Re: w5 GC bug regression? [Re: lua51w4-rvm segmentation fault]

["Adam D. Moss" <adam@...>]

Roberto Ierusalimschy wrote:
> We think it is quite easy for the programmer to implement this flag
> in the most appropriate way for her particular needs.

It isn't hard at all, though it is laborious and, I think,
unnecessary to do manually (is there ever a valid reason to
want to use the value of a post-finalization object from lua?).

> Sorry to ask, but why can't you create B and C before A? If A needs
> them, wouldn't be more natural to create them first? What happens
> if there is an error while creating B (after A is already created)?

With some fiddling I'm sure I can change the creation order -- in this
case at least -- but without awareness of the subtleties of the issue I
didn't realize the hazards of not doing so, and it's a quite unintuitive
way of approaching this particular problem.

Object A is of a class W which requires subsystem X to be activated,
but subsystem X is quite expensive to have running so it is only
activated by the first call to the constructor on an object of class
W.  Subsystem X creates the two objects B,C upon activation,
thus they are created after object A.  Conversely (in an ideal
non-lua_close world) the finalization of the final object (in this
case A again) of class W causes subsystem X to be closed down,
requiring the use of objects B and C to do so.

--Adam
--
Adam D. Moss   -   adam@gimp.org