Are '1st decimal' R rollouts (e.g. 3.2.0) qualitatively different?

Hi all,

With the upcoming 3.2.0 upgrade, the question came up among my students,
how often a regular user who is not a cutting-edge developer "must" upgrade
their R, given that on Windows/Mac this includes the inconvenience of
re-installing dozens of packages.

In this context, I was wondering whether the annual '1st decimal' upgrade
like 3.2.0 is qualitatively different from the interim ones like 3.1.3. In
other words, whether some changes are reserved for those annual upgrades,
or whether all upgrades are essentially equivalent.

In an x.y.z upgrade, our policy is that a change to z should not break
any well-behaved code, but a change to y or x might do so.  In the case
of 3.2.0, I don't know of anything that is broken relative to 3.1.3, but
there are probably a few obscure things.

Changes to z are almost always a good idea to adopt.  They shouldn't
break anything that isn't taking advantage of bugs, but should fix bugs.

So I would say regular users should *always* adopt changes where z
becomes a value greater than 0.

The question of z=0 is harder.  There are likely to be changes in x.y.0
which are not as well tested as those in other versions.  However, R is
an open source project.  You aren't paying anything for using it.  The
reason it works so well is because so many people use it, and report bugs.

If you choose not to use a z=0 version, you are essentially stealing the
effort of the rest of the community who contribute to testing and
development.  You are also making the z=1 version worse, by not
reporting bugs in the z=0 version.

In the particular case of 3.2.0, there are not a lot of new changes, so
probably not a lot of new bugs, and you may well be better off from a
reliability point of view using it instead of 3.1.3.  This isn't always
true, so you should pay attention to the news about changes in the new
version, and make your decision based on your own circumstances.

Duncan Murdoch