10:12 <vlad87> Hi everyone,

10:12 <vlad87> I have a curiosity regarding a simple ruby program stolen from this talk https://youtu.be/ZE6F3drGhA8?t=1811

10:12 <vlad87> <<EOF

10:12 <vlad87> def allocations

10:12 <vlad87>   x = GC.stat(:total_allocated_objects)

10:12 <vlad87>   yield

10:12 <vlad87>   GC.stat(:total_allocated_objects) - x

10:12 <vlad87> end

10:12 <vlad87> p allocations { 1 }

10:12 <vlad87> p allocations { 1 }

10:12 <vlad87> EOF

10:12 <vlad87> Running this program with ruby version 2.7.8 works as I imagined - both calls should print "0" to stdout (considering no allocations are happening in the provided block). However, running it with any ruby version starting from 3.0.7 (may not be the exact one introducing the behavior just what i tested with) I get strange results: first call to

10:12 <vlad87> allocations outputs 1 and then any other future call to it outputs 0 as I initially expected. I get even stranger results with something like:

10:12 <vlad87> <<EOF

10:12 <vlad87> class A end

10:12 <vlad87> p allocations { A.new }

10:12 <vlad87> p allocations { A.new }

10:12 <vlad87> EOF

11:12 <wbooze> allocates only once ?

11:51 <o0x1eef> A.new allocates 4 then 2, that's surprising to me, but I'm not sure you can exclude overhead in those calculations, as in, there might be allocations going on behind the scenes that aren't apparent but contribute to the count.

12:13 <o0x1eef> So I'd probably treat it similar to a yard stick rather than an exact measurement

20:10 <havenwood> o0x1eef: I'd say it was object shapes, but that's 3.2. Probably method inlining optimizations or other pre-shape freeze stuff.

20:12 <havenwood> https://bugs.ruby-lang.org/issues/18776

20:15 <o0x1eef> Makes sense, and nice to see Chris's work continue to have an impact