NAME

B::Utils - Helper functions for op tree manipulation

SYNOPSIS

  use B::Utils;

DESCRIPTION

These functions make it easier to manipulate the op tree.

FUNCTIONS

all_starts

all_roots

Returns a hash of all of the starting ops or root ops of optrees, keyed to subroutine name; the optree for main program is simply keyed to __MAIN__.

Note: Certain ``dangerous'' stashes are not scanned for subroutines: the list of such stashes can be found in @B::Utils::bad_stashes. Feel free to examine and/or modify this to suit your needs. The intention is that a simple program which uses no modules other than B and B::Utils would show no addition symbols.

This does not return the details of ops in anonymous subroutines compiled at compile time. For instance, given

    $a = sub { ... };

the subroutine will not appear in the hash. This is just as well, since they're anonymous... If you want to get at them, use...

anon_subs()

This returns an array of hash references. Each element has the keys ``start'' and ``root''. These are the starting and root ops of all of the anonymous subroutines in the program.

$op->oldname

Returns the name of the op, even if it is currently optimized to null. This helps you understand the stucture of the op tree.

$op->kids

Returns an array of all this op's non-null children, in order.

$op->parent

Returns the parent node in the op tree, if possible. Currently ``possible'' means ``if the tree has already been optimized''; that is, if we're during a CHECK block. (and hence, if we have valid next pointers.)

In the future, it may be possible to search for the parent before we have the next pointers in place, but it'll take me a while to figure out how to do that.

$op->previous

Like $op->next, but not quite.

walkoptree_simple($op, \&callback, [$data])

The B module provides various functions to walk the op tree, but they're all rather difficult to use, requiring you to inject methods into the B::OP class. This is a very simple op tree walker with more expected semantics.

All the walk functions set B::Utils::file and B::Utils::line to the appropriate values of file and line number in the program being examined.

walkoptree_filtered($op, \&filter, \&callback, [$data])

This is much the same as walkoptree_simple, but will only call the callback if the filter returns true. The filter is passed the op in question as a parameter; the opgrep function is fantastic for building your own filters.

walkallops_simple(\&callback, [$data])

This combines walkoptree_simple with all_roots and anon_subs to examine every op in the program. $B::Utils::sub is set to the subroutine name if you're in a subroutine, __MAIN__ if you're in the main program and __ANON__ if you're in an anonymous subroutine.

walkallops_filtered(\&filter, \&callback, [$data])

Same as above, but filtered.

carp(@args)

croak(@args)

Warn and die, respectively, from the perspective of the position of the op in the program. Sounds complicated, but it's exactly the kind of error reporting you expect when you're grovelling through an op tree.

opgrep(\%conditions, @ops)

Returns the ops which meet the given conditions. The conditions should be specified like this:

    @barewords = opgrep(
                        { name => "const", private => OPpCONST_BARE },
                        @ops
                       );

You can specify alternation by giving an arrayref of values:

    @svs = opgrep ( { name => ["padsv", "gvsv"] }, @ops)

And you can specify inversion by making the first element of the arrayref a ``!''. (Hint: if you want to say ``anything'', say ``not nothing'': ["!"])

You may also specify the conditions to be matched in nearby ops.

    walkallops_filtered(
        sub { opgrep( {name => "exec", 
                       next => {
                                 name    => "nextstate",
                                 sibling => { name => [qw(! exit warn die)] }
                               }
                      }, @_)},
        sub { 
              carp("Statement unlikely to be reached"); 
              carp("\t(Maybe you meant system() when you said exec()?)\n");
        }
    )

Get that?

Here are the things that can be tested:

        name targ type seq flags private pmflags pmpermflags
        first other last sibling next pmreplroot pmreplstart pmnext

EXPORT

None by default.

AUTHOR

Simon Cozens, simon@cpan.org

TODO

I need to add more Fun Things, and possibly clean up some parts where the (previous/parent) algorithm has catastrophic cases, but it's more important to get this out right now than get it right.