Run the GC occasionally

[Philip Taylor]

JS won't automatically run the GC until it's run out of memory, which might be an inconvenient time to freeze the simulation. We should probably call it manually at regular intervals, when it's least likely to cause disruption (e.g. during a frame when it's not also busy doing all the simulation computation), to make it more predictable.

[(none)]

Milestone Unclassified deleted

[historic_bruno]

This seems to be partially fixed now? At least for simulation and AI scripts.

I notice a stutter if I'm, for example, scrolling the map during a game. This is due to the GUI GC running and it's enough to be annoying. The larger the runtime size, the more noticeable the stutter. If running the GC more regularly (perhaps as often as the simulation and AI scripts) would fix that, then it's an idea.

Not sure the impact of GC on random map generation, but it's worth examining.

[wraitii]

(not sure how relevant this ticket is anymore).

RM have a way to call for a GC, and my AI v3 API adds a way for AI to call a GC, along with the automatic GC once in a while.

[historic_bruno]

I think the point about controlling when GC occurs is still relevant. We really need to examine its impact on performance.

[Yves]

I'm taking this task, but I won't do anything until we have moved to generational garbage collection, because this will change GC behaviour quite significantly.

[wraitii]

We have, I'm bumping this to ask if it's still needed/already completed?

[Yves]

We have some heuristics to run GC from time to time now (actually since Alpha18 already), so it can be considered fixed. However, SpiderMonkey is still changing in that regard, so it will probably have to be checked again and adapted with each new SpiderMonkey release. Keeping the ticket open to remind us about that doesn't do much good, so I'll close it.