[CentOS] The delays on CentOS 5.6 are causing EPEL incompatibilities

Wed Mar 23 15:40:24 UTC 2011
R P Herrold <herrold at owlriver.com>

On Tue, 22 Mar 2011, Les Mikesell wrote:

> Building the kernel shouldn't be an issue - but look at the 
> SL notes on the srpms that don't build with the listed 
> dependencies as shipped - and they aren't being picky about 
> the library linkages matching the RH binaries like CentOS 
> is.

>  If the RH build links things from source they don't ship, 
> how much can you trust the projects that depend on that 
> source to be able to ship timely updates?

Sometimes looking at the list and the posts, I feel like I am 
watching a group of nuns, talking (speculating) about the life 
issues of Las Vegas showgirls

In trial building the upstream's '6' sources, about the only 
circular build dependency that comes to mind was an openMPI / 
valgrind '-devel' pair that was cross dependent and needed for 
later packages.  It was easy enough to 'bootstrap' around, as 
the dependencies were not 'versioned' such that a prior 
valgrind worked just fine to break the circularity

The compulsive obsession on matching every library version 
exactly is usually just not an issue to most users of any 
distribution, so long as they do not have a third-party (and 
non-LSB conformant) application that absolutely positively 
needs a given library for some reason.  Some of the very high 
end accellerated graphics drivers oriented for some NVidia 
chipsets in certain blade configurations fall over and die 
back to non-accelerated, because the driver vendor is calling 
some non-exposed library interface; some simulation software 
return slightly varying results out several bits of precision. 
Other than that, the Unix that we live in is very forgiving 
with a quick recompile thanks to the FSF / GNU work on the 
autotools

PLUG: if the darn applications were written to a given LSB 
level, these issues would go away.  But frankly for what one 
pays for some of these applications, adding a license from 
upstream is lost in the 'rounding error' of the price /PLUG

I am not against such efforts to match at the library version 
level [it is articulated as part of what CentOS does], but it 
is usually not the end of the world when a person has to port 
around some minor deviation in the build environment

'Mother superior 'Les, later ... 
> they do rely on the upstream which previously was not
> openly hostile to rebuilds

It was not always so ... in the early days, there was pushback 
against the rebuild efforts in general; there is pushback 
toward commercial 'free-riders' now.  This comes and goes, and 
really there is no substitute for actually 'doing' rather than 
talking in the cloister

It is not the end of the world when one hits a build problem, 
as the sources, at the end of the day, are provided, and one 
can study and read.  Indeed, as the collection of Linux 
variants (and thus soliutons of others to study) out there has 
grown, it is much easier these days to solve such issues [I 
solved a cfengine-3.1.4 yesterday with minimal effort]

-- Russ herrold