This is Info file perl.info, produced by Makeinfo version 1.68 from the
input file bigperl.texi.
settitle perl
File: perl.info, Node: perltie, Next: perlbot, Prev: perlobj, Up: Top
how to hide an object class in a simple variable
************************************************
NAME
====
perltie - how to hide an object class in a simple variable
SYNOPSIS
========
tie VARIABLE, CLASSNAME, LIST
$object = tied VARIABLE
untie VARIABLE
DESCRIPTION
===========
Prior to release 5.0 of Perl, a programmer could use dbmopen() to
connect an on-disk database in the standard Unix dbm(3x) format magically
to a %HASH in their program. However, their Perl was either built with
one particular dbm library or another, but not both, and you couldn't
extend this mechanism to other packages or types of variables.
Now you can.
The tie() function binds a variable to a class (package) that will
provide the implementation for access methods for that variable. Once
this magic has been performed, accessing a tied variable automatically
triggers method calls in the proper class. The complexity of the class is
hidden behind magic methods calls. The method names are in ALL CAPS,
which is a convention that Perl uses to indicate that they're called
implicitly rather than explicitly-just like the BEGIN() and END()
functions.
In the tie() call, VARIABLE is the name of the variable to be
enchanted. `CLASSNAME' is the name of a class implementing objects of the
correct type. Any additional arguments in the LIST are passed to the
appropriate constructor method for that class-meaning TIESCALAR(),
TIEARRAY(), TIEHASH(), or TIEHANDLE(). (Typically these are arguments
such as might be passed to the dbminit() function of C.) The object
returned by the "new" method is also returned by the tie() function, which
would be useful if you wanted to access other methods in `CLASSNAME'. (You
don't actually have to return a reference to a right "type" (e.g., HASH or
`CLASSNAME') so long as it's a properly blessed object.) You can also
retrieve a reference to the underlying object using the tied() function.
Unlike dbmopen(), the tie() function will not use or require a module
for you-you need to do that explicitly yourself.
Tying Scalars
-------------
A class implementing a tied scalar should define the following methods:
TIESCALAR, FETCH, STORE, and possibly DESTROY.
Let's look at each in turn, using as an example a tie class for scalars
that allows the user to do something like:
tie $his_speed, 'Nice', getppid();
tie $my_speed, 'Nice', $$;
And now whenever either of those variables is accessed, its current
system priority is retrieved and returned. If those variables are set,
then the process's priority is changed!
We'll use Jarkko Hietaniemi <`jhi@iki.fi'>'s BSD::Resource class (not
included) to access the PRIO_PROCESS, PRIO_MIN, and PRIO_MAX constants
from your system, as well as the getpriority() and setpriority() system
calls. Here's the preamble of the class.
package Nice;
use Carp;
use BSD::Resource;
use strict;
$Nice::DEBUG = 0 unless defined $Nice::DEBUG;
TIESCALAR classname, LIST
This is the constructor for the class. That means it is expected to
return a blessed reference to a new scalar (probably anonymous) that
it's creating. For example:
sub TIESCALAR {
my $class = shift;
my $pid = shift || $$; # 0 means me
if ($pid !~ /^\d+$/) {
carp "Nice::Tie::Scalar got non-numeric pid $pid" if $^W;
return undef;
}
unless (kill 0, $pid) { # EPERM or ERSCH, no doubt
carp "Nice::Tie::Scalar got bad pid $pid: $!" if $^W;
return undef;
}
return bless \$pid, $class;
}
This tie class has chosen to return an error rather than raising an
exception if its constructor should fail. While this is how
dbmopen() works, other classes may well not wish to be so forgiving.
It checks the global variable $^W to see whether to emit a bit of
noise anyway.
FETCH this
This method will be triggered every time the tied variable is accessed
(read). It takes no arguments beyond its self reference, which is the
object representing the scalar we're dealing with. Because in this
case we're using just a SCALAR ref for the tied scalar object, a
simple $$self allows the method to get at the real value stored
there. In our example below, that real value is the process ID to
which we've tied our variable.
sub FETCH {
my $self = shift;
confess "wrong type" unless ref $self;
croak "usage error" if @_;
my $nicety;
local($!) = 0;
$nicety = getpriority(PRIO_PROCESS, $$self);
if ($!) { croak "getpriority failed: $!" }
return $nicety;
}
This time we've decided to blow up (raise an exception) if the renice
fails-there's no place for us to return an error otherwise, and it's
probably the right thing to do.
STORE this, value
This method will be triggered every time the tied variable is set
(assigned). Beyond its self reference, it also expects one (and only
one) argument-the new value the user is trying to assign.
sub STORE {
my $self = shift;
confess "wrong type" unless ref $self;
my $new_nicety = shift;
croak "usage error" if @_;
if ($new_nicety < PRIO_MIN) {
carp sprintf
"WARNING: priority %d less than minimum system priority %d",
$new_nicety, PRIO_MIN if $^W;
$new_nicety = PRIO_MIN;
}
if ($new_nicety > PRIO_MAX) {
carp sprintf
"WARNING: priority %d greater than maximum system priority %d",
$new_nicety, PRIO_MAX if $^W;
$new_nicety = PRIO_MAX;
}
unless (defined setpriority(PRIO_PROCESS, $$self, $new_nicety)) {
confess "setpriority failed: $!";
}
return $new_nicety;
}
DESTROY this
This method will be triggered when the tied variable needs to be
destructed. As with other object classes, such a method is seldom
necessary, because Perl deallocates its moribund object's memory for
you automatically-this isn't C++, you know. We'll use a DESTROY
method here for debugging purposes only.
sub DESTROY {
my $self = shift;
confess "wrong type" unless ref $self;
carp "[ Nice::DESTROY pid $$self ]" if $Nice::DEBUG;
}
That's about all there is to it. Actually, it's more than all there is
to it, because we've done a few nice things here for the sake of
completeness, robustness, and general aesthetics. Simpler TIESCALAR
classes are certainly possible.
