Confirmed users
130
edits
JavaScript:GenerationalGC: Difference between revisions
Jump to navigation
Jump to search
→Initial Steps
No edit summary |
Wmccloskey (talk | contribs) |
||
| Line 64: | Line 64: | ||
* The method JIT would need to be changed to take into account that objects on the JS stack may move during stub calls. So we would need more syncing code. I'm unsure of how the trace JIT would need to change. | * The method JIT would need to be changed to take into account that objects on the JS stack may move during stub calls. So we would need more syncing code. I'm unsure of how the trace JIT would need to change. | ||
* We have to make sure that all other roots are such that we can update them if an object moves. This may involve changing our rooting APIs, as well as their callers. It sounds like it's fairly common for C++ objects to point directly to JSObjects. | |||
* Then we need to implement the generational collector. That is, we need a separate nursery and a facility to collect the nursery without collecting the tenured space. It's hard to see how to reduce this into independent parts. However, getting a very simple implementation working shouldn't be an enormous amount of work---perhaps one or two weeks. Given that we would already have tried a few designs (and actually implemented simplified versions of them), I'm optimistic about coding time. | * Then we need to implement the generational collector. That is, we need a separate nursery and a facility to collect the nursery without collecting the tenured space. It's hard to see how to reduce this into independent parts. However, getting a very simple implementation working shouldn't be an enormous amount of work---perhaps one or two weeks. Given that we would already have tried a few designs (and actually implemented simplified versions of them), I'm optimistic about coding time. | ||