BSDInstall: I want the bikeshed painted plaid

Nathan Whitehorn nwhitehorn at freebsd.org
Sun Jan 2 20:19:52 UTC 2011


On 01/02/11 14:04, Garrett Cooper wrote:
> On Jan 2, 2011, at 11:51 AM, Garrett Cooper wrote:
>
>> On Jan 2, 2011, at 10:49 AM, Nathan Whitehorn wrote:

>>> Paint Comes in Plaid? --------------------- With only a little
>>> more spit and polish, I think this could easily replace
>>> sysinstall for the 9.0 release. It works right now on almost all
>>> architectures for which we ship install media. There are no
>>> dependencies on strange scripting languages, and only one on
>>> tools not currently in the base system.
>>>
>>> bsdinstall depends on the newer dialog and libdialog from Thomas
>>> Dickey (devel/cdialog). This is LGPL2, but that is better than
>>> the current dialog's license, and the software is so much better
>>> to develop with than the current mess that is libdialog it might
>>> be good to have anyway. In addition, I believe he is the sole
>>> author, and it's possible he could be persuaded to relicense it.
>>>
>>> I don't entirely know how all of this relates to pc-sysinstall.
>>> pc-sysinstall has some more features (e.g. installation onto geli
>>> disks), and certainly more thought went into it, but requires an
>>> immense amount of work to make it work on non-x86 systems and a
>>> text front-end is not yet complete. When it is ready, a lot of
>>> the infrastructural changes required for bsdinstall (live CDs for
>>> installation, a new libdialog, revised distfile format, etc.)
>>> will probably help pc-sysinstall as well.
>>
>> Please see pc-sysinstall if you like to see a different effort in
>> chartreuse. No guarantees as to how many x86-isms are in the
>> infrastructure.
>
> Should have read down to the bottom of the email... pc-sysinstall
> really needs to be made non-x86 centric in order to be a usable
> installer backend -- not sure how much effort has been put into
> making this happen, but I _do_ have a PowerPC Mac I could test on so
> we could work together to resolve any x86-isms and I might be able to
> hack something together with mips in my spare time (I have potential
> access to systems in my old work group, but I'd need to clear it with
> them before I did anything with that). text-sysinstall was a WiP but
> abandoned because some folks claimed that they were going to produce
> a solution to that with lynx, a webserver, etc on install media.
> Looks like that idea hasn't panned out though (it's been 2 months and
> I haven't seen a prototype or mumbles on arch@, sysinstall@, etc), so
> a libdialog-like solution is needed. Scripting changes should be
> relatively cakewalk as long as I have a process for producing
> pc-sysinstall based media, but there are other potential areas that
> need to be audited for security issues in pc-sysinstall. Once upon a
> time pc-sysinstall cached passwords on installed media based on some
> discussions I had with Bruce Cran... not secure in the least; other
> issues may lurk behind the scenes. Bruce Cran and I are interested
> parties in whatever will transpire, so please let us know what you
> have planned. I think everyone can agree that sysinstall is dying and
> it's time to sink some nails into its coffin before we send it on its
> way.
 > Thanks, -Garrett

I spent a bunch of time looking at pc-sysinstall before starting to work 
on this. The major problem for non-x86 systems is that it heavily 
assumes that your disks are either MBR+bsdlabel or GPT. If you have 
anything different (APM, VTOC8, or even a raw bsdlabel or MBR 
installation on x86), it breaks in strange and fascinating ways due to a 
random mixture of if (scheme == MBR) else and if (scheme == GPT) else in 
the backend. Some of these are easily fixed, but it looked quite 
difficult to root out all of the places this assumption is made, not to 
mention teaching it about various styles of boot code, that the same 
partitioning scheme may need to be set up different ways on different 
architectures, etc. txt-sysinstall also segfaults when you try to run it 
on powerpc at the moment, though that is likely a simple bug and I 
didn't look into it in detail.

What I intended with bsdinstall is to have something simple, flexible, 
and easily maintained that works immediately on all platforms and will 
be ready for 9.0 for sure. If pc-sysinstall materializes before then, or 
after then, and people like it better, I'm more than happy to get out of 
the way; this is the reason the wiki page is titled "Stopgap Installer". 
In the interim, I hoped to at least start laying out the hammer and 
nails next to sysinstall's coffin.
-Nathan



More information about the freebsd-arch mailing list