Tying Arrays
------------
A class implementing a tied ordinary array should define the following
methods: TIEARRAY, FETCH, STORE, FETCHSIZE, STORESIZE and perhaps DESTROY.
FETCHSIZE and STORESIZE are used to provide `$#array' and equivalent
`scalar(@array)' access.
The methods POP, PUSH, SHIFT, UNSHIFT, SPLICE, DELETE, and EXISTS are
required if the perl operator with the corresponding (but lowercase) name
is to operate on the tied array. The Tie::Array class can be used as a
base class to implement the first five of these in terms of the basic
methods above. The default implementations of DELETE and EXISTS in
Tie::Array simply croak.
In addition EXTEND will be called when perl would have pre-extended
allocation in a real array.
This means that tied arrays are now complete. The example below needs
upgrading to illustrate this. (The documentation in Tie::Array is more
complete.)
For this discussion, we'll implement an array whose indices are fixed at
its creation. If you try to access anything beyond those bounds, you'll
take an exception. For example:
require Bounded_Array;
tie @ary, 'Bounded_Array', 2;
$| = 1;
for $i (0 .. 10) {
print "setting index $i: ";
$ary[$i] = 10 * $i;
$ary[$i] = 10 * $i;
print "value of elt $i now $ary[$i]\n";
}
The preamble code for the class is as follows:
package Bounded_Array;
use Carp;
use strict;
TIEARRAY classname, LIST
This is the constructor for the class. That means it is expected to
return a blessed reference through which the new array (probably an
anonymous ARRAY ref) will be accessed.
In our example, just to show you that you don't *really* have to
return an ARRAY reference, we'll choose a HASH reference to represent
our object. A HASH works out well as a generic record type: the
`{BOUND}' field will store the maximum bound allowed, and the
`{ARRAY}' field will hold the true ARRAY ref. If someone outside the
class tries to dereference the object returned (doubtless thinking it
an ARRAY ref), they'll blow up. This just goes to show you that you
should respect an object's privacy.
sub TIEARRAY {
my $class = shift;
my $bound = shift;
confess "usage: tie(\@ary, 'Bounded_Array', max_subscript)"
if @_ || $bound =~ /\D/;
return bless {
BOUND => $bound,
ARRAY => [],
}, $class;
}
FETCH this, index
This method will be triggered every time an individual element the
tied array is accessed (read). It takes one argument beyond its self
reference: the index whose value we're trying to fetch.
sub FETCH {
my($self,$idx) = @_;
if ($idx > $self->{BOUND}) {
confess "Array OOB: $idx > $self->{BOUND}";
}
return $self->{ARRAY}[$idx];
}
As you may have noticed, the name of the FETCH method (et al.) is the
same for all accesses, even though the constructors differ in names
(TIESCALAR vs TIEARRAY). While in theory you could have the same
class servicing several tied types, in practice this becomes
cumbersome, and it's easiest to keep them at simply one tie type per
class.
STORE this, index, value
This method will be triggered every time an element in the tied array
is set (written). It takes two arguments beyond its self reference:
the index at which we're trying to store something and the value
we're trying to put there. For example:
sub STORE {
my($self, $idx, $value) = @_;
print "[STORE $value at $idx]\n" if _debug;
if ($idx > $self->{BOUND} ) {
confess "Array OOB: $idx > $self->{BOUND}";
}
return $self->{ARRAY}[$idx] = $value;
}
DESTROY this
This method will be triggered when the tied variable needs to be
destructed. As with the scalar tie class, this is almost never
needed in a language that does its own garbage collection, so this
time we'll just leave it out.
The code we presented at the top of the tied array class accesses many
elements of the array, far more than we've set the bounds to. Therefore,
it will blow up once they try to access beyond the 2nd element of @ary, as
the following output demonstrates:
setting index 0: value of elt 0 now 0
setting index 1: value of elt 1 now 10
setting index 2: value of elt 2 now 20
setting index 3: Array OOB: 3 > 2 at Bounded_Array.pm line 39
Bounded_Array::FETCH called at testba line 12
Tying Hashes
------------
As the first Perl data type to be tied (see dbmopen()), hashes have the
most complete and useful tie() implementation. A class implementing a
tied hash should define the following methods: TIEHASH is the constructor.
FETCH and STORE access the key and value pairs. EXISTS reports whether a
key is present in the hash, and DELETE deletes one. CLEAR empties the
hash by deleting all the key and value pairs. FIRSTKEY and NEXTKEY
implement the keys() and each() functions to iterate over all the keys.
And DESTROY is called when the tied variable is garbage collected.
If this seems like a lot, then feel free to inherit from merely the
standard Tie::Hash module for most of your methods, redefining only the
interesting ones. See *Note Tie/Hash: (pm.info)Tie/Hash, for details.
Remember that Perl distinguishes between a key not existing in the hash,
and the key existing in the hash but having a corresponding value of
undef. The two possibilities can be tested with the exists() and
`defined()' functions.
Here's an example of a somewhat interesting tied hash class: it gives
you a hash representing a particular user's dot files. You index into the
hash with the name of the file (minus the dot) and you get back that dot
file's contents. For example:
use DotFiles;
tie %dot, 'DotFiles';
if ( $dot{profile} =~ /MANPATH/ ||
$dot{login} =~ /MANPATH/ ||
$dot{cshrc} =~ /MANPATH/ )
{
print "you seem to set your MANPATH\n";
}
Or here's another sample of using our tied class:
tie %him, 'DotFiles', 'daemon';
foreach $f ( keys %him ) {
printf "daemon dot file %s is size %d\n",
$f, length $him{$f};
}
In our tied hash DotFiles example, we use a regular hash for the object
containing several important fields, of which only the `{LIST}' field will
be what the user thinks of as the real hash.
