Application certification (was Re: [Beowulf] hpl size problems)

[Mark Hahn]

> > Worse, when closed source vendors ship product, they qualify against
> > specific OSes, and will not officially support others.  Usually we get a

this is a huge disservice to the user.  it's obscure enough that they 
don't grasp why, though.  this sort of deeply offensive laziness on the part
of app vendors is one of the reasons that computers SUCK SO MUCH.
it gives rise to things like the Msft monoculture (uniformly bad and 
rarely updated).  it means that the only real multiplatform apps are 
those which ship in source (and companies don't like that - interestingly
because of piracy, even if they claim it's because of their competitors.
(any real competitor can trivially white-room clone anything you do, so 
keeping the source doesn't matter.))

> There's a very good article at LWN about the LSB entitled "The relevance of 
> the Linux Standard Base" at:
> 
> http://lwn.net/Articles/152580/

it points out that an ABI even if accepted, water-tight, and used,
doesn't address an app's dependencies on non-system stuff (whatever that is).
defining what range of interfaces are covered by the ABI is the problem.

> http://www.livejournal.com/users/udrepper/8511.html

I thought this was quite unfortunate, but not too surprising if you've ever
read mailinglist messages from him.  it's probably true that test suites are 
of poor quality.  it's probably true that people who do standards for a
living have an interest that is not necessarily in making the standard better.

but those problems don't mean we can live without an ABI standard.

I'm just thankful that the apps I care most about (say, Pathscale's compiler)
happen to "support" the distro I use (for historic reasons).  and thankful
that most of my users have their own code, rather than wanting to be
button-pushers with some closed app.  even Gaussian users (notoriously
focused on the chemistry, not the computation) tend to be easy because 
that app is available in source...