Re: [IBMiOSS] gmake: *** virtual memory exhausted. Stop.

On Wed, Jul 20, 2016 at 2:12 PM, Kevin Adler <kadler@xxxxxxxxxx> wrote:

Right, but don't these sort of changes only happen going from eg. 1.x ->
2.x (in OPS these would be in separate options).

I wouldn't expect Node 0.12.10 -> 0.12.11 to be a problematic upgrade, but
maybe I'm naive.

With or without stated "SemVer" compliance, in real life you can have
breaking changes in ANY version change. That is just the nature of
change.

I know for sure that it happens in Python. And the Python core
developers are an incredibly talented, dedicated, and conscientious
bunch. Truly world-class.

Not too long ago, there was a heated discussion surrounding the
behavior of a function whose purpose is to generate cryptographically
random bits. (Those of you who are familiar with the issues
surrounding Linux's /dev/urandom will know the gist.) Basically, the
decision was made to slightly change the behavior in 3.5.1. No changes
to this behavior were made in 3.5.2. But then the debate raged over
whether to revert the behavior for 3.5.3. I think Guido ultimately
chose to revert. For most rank-and-file Python developers, all of the
3.5.x versions are equivalent. But for certain cryptographically
sensitive applications, under certain very specific conditions, the
behavior of 3.5.1 and 3.5.2 is different than 3.5.0 (and most likely
3.5.3 will return to what 3.5.0 does).

So yes, it matters.

John Y.