No GC until 1 minute and then GC spike

[matt (gingerbeardman)]

Summary

I'm just not seeing any GC at all for the first minute or so of running my game, and then suddenly there's a ton of GC resulting in a big spike.

I would expect to see regular smaller GC hits, rather than the one large lump.

The source of my GC is known to be allOverlappingSprites()

Comparison

1.0.8

• constant GC, as expected (shows as dark grey)

1.1.0 with frame rate fix, default GC settings

• default GC settings
• custom 5ms GC setting
• custom 5ms plus setGCScaling(x,y) settings

all give the same result.

• no GC at all... (just light grey)

Shortly after the above screen screen grab a sudden GC spike led to this:

• ...then, sudden large GC spike
• approx 10x previous level on the chart

Original post

Using default GC settings I am seeing a big GC hit and resultant frame rate stutter.

• This happens after 1 minute of my game running
• and every 1 minute after that
• I can reproduce it on demand
• changes to setMinimumGCTime() and setGCScaling() make no difference?
• also happens with my recent demo which was built with 1.0.8 SDK

notes

• all timers are removed from my game
• GC hit is usually in allOverlappingSprites(), one of my most called functions, declared as local+const

thoughts

• is this a problem with the way my game works?
• or is it something outside of my control?

Screenshot

Screen recording

• Imgur: The magic of the Internet

ps: I am using a 1.1.1 test version

[matt (gingerbeardman)]

Upgrading this to a Bug Report.

[matt (gingerbeardman)]

Interestingly, if I do manual collectgarbage() calls in my code then I can see GC() calls every frame.

But why are they not happening regularly with default settings?

[marc]

This is fixed in 1.1.1.

When a game requests immediate frame updates with setRefreshRate(0), we weren't giving time for inter-frame garbage collection, even if setMinimumGCTime(x) had been called.
This was causing garbage to build up, and then a long emergency collection would cause a spike in frame time for a few frames in a row. This is also why your manual collectgarbage() calls fix the issue.

For 1.1.1 I've modified this so that games that are using setRefreshRate(0) will get a hard-minimum GC time as requested by a call to setMinimumGCTime(x). I think this feels a bit more true to the API definitions, and probably developer expectations as well.

[matt (gingerbeardman)]

Music to my ears!

Thank you