Unexpected memory growth

[luke-clifton]

Unexpected memory growth

[ocharles]

Nice catch! I've been wanting to play with the new info table profiling in GHC 9.2, maybe I can figure out what's going on!

[ocharles]

Wow, info table profiling is amazing! The source of the leak is this line from switchP:

                Just new -> liftBuildP $ p2 `changeParent` new

Though the info table really points to

changeParent node parent =
    RW.tell $ BuildW (Deps.changeParent node parent, mempty, mempty, mempty)

Understanding why that causes a leak is a different matter... But I can at least say that changing this to Just new -> return () fixes the leak (but may also be changing behavior, I'm not entirely sure).

[ocharles]

This may be a duplicate of #152

[luke-clifton]

Hmm it looks the same yes. Though I thought I'd tried it with never, and not seen the growth. I'm double check that later.

[ocharles]

Yea, this is a bit different due to the event switching into itself. I really want to get this fixed, but it's surprisingly tricky! I'm hoping I can use ghc-debug to visualize the heap and understand who's keeping what alive.

[luke-clifton]

Just noticed you can comment out the reactimate and you still get the growth. (Which is perhaps a different bug?)

If I use never then I do see the growth with the reactimate and no growth without it.