There's no <malloc.h> on OSX

[nrnrnr]

I'm trying to build mosml on OSX. But there's no <malloc.h> file on my system.

There is a <malloc/malloc.h> and a <sys/malloc.h>.

[kfl]

That sound strange. My main platform for developing Moscow ML these days are OSX, but maybe I have polluted/enhanced my local setup.

Can you give me a bit more information about your setup? Are you using gcc or clang to compile the runtime, for instance? (I always use gcc as it makes Moscow ML 5-15% faster for some reason.)

Also, where do you get the compile error? Because we really shouldn't include malloc.h anywhere

[nrnrnr]

That sound strange. My main platform for developing Moscow ML these days
are OSX, but maybe I have polluted/enhanced my local setup.

Can you give me a bit more information about your setup? Are you using gcc
or clang to compile the runtime, for instance? (I always use gcc as it
makes Moscow ML 5-15% faster for some reason.)

I'm using someone else's Mac, about which I know almost nothing.

I'm using something called /usr/bin/gcc but it's clearly an alias for
clang.

I found malloc.h in /usr/include/malloc/malloc.h, but I ran into
enough other issues that I stopped filing them. Including no access
to mallopt().

Also, where do you get the compile error? Because we really shouldn't
include malloc.h anywhere

Maybe the best thing for me to do is to send you the diffs between
what I pulled from github and what eventually compiled? And the joker
in the system is that all of this is on top of my own changes (which
are limited to things like the version string and converting some
warnings to errors).

Norman

[kfl]

It sound like you have a modified version of src/runtime/gc_ctrl.c (*), please check that you are using the gc_ctrl.c from github

(*) At one point we had a hard to track down bug on 64 bit Linux, that triggered when malloc() switched to using mmap(). Thus, the advice back then was use mallopt(), but this advice (and bug) is/was glibc specific and is no longer needed.