USER
whose dot files this object represents
HOME
where those dot files live
CLOBBER
whether we should try to change or remove those dot files
LIST
the hash of dot file names and content mappings
Here's the start of `Dotfiles.pm':
package DotFiles;
use Carp;
sub whowasi { (caller(1))[3] . '()' }
my $DEBUG = 0;
sub debug { $DEBUG = @_ ? shift : 1 }
For our example, we want to be able to emit debugging info to help in
tracing during development. We keep also one convenience function around
internally to help print out warnings; whowasi() returns the function name
that calls it.
Here are the methods for the DotFiles tied hash.
TIEHASH classname, LIST
This is the constructor for the class. That means it is expected to
return a blessed reference through which the new object (probably but
not necessarily an anonymous hash) will be accessed.
Here's the constructor:
sub TIEHASH {
my $self = shift;
my $user = shift || $>;
my $dotdir = shift || '';
croak "usage: @{[&whowasi]} [USER [DOTDIR]]" if @_;
$user = getpwuid($user) if $user =~ /^\d+$/;
my $dir = (getpwnam($user))[7]
|| croak "@{[&whowasi]}: no user $user";
$dir .= "/$dotdir" if $dotdir;
my $node = {
USER => $user,
HOME => $dir,
LIST => {},
CLOBBER => 0,
};
opendir(DIR, $dir)
|| croak "@{[&whowasi]}: can't opendir $dir: $!";
foreach $dot ( grep /^\./ && -f "$dir/$_", readdir(DIR)) {
$dot =~ s/^\.//;
$node->{LIST}{$dot} = undef;
}
closedir DIR;
return bless $node, $self;
}
It's probably worth mentioning that if you're going to filetest the
return values out of a readdir, you'd better prepend the directory in
question. Otherwise, because we didn't chdir() there, it would have
been testing the wrong file.
FETCH this, key
This method will be triggered every time an element in the tied hash
is accessed (read). It takes one argument beyond its self reference:
the key whose value we're trying to fetch.
Here's the fetch for our DotFiles example.
sub FETCH {
carp &whowasi if $DEBUG;
my $self = shift;
my $dot = shift;
my $dir = $self->{HOME};
my $file = "$dir/.$dot";
unless (exists $self->{LIST}->{$dot} || -f $file) {
carp "@{[&whowasi]}: no $dot file" if $DEBUG;
return undef;
}
if (defined $self->{LIST}->{$dot}) {
return $self->{LIST}->{$dot};
} else {
return $self->{LIST}->{$dot} = `cat $dir/.$dot`;
}
}
It was easy to write by having it call the Unix cat(1) command, but it
would probably be more portable to open the file manually (and
somewhat more efficient). Of course, because dot files are a Unixy
concept, we're not that concerned.
STORE this, key, value
This method will be triggered every time an element in the tied hash
is set (written). It takes two arguments beyond its self reference:
the index at which we're trying to store something, and the value
we're trying to put there.
Here in our DotFiles example, we'll be careful not to let them try to
overwrite the file unless they've called the clobber() method on the
original object reference returned by tie().
sub STORE {
carp &whowasi if $DEBUG;
my $self = shift;
my $dot = shift;
my $value = shift;
my $file = $self->{HOME} . "/.$dot";
my $user = $self->{USER};
croak "@{[&whowasi]}: $file not clobberable"
unless $self->{CLOBBER};
open(F, "> $file") || croak "can't open $file: $!";
print F $value;
close(F);
}
If they wanted to clobber something, they might say:
$ob = tie %daemon_dots, 'daemon';
$ob->clobber(1);
$daemon_dots{signature} = "A true daemon\n";
Another way to lay hands on a reference to the underlying object is to
use the tied() function, so they might alternately have set clobber
using:
tie %daemon_dots, 'daemon';
tied(%daemon_dots)->clobber(1);
The clobber method is simply:
sub clobber {
my $self = shift;
$self->{CLOBBER} = @_ ? shift : 1;
}
DELETE this, key
This method is triggered when we remove an element from the hash,
typically by using the delete() function. Again, we'll be careful to
check whether they really want to clobber files.
sub DELETE {
carp &whowasi if $DEBUG;
my $self = shift;
my $dot = shift;
my $file = $self->{HOME} . "/.$dot";
croak "@{[&whowasi]}: won't remove file $file"
unless $self->{CLOBBER};
delete $self->{LIST}->{$dot};
my $success = unlink($file);
carp "@{[&whowasi]}: can't unlink $file: $!" unless $success;
$success;
}
The value returned by DELETE becomes the return value of the call to
delete(). If you want to emulate the normal behavior of delete(),
you should return whatever FETCH would have returned for this key.
In this example, we have chosen instead to return a value which tells
the caller whether the file was successfully deleted.
CLEAR this
This method is triggered when the whole hash is to be cleared,
usually by assigning the empty list to it.
In our example, that would remove all the user's dot files! It's
such a dangerous thing that they'll have to set CLOBBER to something
higher than 1 to make it happen.
sub CLEAR {
carp &whowasi if $DEBUG;
my $self = shift;
croak "@{[&whowasi]}: won't remove all dot files for $self->{USER}"
unless $self->{CLOBBER} > 1;
my $dot;
foreach $dot ( keys %{$self->{LIST}}) {
$self->DELETE($dot);
}
}
EXISTS this, key
This method is triggered when the user uses the exists() function on
a particular hash. In our example, we'll look at the `{LIST}' hash
element for this:
sub EXISTS {
carp &whowasi if $DEBUG;
my $self = shift;
my $dot = shift;
return exists $self->{LIST}->{$dot};
}
FIRSTKEY this
This method will be triggered when the user is going to iterate
through the hash, such as via a keys() or each() call.
sub FIRSTKEY {
carp &whowasi if $DEBUG;
my $self = shift;
my $a = keys %{$self->{LIST}}; # reset each() iterator
each %{$self->{LIST}}
}
NEXTKEY this, lastkey
This method gets triggered during a keys() or each() iteration. It
has a second argument which is the last key that had been accessed.
This is useful if you're carrying about ordering or calling the
iterator from more than one sequence, or not really storing things in
a hash anywhere.
