references ($1, $2 etc.) which are available to construct new filenames.
\*(C`File::Wildcard-\*(C'new( $wildcard, [,option => value,...]);>
my $foo = File::Wildcard->new( path => "/home/me///core"); my $srcfnd = File::Wildcard->new( path => "src///*.cpp", match => qr(^src/(.*?)\.cpp$), derive => ['src/$1.o','src/$1.hpp']);
This is the constructor for File::Wildcard objects. At a simple level, pass a single wildcard string as a path.
For more complicated operations, you can supply your own match regexp, or use the derive option to specify regular expression captures to form the basis of other filenames that are constructed for you.
The $srcfnd example gives you object files and header files corresponding to \*(C+ source files.
Here are the options that are available: This is the input parameter that specifies the range of files that will be looked at. This is a glob spec which can also contain the ellipsis '///' (it could contain more than one ellipsis, but the benefit of this is questionable, and multiple ellipsi would cause a performance hit). Note that the path can be relative or absolute. new will do the right thing, working out that a path starting with '/' is absolute. In order to recurse from the current directory downwards, specify './//foo'. As an alternative, you can supply an arrayref with the path constituents already split. If you do this, you need to tell new if the path is absolute. Include an empty string for an ellipsis. For example: 'foo///bar/*.c' is equivalent to ['foo','','bar','*.c'] You can also construct a File::Wildcard without a path. A call to next will return undef, but paths can be added using the append and prepend methods. This is ignored unless you are using a pre split path. If you are passing a string as the path, new will work out whether the path is absolute or relative. Pass a true value for absolute paths. If your original filespec started with '/' before you split it, specify absolute => 1. absolute is not required for Windows if the path contains a drive specification, e.g. C:/foo/bar. By default, the module will use Filesys::Type to determine whether the file system of your wildcard is defined. This is an optional module (see Module::Optional), and File::Wildcard will guess at case sensitivity based on your operating system. This will not always be correct, as the file system might be \s-1VFAT\s0 mounted on Linux or \s-1ODS-5\s0 on \s-1VMS\s0. Specifying the option \*(C`case_insensitive\*(C' explicitly forces this behaviour on the wildcard. Note that File::Wildcard will use the file system of the current working directory if the path is not absolute. If the path is absolute, you should specify the case_sensitivity option explicitly. You can provide a regexp to apply to any generated paths, which will cause any matching paths not to be processed. If the root of a directory tree matches, no processing is done on the entire tree. This option can be useful for excluding version control repositories, e.g. exclude => qr/.svn/ Optional. If you do not specify a regexp, you get all the files that match the glob; in addition, new will set up a regexp for you, to provide a capture for each wildcard used in the path. If you do provide a match parameter, this will be used instead, and will filter the results. Supply an arrayref with a list of derived filenames, which will be constructed for each matching file. This causes next to return an arrayref instead of a scalar. If given a true value indicates that symbolic links are to be followed. Otherwise, the symbolic link target itself is presented, but the ellipsis will not traverse the link. This module detects a looping symlink that points to a directory higher up, and will only present the tree once. This can take one of the following values: normal, breadth-first, inside-out. The default option is normal. This controls how File::Wildcard handles the ellipsis. The default is a normal depth first search, presenting the name of each containing directory before the contents. The inside-out order presents the contents of directories first before the directory, which is useful when you want to remove files and directories (all O/S require directories to be empty before rmdir will work). See t/03_absolute.t as this uses inside-out order to tidy up after the test. Breadth-first is rarely needed (but I do have an application for it). Here, the whole directory contents is presented before traversing any subdirectories. Consider the following tree:
a/ a/bar/ a/bar/drink a/foo/ a/foo/lish
breadth-first will give the following order: qw(a/ a/bar/ a/foo/ a/bar/drink a/foo/lish). normal gives the order in which the files are listed. inside-out gives the following: qw(a/bar/drink a/bar/ a/foo/lish a/foo/ a/). By default, globbing returns the list of files in the order in which they are returned by the dirhandle (internally). If you specify sort => 1, the files are sorted into \s-1ASCII\s0 sequence (case insensitively if we are operating that way). If you specify a \s-1CODEREF\s0, this will be used as a comparison routine. Note that this takes its operands in @_, not in $a and $b. You can enable a trace of the internal states of File::Wildcard by setting debug to a true value. Set debug_output to an open filehandle to get the trace in a file. If you are submitting bug reports for File::Wildcard, attaching debug trace files would be very useful. debug_output defaults to \s-1STDERR\s0.
