Re: RFC 112 (v3) Asignment within a regex

[Richard Proctor]

 On Fri, 29 Sep 2000 01:02:40 +0100, Hugo wrote:

 It also isn't clear what parts of the expression are interpolated at
 compile time; what should the following leave in %foo?
 
   %foo = ();
   $bar = "one";
   "twothree" =~ / (?$bar=two) (?$foo{$bar}=three) /x;

 It's not just that. You act as if this is assignment takes place
 whenever a submatch succeeds. So:

  "twofour" =~ /(?$bar=two)($foo=three)/;

 Will $bar be set to "two", and $foo undef? I think not. Assignment
 should be postponed to till the very end, when the match finally
 succeeds, as a whole.

In general all assignments should wait to the very end, and then assign
them all.  However before code callouts (?{...}) and enemies, the named
assignments that are currently defined should be made (localised) so that
the code can refer to them by name.  If the expression finally fails the
localised values would unroll.


 Therefore, I think that allowing just any l-value on the left of the "="
 sign, is not practical. Or is it?

I think any simple scalar value is reasonable.


 OTOH I would rather have that all submatches would be assigned to a
 hash, not to global or lexical variables. I have no clue about what
 syntax that would need.

That is in RFC 150, I think there is a case for both.

Richard

Re: RFC 112 (v3) Asignment within a regex

[Hugo]

In [EMAIL PROTECTED], "Richard Proctor" writes:
:In general all assignments should wait to the very end, and then assign
:them all. [...] If the expression finally fails the localised values
:would unroll.

Ah, I hadn't anticipated that - I had assumed you would get whatever
was the last value set. Please can you make sure this is clearly
explained in the next version of the RFC?

Hugo

RFC 112 (v3) Asignment within a regex

[Perl6 RFC Librarian]

This and other RFCs are available on the web at
  http://dev.perl.org/rfc/

=head1  TITLE

Asignment within a regex

=head1 VERSION

  Maintainer: Richard Proctor [EMAIL PROTECTED]
  Date: 16 Aug 2000
  Last Modified: 23 Sep 2000
  Mailing List: [EMAIL PROTECTED]
  Number: 112
  Version: 3
  Status: Developing

=head1 ABSTRACT

Provide a simple way of naming and picking out information from a regex
without having to count the brackets.

=head1 DESCRIPTION

If a regex is complex, counting the bracketed sub-expressions to find the
ones you wish to pick out can be messy.  It is also prone to maintainability
problems if and when you wish to add to the expression.  Using (?:) can be
used to surpress picking up brackets, it helps, but it still gets "complex".  
I would sometimes rather just pickout the bits I want within the regex itself.

Suggested syntax: (?$foo= ... ) would assign the string that is matched by
the patten ... to $foo when the patten matches.  These assignments would be
made left to right after the match has succeded but before processing a 
replacement or other results (or prior to a some (?{...}) or (??{...})
code).  There may be whitespace between the $foo and the "=".  

Potentially the $foo could be any scalar LHS, as in (?$foo{$bar}= ... )!,
likewise the '=' could be any asignment operator.

The camel and the docs include this example:

   if (/Time: (..):(..):(..)/) {
$hours = $1;
$minutes = $2;
$seconds = $3;
}

This then becomes:
 
  /Time: (?$hours=..):(?$minutes=..):(?$seconds=..)/

This is more maintainable than counting the brackets and easier to understand
for a complex regex.  And one does not have to worry about the scope of $1 etc.

=head2 Named Backrefs

The first versions of this RFC did not allow for backrefs.  I now think this
was a shortcoming.  It can be done with (??{quotemeta $foo}), but I find this
clumsy, a better way of using a named back ref might be (?\$foo).

=head2 Scoping

The question of scoping for these assignments has been raised, but I don't
currently have a feel for the "best" way to handle this.  Input welcome.

=head2 Brackets

Using this method for capturing wanted content, it might be desirable to
stop ordinary brackets capturing, and needing to use (?:...).  I therefore
suggest that as an enhancement to regexes that /b (bracket?) ordinary brackets
just group, without capture - in effect they all behave as (?:...).

=head1 CHANGES

V3 - added bit about backrefs, and brackets.

=head1 IMPLENTATION

Currently all $scalars in regexes are expanded before the main regex compiler
gets to analyse the syntax.  This problem also affects several other RFCs
(166 for example).  The expansion of variables in regexes needs for these
(and other RFCs) to be driven from within the regex compiler so that the
regex can expand as and where appropriate.  Changing this should not affect
any existing behaviour.

=head1 REFERENCES

I brought this up on p5p a couple of years ago, but it was lost in the noise...

RFC 166: Alternative lists and quoting of things

Perlstorm #0040