On Fri, Sep 22, 2017 at 04:46:48PM +0100, Ramsay Jones wrote: > >> - int nth_slab, nth_slot; \ > >> + unsigned int nth_slab, nth_slot; \ > > > > I have a feeling that in the long run these should all be size_t, but > > it's probably pretty unlikely to overflow in practice. At any rate, the > > slab index itself is an unsigned, so it probably makes sense to match > > that for now. > > Yes, I briefly considered that, but I didn't want to think about > possible effects of the increased size of 'struct slabname'. I suspect > that it is very unlikely to cause an issue, but I had similar concerns > with the 'ALLOC_GROW' patch, were it would have been more likely to > cause memory pressure issues (to e.g. s/int/size_t/). I decided that it > could be addressed separately, with a patch on top, if necessary.
I _think_ we'd be OK in all of those cases because these size_t's tend to be once-per-big-list, not once-per-list-item. But I agree that it would need looking into closely, and definitely should be addressed separately. Thanks. -Peff
