On May 29, 2012, at 7:31 AM, Brandon Benvie wrote: > Right, but I guess I tend to automatically look at it in terms of what the > amount of exposure is to JSland for guidance in assessing what the magnitude > of implementation difficulties will be. I imagine the ways something like the > WeakMap gc changes break are ones that are mostly areas where they just don't > work efficiently, or don't work at all. Where the ways proxies break ends up > with all sorts of crazy elevated access holes.
The exposure surface of most GC changes is all of JSland... GC bugs are among the hardest to find. GC code is generally "unsafe", GCs operate globally upon the entire heap, it may modify the content of the stack, any object, any internal or user data structure, the noticeable effect of a GC bug is often has significant temporal lag before it manifests , there may be significant non-deterministic behavior in the user program that triggers a GC bug making hard to replicate, Allen _______________________________________________ es-discuss mailing list [email protected] https://mail.mozilla.org/listinfo/es-discuss
