Sergey Organov [off-list ref] writes:

These patches introduce capability to configure the default format of
output of diffs for merge commits by means of new log.diffMerges
configuration variable. The default format is then used by -m,
--diff-merges=m, and new --diff-merges=default options.

In particular,

  git config log.diffMerges first-parent

will change -m option format from "separate" to "first-parent" that
will in turn cause, say,

  git show -m <merge_commit>

to output diff to the first parent only, instead of appending
typically large and surprising diff to the second parent at the end of
the output.

I think that it is a good goal to free a short-and-sweet "-m" from
getting tied forever to the current "two-tree diff for each of the
parent" (aka "separate"), and a configuration to change what the
"-m" option means would be a good approach to do so.  It would help
the interactive use by human end-users, which is the point of having
short-and-sweet options.  Existing scripts may depend on the current
behaviour, so the configuration cannot be introduced right away, but
over time they can be migrated to use the longer and more explicit
option "--diff-merges=separate".

But I do not see much point in adding the "--diff-merges=default".
Who is the target audience?  Certainly not scripts that want to
avoid depending on the 'default' that can be different and easily
vary per user.

Or is the plan to deprecate and remove the short-and-sweet "-m"
option and standardize on "--diff-merges=<style>"?  If so, such a
design makes sense from pureness and completeness standpoint, but I
am not sure if that is a good design for practical use.