Exposing the (full) GC directly in the API is something we are trying
very hard to avoid, as it will be seen as as something generally useful
for the embedding application. However in most cases it just leads to
performance degradations as running the full GC's too often is
expensive.

Having the ability to run GC's to identify actual memory used when
performing heap profiling makes perfect sense, but it should be hidden
somewhere in the Profiling API, and not exposed directly as a general
API.

We do full GC's before scanning the heap in a number of places related
to debugging which have lead to strange bugs as the full heap scan is
somewhat dangerous.

In the long term we should replace the full GC followed by a heap scan
with a heap scanner that will visit only live objects using the actual
mark-sweep algorithm.

http://codereview.chromium.org/159787

--~--~---------~--~----~------------~-------~--~----~
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
-~----------~----~----~----~------~----~------~--~---

Reply via email to