On May 3, 2009, at 2:37 AM, Igor Stasenko wrote: > 2009/5/3 John M McIntosh <[email protected]>: >> Beware: there are two issues here, not one. >> >> (a) the faulty remoteString code is ill behaved. Fine that can be >> fixed or as Igor suggested refactored to oblivion. >> > IMO this should be done sooner or later. Its a pain to see it. > Oh well, too bad, this is not the only thing which requires > attention :) > >> (b) The deeper issue is that we affected how finalization works so >> that this ill behavior now causes application failure. >> If in the past it worked, now it doesn't I don't see anyone really >> having a good answer other than perhaps my guess, >> and how do we get back to the point were ill behaviour by *my* code >> won't cause Socket Failures. >> > I used finalization in my code multiple times. It didn't caused many > problems, if you know how to do it right (by taking an implementation > into account). > >> No doubt people have ugly code that is *silently* busted like >> RemoteString, but they don't know it. >> And as you saw actually finding the culprit is difficult. >> >> Lastly some people DO use finalization to do resource cleanup on >> purpose, not as a safety fallback, so they will >> be impacted I think by the new Pharo behaviour. >> > It looks like i missed this part. Where i can read details about what > is altered in finalization, which could break things?
In this thread or the one related john mentioned it. > > >> Wasn't finalization tagged as bloated and need fixing? Igor I'm sure >> suggested a few things before the flurry of >> code for optimizing semaphores and process switching? >> > Yes, i proposed a little change to VM (just a couple lines of code in > single method + registering additional special object) > which would allow us more direct finalization. > It not makes weak refs to be a full pledged ephemerons, but much > easier to adopt in VM. And besides you wont find any finalization code > in squeak which relies on ephemeron's special behavior -- reference > 'value' slot weakly, only when reference in 'key' slot is dies. > > http://www.nabble.com/An-idea-about-better-finalization-support-td23186805.html > > >> On 2-May-09, at 1:52 PM, Stéphane Ducasse wrote: >> > > > > -- > Best regards, > Igor Stasenko AKA sig. > > _______________________________________________ > Pharo-project mailing list > [email protected] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > _______________________________________________ Pharo-project mailing list [email protected] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
