On 7/1/07, Mark Jason Dominus <mjd@plover.com> wrote:
>
>Dave Mitchell:
>> NB - It has traditionally been the case that 'local $_'
>> was unsafe, and that it was better to do 'local *_'
>> instead. Can anyone remind me why this is, and I'm
>> wondering whether the assorted local() fixes have made
>> this point moot.
>
>       tie $x, "DO_NOT_MODIFY";
>       sub DO_NOT_MODIFY::TIESCALAR { bless [] => $_[0] }
>       sub DO_NOT_MODIFY::FETCH { return rand; }
>       sub DO_NOT_MODIFY::STORE {
>               # system("do-bad-stuff");
>               die("I told you not to modify this!")
>       }
>
>       for ($x) {
>               use_dollar_underbar();
>       }
>
>       sub use_dollar_underbar {
>               local $_;
>               print "okay...\n";
>               $_ = 1;
>       }
>
>
> use_dollar_underbar() wants to operate on $_ without
> tampering with its global value and without causing
> any strange action-at-a-distance.  So it uses "local $_"
> to try to do this.  But "local $_" *itself* tampers
> with the global value and causes a strange
> action-at-a-distance.

And more specifically... Changing it to confess()
instead of die shows that

    local $_;

calls DO_NOT_MODIFY::STORE().

Yves

--
perl -Mre=debug -e "/just|another|perl|hacker/"