Linux emulation version number

[Boris Samorodov]

On Fri, 23 Mar 2007 13:11:40 +0100 Divacky Roman wrote:
> On Fri, Mar 23, 2007 at 11:36:54AM +0100, Alexander Leidinger wrote:
> > Quoting Boris Samorodov <bsam at ipt.ru> (from Fri, 23 Mar 2007 02:36:37 
> > +0300):
> > 
> > >Hi!
> > >
> > >
> > >The FC6 port doesn't install with linux.osrelease=2.4.2. Assuming that
> > >FreeBSD supports only two values (2.6.16 is the second one) I use the
> > >following line at the Makefile:
> > >-----
> > >.include <bsd.port.pre.mk>
> > >
> > >LINUX_OSRELEASE!=       ${SYSCTL} -n compat.linux.osrelease
> > >
> > >.if ${LINUX_OSRELEASE} != "2.6.16"
> > >IGNORE= supported only for compat.linux.osrelease: 2.6.16
> > >.endif
> > >-----
> > >
> > >Something similar should be written to pkg-install script for
> > >packages sake. The Porters Handbook says it's not good to use sysctl.
> > >But such approach should be used for all upcomming fc6 ports...
> > >
> > >What do you think? Thanks.
> > 
> > I suggest to do it the other way around:
> >  .if ${LINUX_OSRELEASE} == "2.4.2"
> >  IGNORE
> >  .endif

> I dont think this is a good idea.... what if someone sets 2.2?

Should we call them innocent?
You shouldn't skip Alexander's comments. ;-)

> from slightly different topic... I think the version should follow these rules

> 1) we should choose 2.6.16 as our target (we already did but its not stated much) and
> implement its FULL functionality, we're almost there lacking basically only the *at syscalls
> (which are almost here as well). when splice() (native one) is finished we should implement
> linux_splice() (a trivial wrapper) and switch to 2.6.17 or possibly 18 as our reference linux version.

> 2) the osrelease should be set to the number of the kernel the actual FC is shipped with if < 16 oterwise
> to 16.

> I plan to lookup what are the major differences between various linux versions and then we can change
> these numbers cause now I know only about the 2.6.16 and splice in 2.6.17

> I think this way we will get most stability because the FCs are tested with the given kernel number
> and using higher might lower stability (glibc not being tested for this kernel) and lower number
> might cripple our ability to emulate..

> opinions?

You may be right. But ATM I think the best default is as Alexander
has proposed. This may change after you do what you have proposed. ;-)


WBR
-- 
Boris Samorodov (bsam)
Research Engineer, http://www.ipt.ru Telephone & Internet SP
FreeBSD committer, http://www.FreeBSD.org The Power To Serve