Front page | perl.perl5.porters |
Postings from February 2013
On 2013-02-17 13:43, demerphq wrote:
> On 17 February 2013 13:14, Dr.Ruud <rvtol+usenet@isolution.nl> wrote:
>> On 2013-02-17 02:15, demerphq wrote:
>>> We changed which level of the perl parser handles escapes intended for
>>> the regex engine.
>>>
>>> Previous to this the \x20 would be resolved to a space, and as far as
>>> the regex engine was concerned the pattern would be " ".
>>>
>>> After this change the \x20 would be delivered to the regex engine
>>> verbatim and the \x20 form would not be recognized by the heuristic
>>> that handles the " " case.
>>>
>>> This change was very desirable for many reasons, and as it doesnt
>>> actually contradict the docs, unless Ricardo says otherwise I consider
>>> this Not A Bug.
>>
>>
>> See some split() cases below. So, #4 should behave as #7..10.
>>
>> So, the PATTERN "\x20" should be compiled as /\x20/, not as " ".
>
> So you dont agree that the original ticket is a bug?
Heheh, "the ticket is a bug".
I am really not sure whether the ticket is wrong or not.
So I will just live with any outcome.
- - - - - - -
If the compile-time "" operation (AKA qq: explicit string extrapolation)
is always done first, like at 'pre-processor' level, then that is
easiest to explain and document, I think.
" " always becomes ' ', and "$x\n" always becomes ($x."\n"), etc.
But then we should compile the split-PATTERN "a*" as /a[*]/.
Also because the split-PATTERN " $" currently leads to:
'Final $ should be \$ or $name at -e line 5, within string'.
So the split-PATTERN ' $' currently behaves different from " $".
Should split( q{\x20} ) then not also differ from split( qq{\x20} )?
The other way is to completely defer extrapolation in split-PATTERN
context, and add some flag to the PATTERN if "" or qq was (not) around.
So I see issues with both ways. IMO the clearest is to not allow any
literal string but a single white space, which then rejects " $" as a
bad split-PATTERN, instead of as a bad string. Sure, that will break
some code.
--
Ruud
Thread Previous
|
Thread Next