SunOS vs. Ultrix comparison

[David Collier-Brown]

alan at shodha.enet.dec.com ( Alan's Home for Wayward Notes File.) writes:
| 	IF you trust the vendor not to change the function and interface
| 	to the parts of shared library, then the Q/A is easy.  As long
| 	as it still runs then it probably runs correctly.  The problems
| 	come if the vendor changes the function of something so that it
| 	isn't quite the same as before.  Or changes the size of data
| 	structure.  Presumably the vendor will have a method of applying
| 	version numbers to shared libraries that the dynamic load checks
| 	to make that things haven't changed (ident mismatch in VMSese).

	This does not appear to be implemented **or implementable** with
	an unchanged sun/bsd interface...

| 	Even if the vendor uses these version numbers correctly, if they
| 	change the version numbers to often (every new release or maybe
| 	every other release) it requires the 3rd party company to re-link
| 	at each library change.  And if you have to re-link you have to
| 	re-test.

	Compatable roll-forward and -back is a known solved problem in
	computer science **for cases of a client program requesting a
	service via an invariant interface**.  The tricky part is evolving
	an interface, which tends to cause one to pore over lint output.

	This may be solved in the domain of formal spcification of the
	interface, where a change in the specification can yield, by a
	somewhat exhaustive procedure, a partial specification of the change
	required to the code.  I just don't know if anyone has done the work
	yet...

--dave
-- 
David Collier-Brown,  | davecb at Nexus.YorkU.CA | lethe!dave
72 Abitibi Ave.,      | 
Willowdale, Ontario,  | Even cannibals don't usually eat their
CANADA. 416-223-8968  | friends.