For our example, we're using a real hash so we'll do just the simple
thing, but we'll have to go through the LIST field indirectly.
sub NEXTKEY {
carp &whowasi if $DEBUG;
my $self = shift;
return each %{ $self->{LIST} }
}
DESTROY this
This method is triggered when a tied hash is about to go out of
scope. You don't really need it unless you're trying to add debugging
or have auxiliary state to clean up. Here's a very simple function:
sub DESTROY {
carp &whowasi if $DEBUG;
}
Note that functions such as keys() and values() may return huge lists
when used on large objects, like DBM files. You may prefer to use the
each() function to iterate over such. Example:
# print out history file offsets
use NDBM_File;
tie(%HIST, 'NDBM_File', '/usr/lib/news/history', 1, 0);
while (($key,$val) = each %HIST) {
print $key, ' = ', unpack('L',$val), "\n";
}
untie(%HIST);
Tying FileHandles
-----------------
This is partially implemented now.
A class implementing a tied filehandle should define the following
methods: TIEHANDLE, at least one of PRINT, PRINTF, WRITE, READLINE, GETC,
READ, and possibly CLOSE and DESTROY. The class can also provide: BINMODE,
OPEN, EOF, FILENO, SEEK, TELL - if the corresponding perl operators are
used on the handle.
It is especially useful when perl is embedded in some other program,
where output to STDOUT and STDERR may have to be redirected in some
special way. See nvi and the Apache module for examples.
In our example we're going to create a shouting handle.
package Shout;
TIEHANDLE classname, LIST
This is the constructor for the class. That means it is expected to
return a blessed reference of some sort. The reference can be used to
hold some internal information.
sub TIEHANDLE { print "\n"; my $i; bless \$i, shift }
WRITE this, LIST
This method will be called when the handle is written to via the
syswrite function.
sub WRITE {
$r = shift;
my($buf,$len,$offset) = @_;
print "WRITE called, \$buf=$buf, \$len=$len, \$offset=$offset";
}
PRINT this, LIST
This method will be triggered every time the tied handle is printed to
with the print() function. Beyond its self reference it also expects
the list that was passed to the print function.
sub PRINT { $r = shift; $$r++; print join($,,map(uc($_),@_)),$\ }
PRINTF this, LIST
This method will be triggered every time the tied handle is printed to
with the printf() function. Beyond its self reference it also
expects the format and list that was passed to the printf function.
sub PRINTF {
shift;
my $fmt = shift;
print sprintf($fmt, @_)."\n";
}
READ this, LIST
This method will be called when the handle is read from via the read
or sysread functions.
sub READ {
my $self = shift;
my $$bufref = \$_[0];
my(undef,$len,$offset) = @_;
print "READ called, \$buf=$bufref, \$len=$len, \$offset=$offset";
# add to $$bufref, set $len to number of characters read
$len;
}
READLINE this
This method will be called when the handle is read from via .
The method should return undef when there is no more data.
sub READLINE { $r = shift; "READLINE called $$r times\n"; }
GETC this
This method will be called when the getc function is called.
sub GETC { print "Don't GETC, Get Perl"; return "a"; }
CLOSE this
This method will be called when the handle is closed via the close
function.
sub CLOSE { print "CLOSE called.\n" }
DESTROY this
As with the other types of ties, this method will be called when the
tied handle is about to be destroyed. This is useful for debugging and
possibly cleaning up.
sub DESTROY { print "\n" }
Here's how to use our little example:
tie(*FOO,'Shout');
print FOO "hello\n";
$a = 4; $b = 6;
print FOO $a, " plus ", $b, " equals ", $a + $b, "\n";
print ;
The untie Gotcha
----------------
If you intend making use of the object returned from either tie() or
tied(), and if the tie's target class defines a destructor, there is a
subtle gotcha you must guard against.
As setup, consider this (admittedly rather contrived) example of a tie;
all it does is use a file to keep a log of the values assigned to a scalar.
package Remember;
use strict;
use warnings;
use IO::File;
sub TIESCALAR {
my $class = shift;
my $filename = shift;
my $handle = new IO::File "> $filename"
or die "Cannot open $filename: $!\n";
print $handle "The Start\n";
bless {FH => $handle, Value => 0}, $class;
}
sub FETCH {
my $self = shift;
return $self->{Value};
}
sub STORE {
my $self = shift;
my $value = shift;
my $handle = $self->{FH};
print $handle "$value\n";
$self->{Value} = $value;
}
sub DESTROY {
my $self = shift;
my $handle = $self->{FH};
print $handle "The End\n";
close $handle;
}
1;
Here is an example that makes use of this tie:
use strict;
use Remember;
my $fred;
tie $fred, 'Remember', 'myfile.txt';
$fred = 1;
$fred = 4;
$fred = 5;
untie $fred;
system "cat myfile.txt";
This is the output when it is executed:
The Start
1
4
5
The End
So far so good. Those of you who have been paying attention will have
spotted that the tied object hasn't been used so far. So lets add an
extra method to the Remember class to allow comments to be included in the
file - say, something like this:
sub comment {
my $self = shift;
my $text = shift;
my $handle = $self->{FH};
print $handle $text, "\n";
}
And here is the previous example modified to use the comment method
(which requires the tied object):
use strict;
use Remember;
my ($fred, $x);
$x = tie $fred, 'Remember', 'myfile.txt';
$fred = 1;
$fred = 4;
comment $x "changing...";
$fred = 5;
untie $fred;
system "cat myfile.txt";
When this code is executed there is no output. Here's why:
When a variable is tied, it is associated with the object which is the
return value of the TIESCALAR, TIEARRAY, or TIEHASH function. This object
normally has only one reference, namely, the implicit reference from the
tied variable. When untie() is called, that reference is destroyed.
Then, as in the first example above, the object's destructor (DESTROY) is
called, which is normal for objects that have no more valid references;
and thus the file is closed.
