CAD-406 fixed space leak in LogBuffer

[CodiePP]

description

• there was a space leak in LogBuffer; this fix was possible with great input from @ksaric

testing

1.) compile using stack --nix --profile
2.) run using stack --nix --profile exec example-complex -- +RTS -hc -L45 -RTS
3.) analyse: hp2ps -c example-complex.hp

checklist

• compiles (cabal new-clean; cabal new-build)
• tests run successfully (cabal new-test)
• documentation added and created (cd docs; nix-shell --run make)
• link to an issue
• link to an epic
• add estimate points
• add milestone (the same as the linked issue)

[CodiePP]

[mrBliss]

Note that this just force the list to WHNF. Using _ to indicate a thunk, this changes csCounters = _ into csCounters = _ : _, so this is not forcing all the elements of the list.

[mrBliss]

If this really is leaking space (we need some evidence for that first), then removing the bang is not the solution. Either use a stricter data structure than a list or force its contents (e.g., Cardano.Prelude.forceElemsToWHNF).

[ksaric]

While this does remove the memory leak, I have to admit that I'm still unable to completely figure out how the leak is occurring.
My assumption would be that the leak is happening because LogObject is referenced here so it cannot be GC'd, but interestingly the bang pattern on lo does no good.

[mrBliss]

Can you show a heap profile of before in addition to the one after? Otherwise, it seems like you're just adding bangs in random places.

[mrBliss]

Can you show a heap profile of before in addition to the one after? Otherwise, it seems like you're just adding bangs in random places.

@CodiePP informed me that the JIRA issue has a heap profile before and after.

Rather annoying that we have to start looking at multiple places now for information, instead of just GitHub, the place we are already using for issues, PRs, and code.