bug#17007: 24.3.50; describe-key/function evaluates documentation function in the wrong buffer

[João Távora]

Hi

So yasnippet has this bit of code (simplified for this report)

    (put 'yas-expand  'function-documentation
         '(yas--expand-from-trigger-key-doc t))
    (defun yas--expand-from-trigger-key-doc (context)
      "A doc synthesizer for `yas-expand'."
      (concat "Expand a snippet before point. If no snippet
    expansion is possible, "
              (let* ((fallback (yas--keybinding-beyond-yasnippet)))
                (or (and fallback
                         (format "call command `%s'"
                                 (pp-to-string fallback)))
                    "do nothing (`yas-expand' doesn't shadow\nanything)."))))

This used to work fine and output something like

    <tab> runs the command yas-expand, which is an alias for
    `yas-expand-from-trigger-key' in `yasnippet.el'.
     
    (yas-expand &optional FIELD)
     
    Expand a snippet before point. If no snippet
    expansion is possible, call command `indent-for-tab-command'

This has stopped working in 24.3.5 since `with-help-window' started
replaced `with-output-to-temp-buffer' with
`with-temp-buffer-window'. The former just binds `standard-output' while
the latter also sets the current buffer to the *Help* buffer.

The result is that the fallback keybinding reported is always
"forward-button", which is almost always wrong.

One could either

1. revert that change (was it just a cleanup?)
2. fix/parametrize that particular behaviour of
`with-temp-buffer-window'
3. pass an extra original-buffer arg to `describe-function-1'
4. dynamically bind some new `help-original-buffer' var.

Even though a better mechanism for "fallback keybindings" is being
discussed (and by then yasnippet can get rid of its own technique, which
is half-baked but working since emacs 22), it'd be nice if the current
function-documentation trick is kept working for upcoming emacs 24.4.

This originated in https://github.com/capitaomorte/yasnippet/issues/468

http://github.com/capitaomorte/autopair does something similar and is
probably also affected, but I plan to deprecate autopair in favor of
24.4's electric-pair-mode anyway.

Thanks,
João
    

In GNU Emacs 24.3.50.1 (i686-pc-mingw32)
 of 2014-01-04 on LEG570