From: Eric Wong Date: 2018-06-07T10:30:17+00:00 Subject: [ruby-core:87441] Re: [Ruby trunk Feature#14767] [PATCH] gc.c: use monotonic counters for objspace_malloc_increase I wrote: > I think the original unsigned + underflow avoidance code > prevented us from accounting free() properly, and we were > triggering GC more as a result. (Warning: thinking out loud while sleep-deprived) What's dangerous about this patch correcting the free()-before-malloc() accounting is that it not only breaks expectations with our current malloc limits; everybody who sets custom limits in env also gets things broken. So, I'm not sure what to do about it... Unsubscribe: