On Sat, Mar 22, 2014 at 8:26 AM, K. Gadd <k...@luminance.org> wrote: > I mentioned self-hosting to illustrate the need for weak references: > It's fine that the JS > environment has no weak reference alternative as long as all the 'hard > stuff' is > written in C++, but it seems as if there are reasons to move and > more functionality into JS in order to create new optimization > opportunities and reduce the security attack surface.
We've been doing this for a while now, and have run into quite a lot of issues in performance, security, correctness, memory management, scalability, etc. This whole discussion is kind of out of scope for this thread, but just keep in mind that not everybody shares the view that this is the future. > Whether or not WRs are the right solution for current JS leak > problems, philosophical disagreements with GC visibility in JS are not > a good reason to rule out being able to solve real application design > problems. They're not at all philosophical. The (very real) problem is that websites can start to depend on GC behavior, and break mysteriously when we try to change it. In a world as wild as the web, behavior that is technically undefined/non-deterministic but in practice deterministic is incredibly dangerous. This is significantly less dangerous (though still bad) with privileged JS shipped with the browser. We still have to track down the buggy JS, but at least we have the power to fix it. bholley. _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
