Front page | perl.perl5.porters |
Postings from May 2003
Nick Ing-Simmons wrote: > I mean the technical how-it-will-work, and how-it-won't break existing > modules with overload etc. discussions ;-) > From last time round those will take a while. > My goal is to make it transparent to existing overload modules, no significant performance hit for other code, and as simple as possible to program. It should also make your teeth whiter and your breath smell minty fresh. YMMV ;~) As a first pass, I want to investigate whether I can set the SMG flag on the overloaded RV and extend the current mg_set code to handle it. The specific callback will be stored with the other package callbacks in the class stash, so it will differ from U-magic in that all objects of that type must have the same callback. What I would like to do would be to extend "overload '=' => &callback" so that I don't need to create a new keyword, but I don't yet know whether that would be possible. The issue, of course, will be clashing metaphors if both the LHS and RHS are themselves magical: $a = new Class::A "1.0"; #implements assignment overload $b = new Class::B "fish"; #also overloaded objects ...sometime later... $a = $b; #here there be dragons The simplest methodology would be to effectively rewrite that last line as $a = "$b"; # safe??? This won't affect the behavior of Class:B (which can predate the change) except that the "$a = $b" will no longer be a shallow copy. This would mean that Class::A would be sticky, but that _is_ the point after all. I suppose we could also allow Class::A to choose whether to use $b->stringify or $b->numify as its input. Hmmm... Basically, I think the logic has to be this: $a = "these aren't the droids you are looking for"; $b = new Something::Wicked 1.3; ...civilizations rise and fall... $a = $b; # shallow copy, old string abandoned ...techtonic plates shift... ++$a; # which is implemented as # $a = $a->clone(undef,""); # $a->incr(undef,""); ...the sun expands in supernova... $a = 42; # called as $a->assign(42) I presume that the current code knows to do the clone() because it can see that more than one object is referring to the same value. Presumably, the same logic can be used to know when to clone() prior to assign(). John -- John Peacock Director of Information Research and Technology Rowman & Littlefield Publishing Group 4720 Boston Way Lanham, MD 20706 301-459-3366 x.5010 fax 301-429-5747Thread Previous | Thread Next