> So of course most of the time you don't have this problem of a suddenly changing libc under MS Windows, because each program has its own libc. Now what happens in case some version contains a security exploitable flaw? Security effort are duplicated in such an environment. (And this is just an example.)
This seems almost a strawman. Criticize Microsoft's policy if you want, but all nye suggested was using symbol versioning for this particular known-to-be-dangerous change. This wouldn't cause any security issues (and may even avoid some).
I agree that it isn't possible to avoid every regression. For example, newer software often has performance regressions on old hardware. However, this seems like a particularly serious regression, so if there was an easy way to stop old versions of software silently corrupting data it may be worth taking.