Research custom pointer implementation for boa gc#91
Conversation
shruti2522
changed the title
nekevss
left a comment
nekevss
left a comment
There was a problem hiding this comment.
This is a pretty good write up. I'm actually super excited to see this concept in action. Can you extend it with some actual example code?
It would be worth attempting to create a new allocator that relies on a custom pointer and implement some level of heap serialization.
|
|
||
| // private | ||
|
|
||
| // TODO(perf): O(n) scan; replace with a sorted index at scale. |
There was a problem hiding this comment.
I left this as a linear O(n) scan for the prototype since pools will be small. For production, we should probably replace this with a sorted index or a direct chunk map for O(1) lookups
| } | ||
|
|
||
| // overflow guard | ||
| if slot_count as u64 > super::MAX_SLOT_IDX as u64 { |
There was a problem hiding this comment.
Added explicit guards against slot_count arithmetic overflows and corrupt live_count data here. Since we are reading raw bytes from disk, we might want to throw a fuzzer at this deserializer in a follow up PR
nekevss
left a comment
nekevss
left a comment
There was a problem hiding this comment.
Accepting this as is, because I think the work here is interesting.
I'm not sure that the tradeoff of not being able to deref is worth it, so we may not continue much further down this route for now. I think a custom pointer type would still be incredibly useful but we may have to reevaluate how it may affect the API.
2 participants
Related to #86