Language change and code breaks

[Terry Reedy]

In the essay that  began this thread, I wrote

"Deletions: If a feature is directly deleted, it presumably is rarely
used and
not too useful."  I made mention of the only pure Python deletion that
I know of, the removal of the unused 'access' keyword.

The discussion, somewhat heated, initially focused on the proposed
division change.
After about a week, Guido mentioned the idea of deleting case
sensitivity from Python.

Since Guido labeled my essay a 'useful summary' and never directly
challenged any of it,
   including the deletion principle laid out above (nor did anyone
else),
since case sensitivity is commonly used by people who consider it
helpful to essential, since its removal would break much code and
outrage many people,
since the above are, to me, bad things to do,
since, to me, it is a 'problem' hardly worth discussing,
since it can instead be accomplished in an editor/IDE without break
and outrage,
since Python, while not a language of intentional duplication, is one
of flexibility, and
since its removal could likely be easily be reversed in a parallel
fork of Python ---

I was puzzled, didn't take the proposal seriously, only skimmed the
resulting discussion, and kept quiet.  After a few days, I noticed
that 'case' heat had replaced 'division' heat, thought Guido clever
for the diversion, and kept quiet to let the 'joke' continue.  It
seems I was wrong about Guido's seriousness.

So I am back to being puzzled and would like to return to my original
topic.  Leaving the case proposal aside, does anyone have a serious
alternative to the delete criterion quoted above?  If so, what, in
general, should be the less-than-finite burden placed on someone
proposing to delete a feature usefully used more than rarely?  For a
hypothetical example, what would be needed to justify removing
underscores (which must confuse beginners as much as case) either from
all public identifiers or just from public builtin identifiers?  (I
personally would prefer that the latter, at least, had been the rule f
rom the beginning.)

Terry J. Reedy