[webkit-dev] Protecting against stale pointers in DrawingBuffer and GraphicsContext3D

[Darin Adler]

On Oct 12, 2010, at 10:12 PM, Darin Fisher wrote:

> Caused more harm than good?


I think it was a good decision.

I don’t have a lot of patience for these abstract debates — I would much prefer to talk about a specific example rather than these vague principles — but here are some of my thoughts.

The weak pointer pattern is still alive and well in WebKit on an ad hoc basis.

Over these first nine or so years of the project we have had many more problems with null pointer dereferences than with storage leaks due to reference cycles. Code following weak pointers such as the pointer from a frame to its page often incorrectly assumes the pointer value is non-zero.

I empathize with your Mozilla experiences with reference cycles, and I think there are places where it’s a good pattern to use a weak pointer. A general-purpose weak pointer template is expensive enough that I don’t think it’s a good thing for us to add.

I think of reference counting as a way to make pointer lifetimes work, but for many objects there’s a separate object lifetime that’s better to manage directly. I can think of three or four times in the last year where we fixed a thorny problem by adding reference counting to a class that started out as single-owner and I think the track record has been good for this in the project so far.

    -- Darin