Re: Pretest next week

[YAMAMOTO Mitsuharu]

>>>>> On Sat, 07 Mar 2009 15:28:21 +0200, Adrian Robert <address@hidden> said:

>> Still, I think Feval calls from read_socket_hook and (emulated)
>> select are really bad.  Most developers assume that these functions
>> (and some higher-level ones such as detect_input_pending) don't
>> call Feval, and they may add some code without noticing that NS
>> breaks such assumptions.

> Although I am still unsure how often in practice there is a problem
> (the tramp menu crash was not reported until now, despite the
> situation causing it on the NS end existing for years), I fully
> agree that the menu event handling under NS should be fixed.  Given
> the menu update model in emacs, the best solution is likely to defer
> the event as other ports including Carbon+AppKit have done.  The
> only thing stopping me is my lack of time right now, and a
> hesitation to implement such a change during pretest.

That's why I suggested putting off the pretest or removing the
Cocoa/GNUstep port from the pretest.

This is not the only thing.  The recent "fail on osx between 2/4/2009
and 2/5/2009" thread reminded me of several "kludges" that should be
fixed before the release:

1. Different interface for existing functionality.

  a. ns-read-file-name vs. x-file-dialog
  b. ns-drag-{file,color,text} event + own handlers vs.
     drag-n-drop event + dnd.el
  c. ns-expand-space vs. line-spacing frame parameter
  d. nsfont_make_fontset_for_font vs.
     :lang/:script/:registry properties in font-spec

2. Interface is the same, but implementation is based on own interpretation.

  a. tooltip (not being an Emacs frame, it cannot make use of Emacs
     display features such as images.)
  b. selection concepts (local/foreign selection, ownership)
  c. rightmost scroll-bar placement (it doesn't consider the case
     where scroll bars in different width, e.g., C-x 2 M-:
     (set-window-scroll-bars nil 11 t) RET)
  d. internal-border-width
  e. fringe and cursor drawing

3. NS-only implementation for features that are not inherently specific to NS.

  a. preferences panel
  b. alpha-component in color specification
  c. color image for stipple (cf. tiling patch by Miles Bader)

4. Suspected fundamental design flaw.

  a. C-g handling
  b. menu bar activation timing

                                     YAMAMOTO Mitsuharu
                                address@hidden