Re: delwin() and deleting windows with subwindows

[Bill Gray]

On 8/10/22 20:08, Thomas Dickey wrote:
> it depends: in the SVr4 code, child-windows have a link to their parent,
> using that to maintain a reference count -- and tell it if there are no
> remaining children, and return an error if any remain.  But that's not
> enough to just delete all of the children and the parent window, which
> would make it nice(r) for applications.

   Well,  yes... nicer,  and apparently we _should_ recursively delete 
to follow the XSI Curses standard ("It [delwin] must delete subwindows 
before deleting their parent.")

   The only downside I see to recursive deletion would be if you did 
the following :

   WINDOW *win = newwin( ...);
   WINDOW *sub = subwin( win, ...);

   delwin( win);
   access something in sub;

   At present,  the delwin() call would fail in ncurses and SVr4.  In 
PDCurses and Solaris X/Open curses,  win would be deleted and sub 
wouldn't.  In either case,  attempts to access something in sub will 
still work.

   Delete subwindows recursively,  and that access attempts to read 
freed memory.  One shouldn't be doing such things;  once a parent window 
is deleted,  you shouldn't try to do anything with the subwindow.  But 
I'll bet you that some code does so,  and recursive deletion would break 
that code.

   I've therefore reluctantly decided against recursive subwindow 
deletion for PDCursesMod,  and have gone back to having delwin() just 
return ERR if you try to delete a parent window or a window that wasn't 
created by the library (just pushed that change to GitHub).  It's not as 
nice,  and doesn't follow the XSI Curses standard.  But it won't break 
existing code.  And at least it's an improvement over PDCursesMod's 
previous behavior of simply freeing parent windows without checking for 
subwindows first...

-- Bill