Re: (?{...}) outside of patterns

[Brian McCauley]

A lot of rubbish has been spoken in this thread - mostly by me. I now 
undersand the main topic of the thread about binding of lexicals within 
(?{}).

However there is still one side-topic I think I'm right about...
Dave Mitchell wrote:
On Mon, Nov 15, 2004 at 06:17:44PM +0000, Brian McCauley wrote:
>>  no re 'eval';
>>  $foo =~ '(?{ $x = 1 })';
In a perfectly consitant world [the above] should be an error.
No it shouldn't. The basic guiding principle is that if the code
is contained within a literal string, then you don't need use re 'eval'.
If it's the result of an expression, then it's suspect.
No. The basic guiding principle is that if the code is contained within 
a literal _regex_, then you don't need use re 'eval'. If it's the result 
of an expression, then you should need use re 'eval'.  Just because an 
expression consists of a literal string constant should not (ideally) 
make a difference.  On the other hand it's not a serious problem if (as 
an artifact of optomisation) it does make a difference since it does not 
introduce an real security problems.

Please consider the following...
  no re 'eval';
  use constant const => 1;  # Inlinable
  sub func { 1 };           # Not inlineable
  'xxx' =~ ( const() ? '(?{ 1 })' : '.' ); # OK
  'xxx' =~ ( func()  ? '(?{ 1 })' : '.' ); # Error
Can you really claim that this is doing "The Right Thing" (as opposed to 
doing an OK thing due to an artifact)?