Sounds good to me.
> -----Original Message-----
> From: sbcl-devel-admin@...
> [mailto:sbcl-devel-admin@... Behalf Of William
> Harold Newman
> Sent: Saturday, December 14, 2002 12:54 PM
> To: sbcl-devel@...
> Subject: [Sbcl-devel] Re: Core files
> On Sat, Dec 14, 2002 at 09:27:12AM -0800, Dave Richards wrote:
> > One difference between UNIX mmap() and Win32 MapViewOfFile()
> (although there
> > are many similarities) is the "allocation granularity" in the process
> > virtual address space. On UNIX you can map regions on a page
> > both as a file offset and as a virtual address space address.
> In Windows,
> > however, the allocation granularity is 64k.
> > (Aside: You can blame this on Alpha, btw. Dave Cutler made it
> 64k because
> > the Alpha architecture Reference Manual claimed that an
> implementation might
> > only support page sizes as small as 64k. This never happened,
> though. The
> > Windows NT group would never have allowed it.)
> > Therefore, Windows-based cores must align sections on 64k
> boundaries. This
> > is easy to do in that the math is easy to drop into the
> existing code. But
> > it is, however, an extra #ifdef/#!+/- test. Would anyone
> objects to making
> > the core section alignment 64k universally, or should I just
> > the code in runtime/save.c and genesis for Win32? My patches
> are currently
> > conditionalized.
> This needs to be visible in genesis, and when reading and writing cores
> in src/runtime/, right? (so perhaps visible in src/runtime/save.c,
> src/runtime/coreparse.c, and src/compiler/generic/genesis.lisp?)
> I think I'd prefer to have it defined in Lisp somewhere, e.g. as
> (defconstant +min-core-alignment-bytes+ ...)
> or something, and then copied (automatically) into sbcl.h, much the
> same way that src/compiler/*/parms.lisp stuff is now. Perhaps it
> could even go in src/compiler/*/parms.lisp, alongside the existing
> #+WHATEVER-OS definitions like
> (def!constant read-only-space-start #x01000000)
> That way the value would be defined once and only once, and then
> referred to by one name thereafter, instead of being explicitly or
> implicitly referred to separately in two or three separate places,
> which is too much maintenance fun for me.
> If you choose to do that, note the hacks in GENESIS which decide
> what Lisp-level constants get copied into sbcl.h as C #defines:
> MAYBE-RECORD-WITH-TRANSLATED-NAME, MAYBE-RECORD-WITH-MUNGED-NAME.
> William Harold Newman <william.newman@...>
> hoping to beat the odds since 1972
> PGP key fingerprint 85 CE 1C BA 79 8D 51 8C B9 25 FB EE E0 C3 E5 7C
> This sf.net email is sponsored by:
> With Great Power, Comes Great Responsibility
> Learn to use your power at OSDN's High Performance Computing Channel
> Sbcl-devel mailing list