I'm going to ExclusiveArch %{ix86} x86_64 ia64 and put a comment in the spec.

# SMBIOS and PCI IRQ Routing Tables only exist on these arches.  It's
# also likely that other arches don't expect the PCI bus to be sorted
# breadth-first, or of so, there haven't been any comments about that
# on LKML.

Marking this as blocking ppc and ppc64 ExcludeArch trackers.

Created attachment 172394 [details]
proposed patch to the specfile

my patch to let it build and install

Created attachment 172396 [details]
revised patch

I just saw the announcement on the poweredge-list. This looks neat! I'll start
the review of this package.

One immediate question : The resulting binary package depends on the shared
libsysfs.so.2 which is in /usr/lib*/ and not /lib*/ so for Harald's patch needs
further changes to get the package to be able to do anything useful before /usr
is mounted.

right... either move libsysfs or make biosdevname static...

(In reply to comment #6)
> right... either move libsysfs or make biosdevname static...

Yup. Current libsysfs-devel still provides the static library, so that would be
an option, but surely moving libsysfs would be cleaner. From the spec file
comment, it seems like the static libsysfs should already have been used, but it
hasn't been. That comment kind of confuses me overall, though :

# some distros won't have a static lib version of libsysfs, which
# would be needed to generate these .so files

I guess the comment is (at least partially) wrong, since even with the static
libsysfs available, the shared one is used.

Matt, please just get rid of libsysfs usage. It's an outdated, unmaintained and
useless library.

OK, I'll nuke the libsysfs usage.  That comes from my use of code from the 
pcmciautils-014 package, but for what is used there, it's easy enough to 
replace with standard C lib calls.

The code can already build a static binary, called biosdevnameS.  I just 
disabled this for now, but if we really need this called before /usr is 
mounted (e.g. inside an initrd or NFS-root or something) it's easy enough to 
build again.

I've applied Harald's RPM spec patch moving this into /lib* and /sbin.  I
removed the libsysfs dependency completely.  I am building, but not RPM
installing, the static copy of the app, in case we decide we need it later.

This builds and runs on Fedora 7 and OpenSUSE 10.3 beta.  rpmlint results are
empty.  The SuSE and Fedora spec files are identical now, and hopefully they can
stay that way and I can nuke one.

Version 0.2.2 is posted now to http://linux.dell.com/files/biosdevname and
pushed into the git tree.

Looking better. Two more questions :
- What is the point of creating a shared library which is used only by the
binary from this package? Since the .so symlink is wiped out, nothing will be
able to link against it, so not using the static version of the binary only
makes us loose the time it takes to run ldconfig, right?
- The dynamic file list confuses me. Some distros would have something else than
the .so.0 and .so.0.0.0 files without changing the spec file? (btw, the comment
regarding libsysfs above it needs updating now ;-))

it's purely a "convenience" library, meaning I want to separate all the table
parsing and lookups from the "frontend" command line application and it's
parsing.  I don't at this point expect other applications to use the
"convenience" library, but at some point in the future, if I clean up the API
and there's good reason something else needs those functions, it can be done.

The reason for the dynamic file list is that some versions of some distros don't
compile libpci.a with -fPIC, so it can't then be included in a shared library. 
There's autoconf code to detect this and automatically sets enable_shared=no
which results in no libbiosdevname.so.* being created.  (OpenSUSE and SLES are
this way right now, as were earlier versions of Fedora).  This spec builds
Fedora, RHEL, OpenSuSE, and SLES at this point, the dynamic file list being the
only magic involved (to add, or not, depending on existance of those .so.*
files, appropriately).

I'm not tied to the idea of the convenience library.  If it's really that ugly,
it can disappear too.

And yes I can fix the spec file comments.

I decided to nuke the libbiosdevname.* stuff in upstream, it isn't really 
needed and only causes confusion.  Updated SRPM and spec posted.

Spec and resulting package look good to go. I'll just need to runtime test the
actual binary package.

The %description could maybe be changed from :

"takes an kernel name"

to

"takes a device kernel name" (or "takes a kernel device name"?)

Spec %description changed to say "takes a kernel device name".

Okay, package approved! Seems like my workstation's BIOS doesn't override much,
though (Precision 490), but the package didn't break anything either ;-)

Ping? Matt, still around? :-)
Please go ahead and request CVS branches creation, import, build, and close ;-)

I haven't forgotten about this, but I've got a showstopper problem before I am 
comfortable doing a build.  I spoke with Kay about this last week, and he 
wanted an enhancement to udev-114 to let biosdevname suggest the new name 
without overriding the values in 70-persistent-net-names.  I haven't made the 
enhancement to udev yet, so that's what's blocking me on this...  Hopefully 
this week I can spend some time on it.

New Package CVS Request
=======================
Package Name: biosdevname
Short Description: Udev helper for naming devices per BIOS names
Owners: mdomsch,mebrown
Branches: FC-6 F-7 EL-4 EL-5
InitialCC: harald
Cvsextras Commits: yes

cvs done.

http://koji.fedoraproject.org/koji/taskinfo?taskID=169642 built successfully.
Closing.