bug#17397: 24.4.50; REGRESSION: `temp-buffer-show-hook' is no longer invoked for `describe-variable'

[Drew Adams]

>  > To get *Help* to DTRT for the code or user that previously used
>  > `temp-buffer-setup-hook' or `temp-buffer-show-hook, is it
>  > appropriate to put the same things on `temp-buffer-window-setup-hook'
>  > and `temp-buffer-window-show-hook' instead?  Or will that maybe
>  > have unwanted effects because `with-temp-buffer-window' is perhaps
>  > used for more than just *Help*?
> 
> Both `with-output-to-temp-buffer' and `with-temp-buffer-window' are
> and can be used for other things than displaying *Help*.

For `with-output-to-temp-buffer':  NOW, yes.  Before, no - it was
pretty much hardwired to help mode.

So there will be users and existing 3rd-party code that depend on
that behavior, which endured for 30 years or so.

This is why I have argued that the right fix for that problem was
not to change the behavior of `with-output-to-temp-buffer', and thus
gratuitously break such code and user expectations, but to make a
new macro for dealing with temporary buffer display that was not
hardwired to help.

Emacs itself could move to using that new macro, of course, but there
should be no reason to break the longstanding behavior of
`with-output-to-temp-buffer'.

>  > Not clear to me.  And I find no help in the doc or NEWS - or in
>  > this bug thread, so far.
> 
> The manual says about `with-temp-buffer-window' that
> 
>       This macro uses the normal hooks `temp-buffer-window-setup-hook'
>       and `temp-buffer-window-show-hook' in place of the analogous
>       hooks run by `with-output-to-temp-buffer'.
> 
> What else do you need?

What did I say?  Yes, I read that in the manual.  What about a function
that has been on `temp-buffer-setup-hook' or `temp-buffer-show-hook',
and that was specifically put there for help, i.e., that expects help
mode?

Is it necessarily appropriate to put that function on the *-window-*
hooks, without modifying it to work around use in a non-help mode?
I don't think that is the case, in general.  And I don't see a hook
that is analogous and specifically for help.

In my case of `fit-frame-if-one-window', this is not a problem AFAICT
- I added it to `temp-buffer-window-show-hook'.  But in the general
case there is a problem in moving a function from a temp-buffer hook to
a temp-buffer-window hook, whenever that function is specifically aimed
at help (expects help mode).

Sure, users and libraries can also change such a function, so that it
tests the buffer or mode to see whether it involves help.  But this is
gratuitous hassling, IMO - it should not be necessary.

I made the further point that none of this is documented, AFAICT.
Nothing about migrating one's use of temp-buffer stuff to whatever
is appropriate for Emacs 24.4+.  Nothing about incompatible
behavior changes for the temp-buffer stuff.

>  > Should I be adding `fit-frame-if-one-window' to `temp-buffer-window'
>  > for Emacs 24.4, to get the effect it has in Emacs 24.3 (and prior)
>  > by being on `temp-buffer-show-hook'?
> 
> I suppose you want to add it to `temp-buffer-window-show-hook'.

I supposed so too, and I did.  I asked the question several times,
hoping to incite adding some migration explanation to the doc.  But
thanks anyway for confirming.  I still hope someone updates the doc.

But again, though that is OK for `fit-frame-if-one-window', since I
want to invoke that regardless of which temp buffer is displayed, it
is not OK in general for functions that have been on
`temp-buffer-show-hook'.

It is likely that at least some such functions are specific to help
mode, since `temp-buffer-show-hook' was dedicated to help previously
(and for so long).

What do you tell library maintainers or users who have a function on
`temp-buffer-show-hook' that is appropriate for help mode but not
for other temp buffers?  Such information should be in the doc, IMO.

>  > If not, how to get the same effect?  If yes, how to deal with other
>  > possible uses of `with-temp-buffer-window', which have nothing to do
>  > with *Help*?  (In the case of `fit-frame-if-one-window', that would
>  > probably not hurt anything, but the question is more general.)
> 
> Emacs has a `temp-buffer-resize-mode' which is tied to temporary buffers
> and not to `help-mode'.  To "deal with other possible uses" of temporary
> buffers, a function run by `temp-buffer-show-hook' or
> `temp-buffer-window-show-hook' should probably check whether the buffer
> is in help mode.

It will have to, now.  Too bad.

> In this regard nothing has changed wrt earlier version where a user
> removed `help-mode-setup' from `temp-buffer-setup-hook' and
> `help-mode-finish' from `temp-buffer-show-hook'.

That is a very roundabout way of saying that IF there previously were
a user who for some reason coded things up to approximate what the
code does now, THEN ... "Nothing has changed"!

So-called temp-buffer display was previously coupled with help display.
Functions on `temp-buffer-show-hook' in existing code or user setups
could well depend on that behavior.  It was expected that the temp
buffer displayed would be in help mode - because it WAS - for 30 years.

