bug#34506: 27.0.50: push-button bug with basic text-property button

["Basil L. Contovounesios" <contovob@tcd.ie>]

Robert Weiner <rsw@gnu.org> writes:

> On Mon, Feb 18, 2019 at 3:51 PM Basil L. Contovounesios <contovob@tcd.ie> wrote:
>
>  Eli Zaretskii <eliz@gnu.org> writes:
>
>  >> From: Robert Weiner <rsw@gnu.org>
>  >> Date: Sun, 17 Feb 2019 18:46:09 -0500
>  >> Cc: 34506@debbugs.gnu.org
>  >> 
>  >> And what about (button-type (button-at (point))) returning
>  >> nil when button-at returns non-nil.  Both of these functions
>  >> operate on push-buttons as the button.el code reflects, right?
>  >> If so, then that should be a bug.  If not, then it could use
>  >> some explanation.
>  >
>  > button-type requires a button as an argument, whereas button-at is
>  > documented to return a marker for text-buttons.  So you cannot safely
>  > invoke button-type if the button at point might be of the text-button
>  > type.
>
>  Buffer positions, markers, and overlays all qualify as "buttons", so
>  button-type works with both text- and overlay-buttons (but not widgets).
>
> But as I think I noted in my first message, my recollection
> is that button-type returned nil when given a marker value
> returned from button-at.

I think what's confusing you is that button-at returns a marker even
when there is no button at point.  If either of the following two
expressions evaluates to nil, then there is no button at point:

  (button-type (button-at (point)))
  (button-type (point))

If there is something that *looks* like a button at point, yet these
expressions evaluate to nil, then you're probably looking at a widget
instead.

>  Since widgets use text-properties,

AFAICT Customize widgets use overlays, not text properties.  I concluded
this by comparing the results of (text-properties-at (point)) and
(overlays-at (point)) with point at a Customize button.

> button-at on a widget can return a non-nil value, so to say that
> widgets and buttons are unrelated ignores the programming API.

Again, that button-at returns a marker for a widget is a coincidence,
not part of either library's API.  Last time I read/skimmed the relevant
manuals I was not given the impression that these libraries were
related, but suggestions for clarification of their text is always
welcome.

> Maybe the solution is to add a more opaque programming abstraction
> atop each type so that they don't expose their underlying
> implementations and cause programming errors.

Can you please elaborate?  I don't see how either library's
implementation is exposed beyond what is documented in its respective
manual.

>  So I'm guessing what you meant is "you cannot safely invoke button-type
>  if the button at point might be a widget rather than a button".
>
> Yes, again noting that widget documentation specifically mentions push-buttons
> (push-button is the function used to activate buttons).

push-button is indeed a button.el function used to activate buttons, but
this has nothing to do with the push-button widget type, which is
documented under '(widget) Introduction' and '(widget) push-button'.

> This negates the idea that the two constructs are wholly independent
> of each other.

I hope I have managed to convince you otherwise.

-- 
Basil