NVIDIA Cg Toolkit [LSARC/2008/495 FastTrack timeout 08/08/08]

[Darren J Moffat]

Richard L. Hamilton wrote:
> Volatile, externally controlled, that's the price of admission on this one, ok.
> That they provide it at all is better than that they don't, and adding it so the
> end user doesn''t have to hunt for it and/or has a similar experience as on other
> platforms, may be desirable.  (I trust it's redistributable even if closed source.)
> 
> Nevertheless, the lack of version numbers on the libs bothers me.  If the
> files are actually like that, I for one would rather it was on the record that
> upstream had been asked (whether they act on it or not is of course up to them).
> And if they were receptive, that version numbers only changed for incompatible
> ABI changes, rather than tracking source version numbers as some dubious practices
> might have it.

Version numbers of the files don't actually help with that, they just 
provide a sad illusion that it might - yes sometimes it does but it all 
goes wrong when multiple copies of the library end up being dragged into 
a process address space.  We have seen this problem several times, one 
very notable one was libXm.

If you really want to track the ABI of a library then the symbol level 
versioning provided by ld is they way to do this.

I completely agree on the dubious practice of tracking source code 
versions, that isn't what matters it is changes/additions/removals to 
the API and ABI that matter.

-- 
Darren J Moffat