@spoofyroot Note that for JS code objects, V8 pages transitions between RW <-> RX today. Changing this to RWX <-> RX should not be a security regression.
Ideally we could get something like MPK on all platforms to make it really secure.
@mraleph @andywingo V8 keeps track of how much native memory is kept alive and factors it into its gc scheduling strategy. The principle idea is to see the native heap as a transitive extension of the managed heap. For some object types (e.g. ArrayBuffers) we distinguish between young and old.
@mraleph@erikcorry @iamtommythorn @pervognsen The write barriers will cost you some throughout but you will get rid of the pause. I guess it ultimately depends on the workload.
@erikcorry @iamtommythorn @pervognsen Concurrent marking is a significant throughout, latency, and memory consumption win trading off overall CPU time.
@Ale_Belli90 @v8js "We expect applications to grow organically out of the research community together with V8 team members at Google." - There is no formal application process. We expect folks to get in touch at conferences, forums, social media etc.
@erikcorry I don’t think they are ugly from an implementation perspective. They unfortunately do have an inherent pause time problem.
The main problem I see is that they make GC observable which creates a new class of bugs complaining that GC is not happening. Won’t fix!
@rifat@bmeurer@mathias@v8js @ulandeg The JavaScript heap is currently limited to 2G of memory.
The task manager has a couple of categories which can show you where memory is going (right click on the title bar to enable them).
Or take a performance trace using chrome://tracing (with memory-infra category enabled)
@rifat@bmeurer@mathias@v8js @ulandeg Got it. I am afraid that we do not have a general mechanism to flush out leaks. Often a leak can be triggered by performing repetitive actions on webpages with a high frequency.
@_developit@_bwhitty@nodejs@reactjs I am wondering whether it measures the accumulative GC time on the main thread per full GC cycle or the main pause of a full GC. A main pause of 100+ms would still be not great. I would love to see a performance trace if this is the case.