Front page | perl.perl5.porters |
Postings from June 2021
On Wed, Jun 30, 2021 at 03:05:04PM +0200, Eirik Berg Hanssen wrote:
> On Wed, Jun 30, 2021 at 12:42 PM Nicholas Clark <nick@ccl4.org> wrote:
>
> > It seems that if $ is matched, it needs to be the (logical) end of the
> > pattern, else the entire pattern matches. By "logical", I mean that other
> > text follow in the regex, but it needs to be other alternations, or
> > something
> > else that doesn't contribute to matching. It can't be more characters, even
> > if they are present on the next line.
> >
>
> You can match more characters, but you need to consume the newline
> first. (The $ is a zero-width lookahead assertion – it does not consume
> anything.)
D'oh! Thanks. Yes. That was what I goofed.
So, based on your suggestion I've written what I think is a fix, and
pushed it to https://github.com/nwc10/perl5/tree/DD-84569
The pure-Perl implementation is this:
$pat =~ s <
(\\.) # anything backslash escaped
| (\$)(?![)|]|\z) # any unescaped $, except $| $) and end
| / # any unescaped /
>
{
$1 ? $1
: $2 ? '${\q($)}'
: '\\/'
}gex;
The C implementation might actually be easier to read. :-)
It passes tests on blead. I've not *yet* tested it against older Perls,
so I'm not going to roll a CPAN dev release yet.
(Also, a fresh head tomorrow might spot some bugs or corner cases that I
missed)
Thanks for your insights on this.
Nicholas Clark
Thread Previous
|
Thread Next