In the second example, however, we have stored another reference to the
tied object in $x. That means that when untie() gets called there will
still be a valid reference to the object in existence, so the destructor
is not called at that time, and thus the file is not closed. The reason
there is no output is because the file buffers have not been flushed to
disk.
Now that you know what the problem is, what can you do to avoid it?
Well, the good old -w flag will spot any instances where you call untie()
and there are still valid references to the tied object. If the second
script above this near the top `use warnings 'untie'' or was run with the
-w flag, Perl prints this warning message:
untie attempted while 1 inner references still exist
To get the script to work properly and silence the warning make sure
there are no valid references to the tied object before untie() is called:
undef $x;
untie $fred;
SEE ALSO
========
See *Note DB_File: (pm.info)DB_File, or *Note Config: (pm.info)Config,
for some interesting tie() implementations.
BUGS
====
Tied arrays are *incomplete*. They are also distinctly lacking
something for the `$#ARRAY' access (which is hard, as it's an lvalue), as
well as the other obvious array functions, like push(), pop(), shift(),
unshift(), and splice().
You cannot easily tie a multilevel data structure (such as a hash of
hashes) to a dbm file. The first problem is that all but GDBM and
Berkeley DB have size limitations, but beyond that, you also have problems
with how references are to be represented on disk. One experimental
module that does attempt to address this need partially is the MLDBM
module. Check your nearest CPAN site as described in *Note Perlmodlib:
perlmodlib, for source code to MLDBM.
AUTHOR
======
Tom Christiansen
TIEHANDLE by Sven Verdoolaege <`skimo@dns.ufsia.ac.be'> and Doug
MacEachern <`dougm@osf.org'>
File: perl.info, Node: perltodo, Next: perlhack, Prev: perlintern, Up: Top
Perl TO-DO List
***************
NAME
====
perltodo - Perl TO-DO List
DESCRIPTION
===========
This is a list of wishes for Perl. It is maintained by Nathan
Torkington for the Perl porters. Send updates to
*perl5-porters@perl.org*. If you want to work on any of these projects,
be sure to check the perl5-porters archives for past ideas, flames, and
propaganda. This will save you time and also prevent you from
implementing something that Larry has already vetoed. One set of archives
may be found at:
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/
Infrastructure
==============
Mailing list archives
---------------------
Chaim suggests contacting egroup and asking them to archive the other
perl.org mailing lists. Probably not advocacy, but definitely
perl6-porters, etc.
Bug tracking system
-------------------
Richard Foley *richard@perl.org* is writing one. We looked at several,
like gnats and the Debian system, but at the time we investigated them,
none met our needs. Since then, Jitterbug has matured, and may be worth
reinvestigation.
The system we've developed is the recipient of perlbug mail, and any
followups it generates from perl5-porters. New bugs are entered into a
mysql database, and sent on to perl5-porters with the subject line
rewritten to include a "ticket number" (unique ID for the new bug). If
the incoming message already had a ticket number in the subject line, then
the message is logged against that bug. There is a separate email
interface (not forwarding to p5p) that permits porters to claim,
categorize, and close tickets.
There is also a web interface to the system at http://bugs.perl.org.
The current delay in implementation is caused by perl.org lockups. One
suspect is the mail handling system, possibly going into loops.
We still desperately need a bugmaster, someone who will look at every
new "bug" and kill those that we already know about, those that are not
bugs at all, etc.
Regression Tests
----------------
The test suite for Perl serves two needs: ensuring features work, and
ensuring old bugs have not been reintroduced. Both need work.
Brent LaVelle (lavelle@metronet.com) has stepped forward to work on
performance tests and improving the size of the test suite.
Coverage
Do the tests that come with Perl exercise every line (or every block,
or ...) of the Perl interpreter, and if not then how can we make them
do so?
Regression
No bug fixes should be made without a corresponding testsuite
addition. This needs a dedicated enforcer, as the current pumpking
is either too lazy or too stupid or both and lets enforcement wander
all over the map. :-)
__DIE__
Tests that fail need to be of a form that can be readily mailed to
perlbug and diagnosed with minimal back-and-forth's to determine
which test failed, due to what cause, etc.
suidperl
We need regression/sanity tests for suidperl
The 25% slowdown from perl4 to perl5
This value may or may not be accurate, but it certainly is
eye-catching. For some things perl5 is faster than perl4, but often
the reliability and extensability have come at a cost of speed. The
benchmark suite that Gisle released earlier has been hailed as both a
fantastic solution and as a source of entirely meaningless figures.
Do we need to test "real applications"? Can you do so? Anyone have
machines to dedicate to the task? Identify the things that have grown
slower, and see if there's a way to make them faster.
Configure
=========
Andy Dougherty maintain(ed|s) a list of "todo" items for the configure
that comes with Perl. See Porting/pumpkin.pod in the latest source
release.
Install HTML
------------
Have "make install" give you the option to install HTML as well. This
would be part of Configure. Andy Wardley (certified Perl studmuffin) will
look into the current problems of HTML installation-is 'installhtml'
preventing this from happening cleanly, or is pod2html the problem? If
the latter, Brad Appleton's pod work may fix the problem for free.
Perl Language
=============
our ($var)
----------
Declare global variables (lexically or otherwise).
64-bit Perl
-----------
Verify complete 64 bit support so that the value of sysseek, or -s, or
stat(), or tell can fit into a perl number without losing precision. Work
with the perl-64bit mailing list on perl.org.
Prototypes
----------
Named prototypes
Add proper named prototypes that actually work usefully.
Indirect objects
Fix prototype bug that forgets indirect objects.
Method calls
Prototypes for method calls.
Context
Return context prototype declarations.
Scoped subs
lexically-scoped subs, e.g. my sub
Perl Internals
==============
magic_setisa
------------
magic_setisa should be made to update %FIELDS [???]
