Re: FreeBSD 5.3 setup help

[Richard Frith-Macdonald]

Forgive me for butting in, but you and David appear to be talking 
without
connecting ... at least you appear to be ignoring his points.

Your point seems to be basically that use of '/proc' is not portable 
and that
the /proc api is bad.  That may be true, but it's not a useful point 
since we
have to work with the features actually available on a system. 
Certainly most
of the ms-windows api is neither portable nor good, but we still have 
to use it
on a windows system.

Davids point was that he would like to see examples of specific cases 
where
the existing code does not work as it should on some systems, 
preferably with
patches to fix it on those systems (ones which don't break other 
platforms), so
we can improve things. Apart from your 'side note' below, you seem to 
have
ignored this request.

On 2 Feb 2005, at 13:12, Chris Vetter wrote:
> I said, even though there is no /proc on FreeBSD 5.x, you can always
> use --disable-proc (or something similar) on 'configure' since BSD
> does have libkvm.
>
> The problem with NSProcessInfo is, that if libkvm does NOT exist, and
> tries /proc instead, it most likely will fail, because NSProcessInfo
> tries to open and read file '/proc/<pid>/cmdline' which may well exist
> on Linux, but does not necessarily exist on other systems. Or if it 
> does,
> it does not necessarily have the expected information.

Of course code would fail if it tried to use system-specific features 
on the
wrong system.  David probably considered that too obvious a point to be
worth mentioning.  The 'configure' scripts are used to detect what 
features
are available on a particular system and cause the NSProcessInfo code
to be compiled with the appropriate feature support.

> On a side note regarding NSProcessInfo+load:
> The arguments passed to kvm_open() differ from what Solaris' kvm_open()
> expects and kvm_getprocs() doesn't even exist on Solaris. Contrary to
> what the opening comment states about being compatible.

That's interesting information.  Do you have code which *does* work on 
Solaris
and a patch to alter the configure script to detect the difference that 
you can contribute?
I suspect that the author of the kvm code developed it on BSD rather 
than Solaris.

Does this KVM code actually get used on Solaris?  I would have thought 
that, if code
containing a non-existent function was used, we would get compile 
errors on a solaris
machine, and I haven't heard of them.