Re: Upgrade experience 5.1p10->5.2 Beta

I can't speak to most of these issues, but...

On Sat, 29 Nov 2003, Justin Smith wrote:

> First I tried making world and kernel from the cvsup'ed standard 
> supfile. This went OK.
> 
> Then I did installkernel and installworld (in that order) and the second 
> step crashed with messages that shared libraries were missing. (It's as 
> if it installed some files before installing the shared libraries they 
> needed).

Did you reboot between installkernel and installworld?  It looks like
there may be some source upgrade nits, but my understanding was that if
you follow the complete directions in UPDATING, things should generally
work well.  The only difficulty I ran into upgrading from about
5.1-RELEASE to 5.2-JUSTBEFOREBETA a couple of weeks ago was that although
the net 5.2-BETA / contents are much smaller than 5.1, during the upgrade
you overlap lots of larger binaries, /rescue, and /lib, and the result is
you briefly need more space rather than less space.  The box in question
was installed as 4.0-CURRENT, and had a very small / file system, and
required a bit of manual adjustment to get forward as a result.

> 1. Will the AGP performance be brought up to the level of 5.1p10 before
> 5.2 is released (so I don't have to slow the AGP interface and can do
> OpenGL)?

Are you running with GENERIC from 5.2-BETA?  If so, you may wish to
recompile your kernel with WITNESS and INVARIANTS turned off and see if
that helps.  These debugging features are critical during the development
process, but also seriously impact performance and might make the
difference you're seeing.

> 2. Why does the system fail in such an non-robust and uninformative
> fashion? Why no error messages? 

Unfortunately, some classes of hardware/driver failures are very hard to
convert to explicitly caught panics, especially where they involve the
mechanism by which panics are displayed (system console), or low level
memory management, etc.  Many developers run with serial consoles so that
they can still get access to console debugging information even when in X
Windows, etc. 

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research