Garbage Collection
------------------
There was talk of a mark-and-sweep garbage collector at TPC2, but the
(to users) unpredictable nature of its behaviour put some off. Sarathy, I
believe, did the work. Here's what he has to say:
Yeah, I hope to implement it someday too. The points that were raised
in TPC2 were all to do with calling DESTROY() methods, but I think we can
accomodate that by extending bless() to stash extra information for
objects so we track their lifetime accurately for those that want their
DESTROY() to be predictable (this will be a speed hit, naturally, and will
therefore be optional, naturally. :)
[N.B. Don't even ask me about this now! When I have the time to write
a cogent summary, I'll post it.]
Reliable signals
----------------
Sarathy and Dan Sugalski are working on this. Chip posted a patch
earlier, but it was not accepted into 5.005. The issue is tricky, because
it has the potential to greatly slow down the core.
There are at least three things to consider:
Alternate runops() for signal despatch
Sarathy and Dan are discussed this on perl5-porters.
Figure out how to die() in delayed sighandler
Add tests for Thread::Signal
Automatic tests against CPAN
Is there some way to automatically build all/most of CPAN with the
new Perl and check that the modules there pass all the tests?
Interpolated regex performance bugs
-----------------------------------
while (<>) {
$found = 0;
foreach $pat (@patterns) {
$found++ if /$pat/o;
}
print if $found;
}
The qr// syntax added in 5.005 has solved this problem, but it needs
more thorough documentation.
Memory leaks from failed eval/regcomp
-------------------------------------
The only known memory leaks in Perl are in failed code or regexp
compilation. Fix this. Hugo Van Der Sanden will attempt this but won't
have tuits until January 1999.
Make XS easier to use
---------------------
There was interest in SWIG from porters, but nothing has happened
lately.
Make embedded Perl easier to use
--------------------------------
This is probably difficult for the same reasons that "XS For Dummies"
will be difficult.
Namespace cleanup
-----------------
CPP-space: restrict CPP symbols exported from headers
header-space: move into CORE/perl/
API-space: begin list of things that constitute public api
env-space: Configure should use PERL_CONFIG instead of CONFIG etc.
MULTIPLICITY
------------
Complete work on safe recursive interpreters `Perl->new()'. Sarathy
says that a reference implementation exists.
MacPerl
-------
Chris Nandor and Matthias Neeracher are working on better integrating
MacPerl into the Perl distribution.
Documentation
=============
There's a lot of documentation that comes with Perl. The quantity of
documentation makes it difficult for users to know which section of which
manpage to read in order to solve their problem. Tom Christiansen has
done much of the documentation work in the past.
A clear division into tutorial and reference
--------------------------------------------
Some manpages (e.g., perltoot and perlreftut) clearly set out to
educate the reader about a subject. Other manpages (e.g., perlsub) are
references for which there is no tutorial, or are references with a slight
tutorial bent. If things are either tutorial or reference, then the
reader knows which manpage to read to learn about a subject, and which
manpage to read to learn all about an aspect of that subject. Part of the
solution to this is:
Remove the artificial distinction between operators and functions
-----------------------------------------------------------------
History shows us that users, and often porters, aren't clear on the
operator-function distinction. The present split in reference material
between perlfunc and perlop hinders user navigation. Given that perlfunc
is by far the larger of the two, move operator reference into perlfunc.
More tutorials
--------------
More documents of a tutorial nature could help. Here are some
candidates:
Regular expressions
Robin Berjon (r.berjon@ltconsulting.net) has volunteered.
I/O
Mark-Jason Dominus (mjd@plover.com) has an outline for perliotut.
pack/unpack
This is badly needed. There has been some discussion on the subject
on perl5-porters.
Debugging
Ronald Kimball (rjk@linguist.dartmouth.edu) has volunteered.
Include a search tool
---------------------
perldoc should be able to 'grep' fulltext indices of installed POD
files. This would let people say:
perldoc -find printing numbers with commas
and get back the perlfaq entry on 'commify'.
This solution, however, requires documentation to contain the keywords
the user is searching for. Even when the users know what they're looking
for, often they can't spell it.
Include a locate tool
---------------------
perldoc should be able to help people find the manpages on a particular
high-level subject:
perldoc -find web
would tell them manpages, web pages, and books with material on web
programming. Similarly `perldoc -find databases', `perldoc -find
references' and so on.
We need something in the vicinity of:
% perl -help random stuff
No documentation for perl function `random stuff' found
The following entry in perlfunc.pod matches /random/a:
=item rand EXPR
=item rand
Returns a random fractional number greater than or equal to C<0> and less
than the value of EXPR. (EXPR should be positive.) If EXPR is
omitted, the value C<1> is used. Automatically calls C unless
C has already been called. See also C.
(Note: If your rand function consistently returns numbers that are too
large or too small, then your version of Perl was probably compiled
with the wrong number of RANDBITS.)
The following pod pages seem to have /stuff/a:
perlfunc.pod (7 hits)
perlfaq7.pod (6 hits)
perlmod.pod (4 hits)
perlsyn.pod (3 hits)
perlfaq8.pod (2 hits)
perlipc.pod (2 hits)
perl5004delta.pod (1 hit)
perl5005delta.pod (1 hit)
perlcall.pod (1 hit)
perldelta.pod (1 hit)
perlfaq3.pod (1 hit)
perlfaq5.pod (1 hit)
perlhist.pod (1 hit)
perlref.pod (1 hit)
perltoc.pod (1 hit)
perltrap.pod (1 hit)
Proceed to open perlfunc.pod? [y] n
Do you want to speak perl interactively? [y] n
Should I dial 911? [y] n
Do you need psychiatric help? [y] y
Hi, what bothers you today?
A Python programmer in the next cubby is driving me nuts!
Hmm, thats fixable. Just [rest censored]
Separate function manpages by default
-------------------------------------
Perl should install 'manpages' for every function/operator into the 3pl
or 3p manual section. By default. The splitman program in the Perl
source distribution does the work of turning big perlfunc into little 3p
pages.
Users can't find the manpages
-----------------------------
Make perldoc tell users what they need to add to their .login or .cshrc
to set their MANPATH correctly.
