On Thu, Nov 14, 2013 at 2:30 PM, Lesley De Cruz <lesley.decruz@xxxxxxxxx> wrote: > On 14 November 2013 11:23, Alexander Gladysh <agladysh@xxxxxxxxx> wrote: >> On Thu, Nov 14, 2013 at 12:29 PM, Alexander Gladysh <agladysh@xxxxxxxxx> >> wrote: >>> We observe intermittent failures in ff.new, when attempting to >>> allocate a largeish chunk of memory on Linux (see below). C calls like >>> calloc work fine. >> >> In fact, the failures are _not_ intermittent. I guess, first >> allocation is freed by GC when second fails: > [...] > > ffi.new is also subject to the current 2GB limit (as the objects > created by it are also GC'ed, I suppose?). > See: > //www.freelists.org/post/luajit/Any-plans-to-increase-the-GC-upper-limit-of-LuaJIT-on-Linux-x86-64,3 > "If you allocate these with malloc() and not ffi.new(), they get > access to the full x64 address range." > > Also, to avoid confusion, the "LuaJIT 2.1" referred to in that post is > now LuaJIT 3.0. Thank you for the reference! Mike, is this ffi.new limit documented somewhere on the Wiki? (I also did not find "official" documentation for the memory limits — but I probably did lot look for them well enough.) Best, Alexander.