| 14:12 | <Himanshu Shubham> | Thanks for resource. really appreciate it. May I know why you said |
| 17:03 | <shu> | by that i meant that we'd prefer that this optimization didn't exist |
| 17:05 | <shu> | because it is fundamentally kind of dangerous: you used to have a GC object in the heap at location p, and now you've split it into two |
| 17:05 | <shu> | this kind of object surgery is easy to get wrong and easy to forget in other parts of the system, adding complexity |
| 17:06 | <shu> | for example, if you're marking the heap concurrently with the mutator in another thread, what if left trimming happens at the same time for an object you're scanning? |
| 21:20 | <Himanshu Shubham> | While I might not grasp all the technical details, I do get the main idea. So, does SpiderMonkey's optimization in this area also introduce potential bugs, or did they manage it differently? The article I was referring to https://jandemooij.nl/blog/some-spidermonkey-optimizations-in-firefox-quantum/ |
| 21:28 | <bakkot> | All engine optimizations introduce potential bugs. |