bug#24340: insert-file-contents calls before-change-functions too late

[Stefan Monnier <monnier@iro.umontreal.ca>]

> I'm sorry, but I must insist.  The change I proposed is the only one
> in insert-file-contents for that use case that I'm prepared to
> consider in the current situation.  (We could also leave this bug
> open, until such time as a more thorough refactoring is done of the
> related functionalities.)  Deeper changes are exactly the can of worms
> that I don't want to open to fix just this one use case, especially
> since Emacs 25.2 will almost certainly be branched from what now is
> the master branch.  Such deeper changes were what Alan proposed in the
> first place (with a similar patch), and I already said I didn't want
> to do that.

Right: I do not intend for my patch to go to 25.2.  But for Emacs-26,
it seems there's plenty of time to find and fix any possible fallout.

> I think a cleaner way is to change the model of how we partition
> piecemeal changes, when we signal the changes and when don't, when we
> ask the user about supersession-threat, etc.  The current model is
> fundamentally flawed, if we want to use the buffer-change hooks in the
> ways that emerged from these discussions.

All I care about is for any change to the buffer to be announced by
a prior b-c-f.


        Stefan