Re: additional sysconf variables

On Tue, Jul 12, 2011 at 8:41 PM, Christopher Cramer
<[email protected]> wrote:
> On Tue, Jul 12, 2011 at 04:56:22PM -0400, Matthew Fluet wrote:
>> For the most part, we've been following the Single Unix Specification
>> for the POSIX functionality.  The currently supported sysconf
>> variables correspond to those that are required by the specification
>> (http://pubs.opengroup.org/onlinepubs/009695399/functions/sysconf.html).
>>  I'm not sure the best way of handling variables above and beyond the
>> required set.  Adding them piecemeal like this seems to be an onerous
>> process.  SML/NJ tries to mine <unistd.h> for _SC* names and generates
>> a table specific to the target platform.
>
> Well... I don't think it's that onerous. I just copied the few extra ones
> from the Linux docs. Checking the other platforms supported by MLton,
> Solaris also has SC_NPROCESSORS_MAX, and then HPUX has a about a dozen
> others. FreeBSD, OSX, AIX, OpenBSD, and NetBSD don't have any others
> (most of them have just SC_NPROCESSORS_CONF and SC_NPROCESSORS_ONLN -
> SC_NPROCESSORS_ONLN is in fact the one I want).
>
> Personally, I don't care about anything except Linux, but adding
> SC_NPROCESSORS_MAX is not a big deal. The HPUX ones are mostly for
> checking CPU characteristics, and I don't think they'll ever be useful
> for a program written in SML. If you'd really like, I could write a
> thing to do it the SML/NJ way though.

Fair enough; the NPROCESSORS_{CONF,ONLN} and {,AV}PHYS_PAGES seem to
be fairly widely supported.  I've committed the patch.

I will note that my glibc docs (for Fedora 15; also see
http://www.gnu.org/software/libc/manual/html_node/Constants-for-Sysconf.html#Constants-for-Sysconf)
include quite a few more _SC_* variables, though most of them seem to
be of limited use (e.g., _SC_INT_MAX).