On Wed, Sep 14, 2016 at 05:41:55AM +0200, Theo Buehler wrote: > On Tue, Sep 13, 2016 at 01:29:17PM +0200, Otto Moerbeek wrote: > > As often, real life came in between. Did anybody do measurements? I > > really would like to to see hard data. > > It seems that the price is relatively modest. > > I ran several 'make build's: > > 3rd gen X1, i7 5500U, 2.4GHz (amd64): > no malloc.conf, malloc.c r1.195: > 1525.04 real 2505.98 user 1362.47 sys > no malloc.conf, malloc.c + diff: > 1532.15 real 2540.63 user 1356.98 sys > > for comparison: > malloc.conf -> J, malloc.c + diff: > 1554.76 real 2596.40 user 1353.26 sys > > Acer Aspire 5633WLMi, Core 2 Duo 1.66 GHz (i386): > no malloc.conf, malloc.c, r1.195: > 5020.77 real 5725.21 user 1609.28 sys > no malloc.conf, malloc.c + diff: > 5088.07 real 5865.80 user 1572.12 sys
Did some more thinking about this. Since this diff interferes with the effort to move canary bytes closer to the end of the requested size (See tedu's diff from December https://marc.info/?l=openbsd-tech&m=144966528402282&w=2), I like to revisit that diff and fix it, and then see if and how we can to random junking. tb@ pointed out at least realloc is broken with that diff. I have to go and see if I can spot the actual problem. -Otto