Install ALL Documentation
-------------------------
Make the standard documentation kit include the VMS, OS/2, Win32,
Threads, etc information. installperl and pod/Makefile should know enough
to copy README.foo to perlfoo.pod before building everything, when
appropriate.
Outstanding issues to be documented
-----------------------------------
Tom has a list of 5.005_5* features or changes that require
documentation.
Create one document that coherently explains the delta between the last
camel release and the current release. perldelta was supposed to be that,
but no longer. The things in perldelta never seemed to get placed in the
right places in the real manpages, either. This needs work.
Adapt www.linuxhq.com for Perl
------------------------------
This should help glorify documentation and get more people involved in
perl development.
Replace man with a perl program
-------------------------------
Can we reimplement man in Perl? Tom has a start. I believe some of
the Linux systems distribute a manalike. Alternatively, build on perldoc
to remove the unfeatures like "is slow" and "has no apropos".
Unicode tutorial
----------------
We could use more work on helping people understand Perl's new Unicode
support that Larry has created.
Modules
=======
Update the POSIX extension to conform with the POSIX 1003.1 Edition 2
---------------------------------------------------------------------
The current state of the POSIX extension is as of Edition 1, 1991,
whereas the Edition 2 came out in 1996. ISO/IEC 9945:1-1996(E), ANSI/IEEE
Std 1003.1, 1996 Edition. ISBN 1-55937-573-6. The updates were legion:
threads, IPC, and real time extensions.
Module versions
---------------
Automate the checking of versions in the standard distribution so it's
easy for a pumpking to check whether CPAN has a newer version that we
should be including?
New modules
-----------
Which modules should be added to the standard distribution? This ties
in with the SDK discussed on the perl-sdk list at perl.org.
Profiler
--------
Make the profiler (Devel::DProf) part of the standard release, and
document it well.
Tie Modules
-----------
VecArray
Implement array using vec(). Nathan Torkington has working code to
do this.
SubstrArray
Implement array using substr()
VirtualArray
Implement array using a file
ShiftSplice
Defines shift et al in terms of splice method
Procedural options
------------------
Support procedural interfaces for the common cases of Perl's
gratuitously OOO modules. Tom objects to "use IO::File" reading many
thousands of lines of code.
RPC
---
Write a module for transparent, portable remote procedure calls. (Not
core). This touches on the CORBA and ILU work.
y2k localtime/gmtime
--------------------
Write a module, Y2k::Catch, which overloads localtime and gmtime's
returned year value and catches "bad" attempts to use it.
Export File::Find variables
---------------------------
Make File::Find export $name etc manually, at least if asked to.
Ioctl
-----
Finish a proper Ioctl module.
Debugger attach/detach
----------------------
Permit a user to debug an already-running program.
Regular Expression debugger
---------------------------
Create a visual profiler/debugger tool that stepped you through the
execution of a regular expression point by point. Ilya has a module to
color-code and display regular expression parses and executions. There's
something at http://tkworld.org/ that might be a good start, it's a Tk/Tcl
RE wizard, that builds regexen of many flavours.
Alternative RE Syntax
---------------------
Make an alternative regular expression syntax that is accessed through
a module. For instance,
use RE;
$re = start_of_line()
->literal("1998/10/08")
->optional( whitespace() )
->literal("[")
->remember( many( or( "-", digit() ) ) );
if (/$re/) {
print "time is $1\n";
}
Newbies to regular expressions typically only use a subset of the full
language. Perhaps you wouldn't have to implement the full feature set.
Bundled modules
---------------
Nicholas Clark (nick@flirble.org) had a patch for storing modules in
zipped format. This needs exploring and concluding.
Expect
------
Adopt IO::Tty, make it as portable as Don Libes' "expect" (can we link
against expect code?), and perfect a Perl version of expect. IO::Tty and
expect could then be distributed as part of the core distribution,
replacing Comm.pl and other hacks.
GUI::Native
-----------
A simple-to-use interface to native graphical abilities would be
welcomed. Oh, Perl's access Tk is nice enough, and reasonably portable,
but it's not particularly as fast as one would like. Simple access to the
mouse's cut buffer or mouse-presses shouldn't required loading a few
terabytes of Tk code.
Update semibroken auxiliary tools; h2ph, a2p, etc.
--------------------------------------------------
Kurt Starsinic is working on h2ph. mjd has fixed bugs in a2p in the
past. a2p apparently doesn't work on nawk and gawk extensions. Graham
Barr has an Include module that does h2ph work at runtime.
POD Converters
--------------
Brad's PodParser code needs to become part of the core, and the Pod::*
and pod2* programs rewritten to use this standard parser. Currently the
converters take different options, some behave in different fashions, and
some are more picky than others in terms of the POD files they accept.
pod2html
--------
A short-term fix: pod2html generates absolute HTML links. Make it
generate relative links.
Podchecker
----------
Something like lint for Pod would be good. Something that catches
common errors as well as gross ones. Brad Appleton is putting together
something as part of his PodParser work.
Tom's Wishes
============
Webperl
-------
Design a webperl environment that's as tightly integrated and as
easy-to-use as Perl's current command-line environment.
Mobile agents
-------------
More work on a safe and secure execution environment for mobile agents
would be neat; the Safe.pm module is a start, but there's a still a lot to
be done in that area. Adopt Penguin?
POSIX on non-POSIX
------------------
Standard programming constructs for non-POSIX systems would help a lot
of programmers stuck on primitive, legacy systems. For example, Microsoft
still hasn't made a usable POSIX interface on their clunky systems, which
means that standard operations such as alarm() and fork(), both critical
for sophisticated client-server programming, must both be kludged around.
I'm unsure whether Tom means to emulate alarm( )and fork(), or merely
to provide a document like perlport.pod to say which features are portable
and which are not.
Portable installations
----------------------
Figure out a portable semi-gelled installation, that is, one without
full paths. Larry has said that he's thinking about this. Ilya pointed
out that perllib_mangle() is good for this.
