Re: HEADS UP: sysinstall is no longer the default installer

On 03/14/11 10:44, John Baldwin wrote:
> On Monday, March 14, 2011 10:13:30 am Nathan Whitehorn wrote:
>> I just committed (r219641) changes that make the release infrastructure
>> (src/release/Makefile) use bsdinstall by default instead of sysinstall
>> on install media. A big thank you is in order to everyone who provided
>> advice, criticism, and testing for this project over the last few months!
>>
>> Along with sysinstall, the original sysinstall build stuff has been
>> preserved (now /usr/src/release/Makefile.sysinstall) and will continue
>> to be for the lifetime of the 9.x release series, although it will not
>> be used by default. This change modifies the process of building
>> releases somewhat, so I'll outline changes that people who run snapshot
>> buildbots will have to make below, and some next steps planned with the
>> installer.
> Please consider supporting using SVN or CVS to obtain docs, ports, and source
> trees.  I have a custom SVN repo at work that is not exported to CVS and
> available via csup and am able to use the existing SVNROOT SVNBRANCH variables
> with 'make release'.  Having support for this sort of thing would be useful.
> I have also made much use of LOCAL_PATCHES in the past for building releases,
> so having support for that would be useful as well.

SVNBRANCH works now, and source comes over SVN, the others via cvsup. 
Support for a different SVNROOT and regular cvs for ports and docs can 
certainly be added. In the case of LOCAL_PATCHES, you can just use the 
regular makefile on your patched tree -- I don't think the chroot and 
checkouts make much sense in this case.

> I think for re_at_ especially it is nice to just do 'make release TAG=7.2' (or
> some such) and have it DTRT to check out matching ports, doc, and src into the
> chroot, etc.  I think the new process should be similarly automated.

The generate-release.sh script likely needs some work. It exists almost 
purely for the benefit of re_at_, and I don't know exactly what their 
requirements are. A list (or patches) would be very welcome. The feature 
you want here, though, can be obtained now by the CVSUP_TAG and svn 
branch arguments to generate-release.sh.

> Have you tested network installs using PXE or the like?  This was fairly easy
> before as you could copy the '/boot' directory from a bootable ISO and the
> mfsroot was self-contained.  Do you now have to put the entire contents of
> release.iso up via NFS?  Is there a subset you put in the NFS root and then do
> an NFS or FTP install?
>

Yes, I have, and it works well (tested on i386, sparc64, and powerpc). 
Right now, you need the whole system (which is a regular installworld + 
the rc.local to give the installer menu, and, optionally the distfiles). 
For the future, the set of things the installer needs from the userland 
is intentionally fairly small. I need to do some work anyway to make a 
minimal system for bootonly CDs and the like, which should also a 
smaller system for PXE as well.
-Nathan