[compiler-rt] [scudo] On Android, only release when changing M_DECAY_TIME to 0. (PR #154194)

[via llvm-commits]

ChiaHungDuan wrote:

> While adding tracing, I found one place where the release to OS call happens early in the process life and it doesn't make a lot of sense. That's why I'm removing it.

I think the reason for doing a force-purge is that when switching back from 0 to 1, we want to make sure the "heap" is back to the managed status so our page releasing heuristic won't be impacted. Switching to 0 means the user is aware of the impact of unreleased pages thus doing a purge before that seems not showing much value ?! Not sure why I did it for both cases (maybe for consistency?). Given this reason, what do you think if we only do it for switch to 1?

https://github.com/llvm/llvm-project/pull/154194