On 24 October 2015 at 21:39, Bart Schaefer wrote: > OK, let's try it this way. This retains MOST of 36834, but puts back > scanning the entire heap if we intend to free the block to which the > current fheap pointer refers. > > (This is a patch against 8e9a68ad but applies to current head.) My results. 36943 has string_test speed of current head (so it's slightly better than in your results). And memory usage is in check, so that's a fix. I calibrated for string_test, it's interesting how much slower array_test is on OS X. My machine is 2.3 GHz, maybe that's some reason (your is faster IIRC). Best regards, Sebastian Gniazdowski