Sure, someone could have jumped through hoops to use
`with-output-to-temp-buffer' without help mode.  But that would have
been relatively rare.  What percentage of the uses of
`with-output-to-temp-buffer' in the Emacs code corresponded to this?
Not more than 1% would be my guess - maybe 0%.

It is hardly the common use case of `with-output-to-temp-buffer'.

To suggest that it is, that "nothing has changed", would be
disingenuous.  Yes, I note that you did not claim that in general -
you qualified it as being the case only for such exceptional uses
(without acknowledging that they are exceptional).  But the
impression can be got from a cursory reading that you are saying
that nothing has changed in general, i.e., for most uses of
`with-output-to-temp-buffer'.

In fact, I have jumped through that hoop that in some of my code,
so I do recognize such an exception:

(defmacro bmkp-with-output-to-plain-temp-buffer (buf &rest body)
  "Like `with-output-to-temp-buffer', but with no *Help* navigation stuff."
  `(unwind-protect
    (progn
      (remove-hook 'temp-buffer-setup-hook 'help-mode-setup)
      (remove-hook 'temp-buffer-show-hook  'help-mode-finish)
      (with-output-to-temp-buffer ,buf ,@body))
    (add-hook 'temp-buffer-setup-hook 'help-mode-setup)
    (add-hook 'temp-buffer-show-hook  'help-mode-finish)))

But doing things like this was no doubt uncommon.  That was the point
of my original bug report asking that Emacs decouple temporary buffer
display from help display, i.e., that it provide a real temp-buffer
display that does not involve help (as does the macro above, in a
workaround way).

What was wrong was the way this decoupling was done in Emacs:
changing the behavior of macro `with-output-to-temp-buffer'.  That
was misguided, IMHO, and not necessary.  That boat has apparently
sailed, however.

> Maybe you could also use `help-mode-hook' to do something for help
> buffers exclusively and avoid other temporary buffers.
> 
>  > The existing doc is anyway wrong in some cases, AFAICT: The doc
>  > string of `with-temp-buffer-window' says:
>  >
>  >    This construct is similar to `with-output-to-temp-buffer'
>  >    but, neither runs `temp-buffer-setup-hook' which usually puts
>  >    the buffer in Help mode, nor `temp-buffer-show-function' (the
>  >    ACTION argument replaces this).
> 
> What is wrong here?

`temp-buffer-setup-hook' no longer "usually puts the buffer in Help
mode".

>  > And the doc string of `temp-buffer-setup-hook', likewise, says:
>  >
>  >    This hook is normally set up with a function to put the buffer
>  >    in Help mode.
> 
> This is still the case for the release version.  IIRC Leo Liu changed it
> on the trunk so the doc-string should be probably updated there.

The doc string is not updated in the trunk build I cited, from April 29.
I don't have a more recent build.  But yes, the doc should be updated -
that was what I was pointing out.

>  > I don't get the impression that either of those statements is
>  > true anymore.  It seems like there is *no relation* anymore
>  > between such temp-buffer things and Help mode.  Furthermore,
>  > grepping for `temp-buffer-setup-hook' in the Emacs sources
>  > shows that it is not used for this at all anymore.
> 
> The connection between temporary buffers and help mode is established
> by `with-help-window' which uses `with-temp-buffer-window'.

Yes, I know.

And previously such a connection was hardwired in
`with-output-to-temp-buffer'.  That behavior should not have changed.
Not because it was good to couple the two behaviors, but because they
have been so coupled for a long time in `with-output-to-temp-buffers'.

The right approach would have been to (a) leave
`with-output-to-temp-buffer' the way it was, (b) decouple temp-buffer
display from help mode by coming up with new macros, and (c) use the
new macros in the Emacs source code (this has been done).

That way, the Emacs code would no longer use `with-output-to-temp-buffer',
and thus would no longer depend on a coupling of temp-buffer display
with help, BUT users would not be bothered and existing 3rd-party code
would not be broken by a change to `with-output-to-temp-buffer': it would
continue to work as usual, coupling temp with help as before.  Eventually,
`with-output-to-temp-buffer' would be abandoned everywhere, naturally.

I pointed this out in my original bug report.  That is not what was
done.  We now have to live with the consequences of the incompatible
change.  That calls for doc that explains the change and how to deal
with it.

The doc (manual and/or NEWS) should clearly point out (a) what
`with-output-to-temp-buffer' does now, and (b) that it does not do what
it has always done before, and (c) this is how to migrate existing code
that uses it: A, B, C,...Z.

For Emacs's source code the change was trivial: replace uses of
`with-output-to-temp-buffer' with uses of the new macro.  But for
3rd-party code that supports multiple Emacs versions the change is not
so trivial.

It is too bad that we have arrived here, but we have.  Now let's patch
things up by at least letting users know about the damage done and how
to make do with it.