bug#30186: 27.0.50; Password is not hidden in read-passwd

[Alan Mackenzie]

Hello, Juri.

Firstly, apologies for answering your first post without having
understood it properly.

On Thu, Jan 25, 2018 at 23:15:41 +0200, Juri Linkov wrote:
> >> Do you agree we could remove the effect of
> >> with-silent-modifications around set-text-properties,
> >> and leave it only on remove-list-of-text-properties?
> >> This will help to fix the reported regression.

> > I'm not sure about this.  Doesn't `set-text-properties' need to be
> > "protected", too?

> I'm not sure either.  Do you think that `set-text-properties' without
> `with-silent-modifications' will cause the same problem that you
> described in http://lists.gnu.org/r/emacs-devel/2015-04/msg00506.html

set-text-properties outwith a with-silent-modifications will most
definitely cause that problem.

> > I'm not sure why you want to do this.  Why do you want to do this?

> Doing yank `C-y' in the minibuffer of `read-passwd' puts dots `.'
> over the yanked characters using `display' properties, then later
> `set-text-properties' removes all properties (exposing the yanked
> characters), but without `with-silent-modifications' it used to put
> `display' properties back.

OK, I understand this, now.

> After the change that added `with-silent-modifications',
> the hook that puts `display' properties back doesn't run.

Yes.

[ .... ]

The problem here appears to be a fundamental design bug in Emacs: that
text properties are regarded as part of the buffer rather than something
accompanying the buffer, as overlays are.  before/after-change-functions
are applied alike to proper buffer changes and text property changes.

read-passwd absolutely needs an after-change-function to run on the
text-property changes in remove-yank-excluded-properties.  This seems
fair enough.  CC Mode absolutely requires the change hooks _not_ to run
on the text-property changes in remove-yank-excluded-properties.  The
two conflict with eachother.

A workaround might be for read-passwd to use an overlay rather than a
text property for display.  But this doesn't solve the underlying
problem.  I now see that the invocation of with-silent-modifications in
remove-yank-excluded-properties is not the Right Thing.  It breaks the
model of text properties being an integral part of the buffer.

Perhaps an alternative would be for Emacs to provide a flag which
indicates to a before/after-change-function whether the current change
is a "proper" change or merely a text property change.  Change hook
functions could then test this flag and, for example, refrain from doing
anything for a text property change.

-- 
Alan Mackenzie (Nuremberg, Germany).