On Fri, Jan 18, 2008 at 10:08:40AM -0200, Iruata Souza wrote: > On Jan 18, 2008 9:48 AM, Simon Kuhnle <[EMAIL PROTECTED]> wrote: > > > > On Thu, Jan 17, 2008 at 02:06:01PM -0800, Josh Elsasser wrote: > > > On Sun, Jan 13, 2008 at 05:15:40PM +0100, Simon Kuhnle wrote: > > > > On Fri, Jan 11, 2008 at 06:30:47PM -0800, Josh Elsasser wrote: > > > > > On Fri, Jan 11, 2008 at 06:21:44PM -0800, Josh Elsasser wrote: > > > > > > This is a port of the SBCL native code compiler, based on patches > > > > > > which were recently committed to SBCL CVS. Only i386 is supported > > > > > > at > > > > > > the moment. This is my first port so feedback is more than welcome, > > > > > > in particular regarding how it has to download a 7 meg binary to > > > > > > bootstrap itself from. > > > > > > > > > > > > Note that you will see several timer regression tests fail unless > > > > > > you > > > > > > are running a recent -current snapshot, sys/kern/kern_time.c 1.63 is > > > > > > needed for them to pass. > > > > > > > > > > I guess it would be helpful to mention that the port is at > > > > > http://www.elsasser.org/openbsd/sbcl-port-1.0.13.tgz > > > > > > > > The build fails for me with the following error: > > > > mmap: Cannot allocate memory > > > > ensure_space: failed to validate 805306368 bytes at 0x4c000000 > > > > > > > > Complete buildlog here: http://blarzwurst.de/openbsd/sbcl-1.0.13.log > > > > > > > > Note that I uncommented tetex in build depends, but that shouldn't be > > > > the problem (is this really needed by the way?). > > > > > > > > Running i386 snapshot from January 11th. > > > > > > Sorry about this, I forgot to mention the annoying virtual memory > > > requirements to build this. The easiest way to make sure your > > > resource limits are high enough is to build as root or a user in the > > > staff login class. I've updated the above url with a new version of > > > the port which sets VMEM_WARNING. > > > > > > I've also rolled a new bootstrap tarball since I realized the old one > > > was out of date. > > > > I didn't look at the build system, but isn't there another way to built > > this, without requiring so much memory? > > > > I may be wrong here, but a bootstrap sbcl is required to run for it to > compile and sbcl requires big memory areas.
This is correct, the way SBCL currently does memory allocation is to mmap() a huge chunk of it at a fixed address and then grab little bits of that as needed. Since UVM checks resource limits when the initial mmap() call is done you have to have a high datasize limit even if most of the memory is not going to be used. I suppose the build script could be patched to pass --dynamic-space-size to sbcl, but since it isn't really using most of the memory I didn't see it as much of a problem. In answer to your earlier question, tetex is used to build some of the documentation. -jre
