2/9/07, Abigail <abigail (AT) abigail (DOT) bewrote:
If a regular expression contains two named captures with the same
name, $+ {NAME} returns the leftmost *defined* capture. I'm not
sure how useful that is - I think I'd prefer the lefmost capture,
whether defined or not.
Consider the following code:
my $re = qr /(?<integer>\d+)(?:\.(?<fraction>\d+))?/;
if ("1.2 3.4" =~ /$re $re/) {
print $+ {integer}, " ", $+ {fraction} // "UNDEF", "\n";
}
This prints "1 2", as expected. But if we change it to:
my $re = qr /(?<integer>\d+)(?:\.(?<fraction>\d+))?/;
if ("1 3.4" =~ /$re $re/) {
print $+ {integer}, " ", $+ {fraction} // "UNDEF", "\n";
}
it prints "1 4", getting something from the first $re and something from
the second. I would have expected "1 UNDEF".
Just to repeat what I said on IRC today:
For this you can use %-, which will given an array of the results of
each capture buffer of a given name. So the above example could be
written:
my $re = qr /(?<integer>\d+)(?:\.(?<fraction>\d+))?/;
if ("1 3.4" =~ /$re $re/) {
foreach my $set (01) {
print $- {integer}[$set], " ", $- {fraction}[$set] // "UNDEF", "\n";
}
}
You'd get the results from each pair correctly ordered.
Although this reminds me,
STMT for keys %-;
STMT for keys %+;
Both warn about being ambiguous, which is annoying.
D:\dev\perl\ver\zoro\win32>perl -e"$x++ for keys %+"
Warning: Use of "keys" without parentheses is ambiguous at -e line 1.
D:\dev\perl\ver\zoro\win32>perl -e"$x++ for keys %-"
Warning: Use of "keys" without parentheses is ambiguous at -e line 1.
D:\dev\perl\ver\zoro\win32>perl -e"$x++ for keys %x"
Now, my guess is that returning the leftmost defined capture is useful
in cases like:
/(?<foo>PAT1)bar|(?<foo>PAT2)baz/
But you could make use of (?|) then:
/(?|(?<foo>PAT1)bar|(?<foo>PAT2))/
That is, if you have the same NAME repeated, return the leftmost capture
regardless whether defined or not, and use (?| ) if you want the leftmost
defined one.
Yes, i guess so.
Although I think that given that we have %- this just becomes TMTWTDI.
Alternatively, leave it as is, and have a way of getting to all the named
captures, if if they share the same name. For instance after:
my $re = qr /(?<integer>\d+)(?:\.(?<fraction>\d+))?/;
if ("1 3.4" =~ /$re $re/) {
print $+ {integer}, " ", $+ {fraction} // "UNDEF", "\n";
}
$+ {"integer"} eq '1' # Leftmost defined.
$+ {"fraction"} eq '4' # Leftmost defined.
$+ {"integer.1"} eq '1' # First capture of 'integer'
$+ {"fraction.1"} eq undef # First capture of 'fraction'
$+ {"integer.2"} eq '3' # Second capture of 'integer'
$+ {"fraction.2"} eq '4' # Second capture of 'fraction'
Yep, thats pretty much what you can do but the syntax on the last four would be
$-{$name}[$idx]
Another issue, the NAME of named captures have similar constraints on
the name as identifiers - except that you cannot use '::' inside them.
That's a pity because with '::' there would be an obvious way of using
name spaces in your NAMEs.
I'm open to modifications of the naming rule so long as it ensures
that names are easy to parse and can't be confused with integers
(possibly signed) so that \g{} stays unambiguous.
Cheers,
Yves