Win32 Stuff
===========
Rename new headers to be consistent with the rest
-------------------------------------------------
Sort out the spawnvp() mess
---------------------------
Work out DLL versioning
-----------------------
Style-check
-----------
Would be nice to have
=====================
`pack "(stuff)*"'
Contiguous bitfields in pack/unpack
lexperl
Bundled perl preprocessor
Use posix calls internally where possible
format BOTTOM
-i rename file only when successfully changed
All ARGV input should act like <>
report HANDLE [formats].
support in perlmain to rerun debugger
lvalue functions
Tuomas Lukka, on behalf of the PDL project, greatly desires this and
Ilya has a patch for it (probably against an older version of Perl).
Tuomas points out that what PDL really wants is lvalue methods, not
just subs.
Possible pragmas
================
'less'
------
(use less memory, CPU)
Optimizations
=============
constant function cache
-----------------------
foreach(reverse...)
-------------------
Cache eval tree
---------------
Unless lexical outer scope used (mark in &compiling?).
rcatmaybe
---------
Shrink opcode tables
--------------------
Via multiple implementations selected in peep.
Cache hash value
----------------
Not a win, according to Guido.
Optimize away @_ where possible
-------------------------------
Optimize sort by { $a <=> $b }
------------------------------
Greg Bacon added several more sort optimizations. These have made it
into 5.005_55, thanks to Hans Mulder.
Rewrite regexp parser for better integrated optimization
--------------------------------------------------------
The regexp parser was rewritten for 5.005. Ilya's the regexp guru.
Vague possibilities
===================
ref function in list context
This seems impossible to do without substantially breaking code.
make tr/// return histogram in list context?
Loop control on do{} et al
Explicit switch statements
Nobody has yet managed to come up with a switch syntax that would
allow for mixed hash, constant, regexp checks. Submit implementation
with syntax, please.
compile to real threaded code
structured types
Modifiable $1 et al
The intent is for this to be a means of editing the matched portions
of the target string.
To Do Or Not To Do
==================
These are things that have been discussed in the past and roundly
criticized for being of questionable value.
Making my() work on "package" variables
---------------------------------------
Being able to say my($Foo::Bar), something that sounds ludicrous and
the 5.6 pumpking has mocked.
"or" testing defined not truth
------------------------------
We tell people that `||' can be used to give a default value to a
variable:
$children = shift || 5; # default is 5 children
which is almost (but not):
$children = shift;
$children = 5 unless $children;
but if the first argument was given and is "0", then it will be
considered false by `||' and 5 used instead. Really we want an `||'-like
operator that behaves like:
$children = shift;
$children = 5 unless defined $children;
Namely, a `||' that tests defined-ness rather than truth. One was
discussed, and a patch submitted, but the objections were many. While
there were objections, many still feel the need. At least it was decided
that `??' is the best name for the operator.
"dynamic" lexicals
------------------
my $x;
sub foo {
local $x;
}
Localizing, as Tim Bunce points out, is a separate concept from whether
the variable is global or lexical. Chip Salzenberg had an implementation
once, but Larry thought it had potential to confuse.
"class"-based, rather than package-based "lexicals"
---------------------------------------------------
This is like what the Alias module provides, but the variables would be
lexicals reserved by perl at compile-time, which really are indices
pointing into the pseudo-hash object visible inside every method so
declared.
Threading
=========
Modules
-------
Which of the standard modules are thread-safe? Which CPAN modules?
How easy is it to fix those non-safe modules?
Testing
-------
Threading is still experimental. Every reproducible bug identifies
something else for us to fix. Find and submit more of these problems.
$AUTOLOAD
---------
exit/die
--------
Consistent semantics for exit/die in threads.
External threads
----------------
Better support for externally created threads.
Thread::Pool
------------
thread-safety
-------------
Spot-check globals like statcache and global GVs for thread-safety.
"*Part done*", says Sarathy.
Per-thread GVs
--------------
According to Sarathy, this would make @_ be the same in threaded and
non-threaded, as well as helping solve problems like filehandles (the same
filehandle currently cannot be used in two threads).
Compiler
========
Optimization
------------
The compiler's back-end code-generators for creating bytecode or
compilable C code could use optimization work.
Byteperl
--------
Figure out how and where byteperl will be built for the various
platforms.
Precompiled modules
-------------------
Save byte-compiled modules on disk.
Executables
-----------
Auto-produce executable.
Typed lexicals
--------------
Typed lexicals should affect B::CC::load_pad.
Win32
-----
Workarounds to help Win32 dynamic loading.
END blocks
----------
END blocks need saving in compiled output, now that CHECK blocks are
available.
_AUTOLOAD
---------
_AUTOLOAD prodding.
comppadlist
-----------
Fix comppadlist (names in comppad_name can have fake SvCUR from where
newASSIGNOP steals the field).
Cached compilation
------------------
Can we install modules as bytecode?
Recently Finished Tasks
=======================
Figure a way out of $^(capital letter)
--------------------------------------
Figure out a clean way to extend $^(capital letter) beyond the 26
alphabets. (${^WORD} maybe?)
Mark-Jason Dominus sent a patch which went into 5.005_56.
Filenames
---------
Keep filenames in the distribution and in the standard module set be
8.3 friendly where feasible. Good luck changing the standard modules,
though.
Foreign lines
-------------
Perl should be more generous in accepting foreign line terminations.
Mostly done in 5.005.
Namespace cleanup
-----------------
symbol-space: "pl_" prefix for all global vars
"Perl_" prefix for all functions
CPP-space: stop malloc()/free() pollution unless asked
ISA.pm
------
Rename and alter ISA.pm. *Done*. It is now base.pm.
gettimeofday
------------
See Time::HiRes.
autocroak?
----------
This is the Fatal.pm module, so any builtin that that does not return
success automatically die()s. If you're feeling brave, tie this in with
the unified exceptions scheme.