Re: Is amd64 ready for the desktop?

["Steven M. Bellovin" <smb%cs.columbia.edu@localhost>]

On Mon, 17 Mar 2008 14:40:40 -0400
"D'Arcy J.M. Cain" <darcy%NetBSD.org@localhost> wrote:

> My experience is no but before I switch back to i386 I thought I would
> see what others' experiences are.
> 
> First of all, the NetBSD system itself seems to be extremely stable.
> As a server system I suspect it is just fine.  All of my problems seem
> to be with add-on packages and only a handful at that.  Unfortunately
> those few packages are the ones that I need on the desktop.
> 
I'm running amd64 on my laptop.  I agree with your assessment -- the OS
is very solid, but some apps are problematic.

The biggest single problem I've found, in a number of packages, is the
implicit assumption that ints and pointers are the same size.  This in
turn tends to happen when the declaration for subroutines is omitted.
(In a few cases, the problem was due to inappropriate declaration of a
feature set macro; NetBSD stuck to the letter of the spec while Linux
does not.  This was exacerbated by bugs in the Debian/Ubuntu man pages
for some such routines -- I did the Ubuntu equivalent of filing PRs on
the man pages, but that doesn't magically fix the buggy apps.)

Where I've tracked these down -- claws-mail, liferea, xfig -- I've
reported the problem upstream and/or committed a patch.

In retrospect, I realize that there were often compiler warnings.  But
with pkgsrc, those tend to fly right by.  Perhaps setting the
appropriate flag -- -Wall, perhaps? -- when rebuilding would catch
these.

The two programs I'm having trouble with right now are gtkpod and
Openoffice.  The former is get a mutex error; I suspect that that's
more multicore-related.  I haven't had time to track it down yet,
though.  (Nor does it pay to file a PR, since I'm the package
maintainer.)  I don't know what to do about Openoffice -- that's
serious enough that I contemplated switching back to i386.  In some
sense, I've given up -- last week, I ordered a Mac Mini and a KVM
switch, so I can at least encrypt/decrypt Word and Powerpoint in my
office...  One person thought it was a gcc bug fixed in a later release
than is available for NetBSD.

Neither lsof nor asapm compile on amd64.  I plan to fix the latter, but
haven't gotten around to it yet.

                --Steve Bellovin, http://www.cs.columbia.edu/~smb