my $foo_re = $foo->match; $foo->match('bar/core');
This is a get and set method that gives access to the match regexp that the File::Wildcard object is using. It is possible to change the regex on the fly in the middle of a search (though I don't know why anyone would want to do this).
$foo->append(path => '/home/me///*.tmp');
appends a path to an object's todo list. This will be globbed after the object has finished processing the existing wildcards.
$srcfnd->prepend(path => $include_file);
This is similar to append, but prepends the path to the todo list. In other words, the current wildcard operation is interrupted to serve the new path, then the previous wildcard operation is resumed when this is exhausted.
while (my $core = $foo->next) { unlink $core; } my ($src,$obj,$hdr) = @{$srcfnd->next};
The \*(C`next\*(C' method is an iterator, which returns successive files. Returns matching files if there was no derive option passed to new. If there was a derive option, returns an arrayref containing the matching filespec and all derived filespecs. The derived filespecs do not have to exist.
Note that \*(C`next\*(C' maintains an internal cursor, which retains context and state information. Beware if the contents of directories are changing while you are iterating with next; you may get unpredictable results. If you are intending to change the contents of the directories you are scanning (with unlink or rename), you are better off deferring this operation until you have processed the whole tree. For the pending delete or rename operations, you could always use another File::Wildcard object - see the spike example below:
my @cores = $foo->all;
\*(C`all\*(C' returns an array of matching files, in the simple case. Returns an array of arrays if you are constructing new filenames, like the $srcfnd example.
Beware of the performance and memory implications of using \*(C`all\*(C'. The method will not return until it has read the entire directory tree. Use of the \*(C`all\*(C' method is not recommended for traversing large directory trees and whole file systems. Consider coding the traversal using the iterator \*(C`next\*(C' instead.
\*(C`reset\*(C' causes the wildcard context to be set to re-read the first filename again. Note that this will cause directory contents to be re-read.
Note also that this will cause the path to revert to the original path specified to new. Any additional paths appended or prepended will be forgotten.
Release all directory handles associated with the File::Wildcard object. An object that has been closed will be garbage collected once it goes out of scope. Wildcards that have been exhausted are automatically closed, (i.e. \*(C`all\*(C' was used, or c<next> returned undef).
Subsequent calls to \*(C`next\*(C' will return undef. It is possible to call \*(C`reset\*(C' after \*(C`close\*(C' on the same File::Wildcard object, which will cause it to be reopened.
The spike my $todo = File::Wildcard->new;
...
$todo->append(path => $file);
...
while (my $file = $todo->next) { ... } You can use an empty wildcard to store a list of filenames for later processing. The order in which they will be seen depends on whether append or prepend is used.
Shell style globbing my $wc_args = File::Wildcard->new;
$wc_args->append(path => $_) for @ARGV;
while ($wc_args->next) { ... } On Unix, file wildcards on the command line are globbed by the shell before perl sees them, unless the wildcards are escaped or quoted. This is not true of other operating systems. MS-DOS does no globbing at all for example. File::Wildcard gives you the bonus of elliptic globbing with '///'.
This module takes \s-1POSIX\s0 filenames, which use forward slash '/' as a path separator. All operating systems that run Perl can manage this type of path. The module is not designed to work with native file specs. If you want to write code that is portable, convert native filespecs to the \s-1POSIX\s0 form. There is of course no difference on Unix platforms.
Please report bugs to http://rt.cpan.org
Ivor Williams ivorw-file-wildcard010 at xemaps.com
This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
The full text of the license can be found in the \s-1LICENSE\s0 file included with this module.
glob\|(3), File::Find, File::Find::Rule.