[OpenAFS-GK] Re: [OpenAFS-win32-devel] rxk5 openafs for windows work in progress--seems to work

[Marcus Watts]

...
> > //2 - com_err issues.
> > 	Each of these might exist and is likely to be an afs issue in a
> > 	given environment.  They all has gratuitous incompatibilities, and
> > 	often also depending on version & build configuration:
> > 		e2fsprogs	- tytso, debian linux
> > 		heimdal		- kth, various
> > 		mit		- from source for unix,
> > 				also windows, macosx, solaris
> > 	and especially (!) openafs
> >
> > 	problems include:
> > 		compiling error tables
> 
> Where is compiling a problem?

You mean entirely inside openafs today, linking external things
against openafs, or in tomorrow's bright shiny new world?

Today, openafs comes with its own compile_et, which is of
course 100% compatible with the comerr that's inside of openafs.
No problem.  :-)

Linking external things against openafs means using the same compile_et
that came with openafs.  That's a packaging decision, which means
various distributions pick different versions of compile_et from the
above list (and probably more) of potential candidates.  My bet is that
openafs usually winds up the loser.  That's probably fair considering
that afs doesn't make /usr/lib/libcom_err.a but instead makes
/usr/lib/afs/libcom_err.a.  People who install openafs probably wind up
with the openafs compile_et since it's probably the last copy installed
- if they install it.

In tomorrow's shiny new world?  Maybe openafs shouldn't be providing
its own com_err but should instead link against some external comerr.
That means using somebody else's compile_et.  This could either be very
very good, or very very bad.

> 
> > 		non-standard bases - or openafs special errors
> 
> We eliminated the odd bases in OpenAFS by switching to error tables of the 
> correct case name. The special errors still suck.

You're got this slightly confused with lower-case base names.
Yes, it's good the lower-case thing was fixed.
The special errors & non-standard bases are
	-457	rxgen		afs/rxgen_const.h
	-8	rx		rx/rx.h
	101	vice

heimdal has "non-standard base" support which makes it possible
to do something useful here.  Ie, error_message() can be vanilla code
and not have special hacks to deal with AFSness.  The error_mesage()
that's in rxk5 already does this.  I think I talked tytso into doing
the same thing, so e2fsprogs may be right as well.

...

				-Marcus