wxruby users - Aug 2005 - New inheritance

A good number of the samples I have tested are outright or subtly broken 
on Windows now.  I''m trying to determine what the cause is.  Mostly it 
seems related to the initialize function not being created when it 
should be.

Case in point:  ClientDC.i

The caret example won''t work at all because initialize isn''t
declared
any longer.  If you comment out the import lines, Swig does create an 
initialize.  For ClientDC this fixes things so you can at least allocate 
one.

Second case:  Choice.i

This is missing an %import for wxControlWithItems.  However, if you add 
this, the constructor goes away.   It''s too hot in my house with the
air
conditioner out for me to attempt to puzzle this out later.  If cooler 
heads haven''t figured this out later I may revisit it.

Roy

Roy Sutton wrote:
> A good number of the samples I have tested are outright or subtly broken 
> on Windows now.  
Argh! You''re right. wxruby2 is pretty thoroughly broken at the moment.

The problem has to do with pure virtual functions and abstract base classes.
> Case in point:  ClientDC.i
Right. It inherits from WindowDC which inherits from DC. DC itself 
is/should be abstract because it has a large number of pure virtual 
functions.

After quite a bit of research, though, and experimentation, I think I 
see the answer:

1. Insert the word "virtual" at the beginning of each pure virtual 
method declaration. I don''t think swig actually cares, but
it''s the
right thing to do.

2. Stop generating directors for any abstract base class. Fortunately, 
there only seem to be 8 of these, and I''ve already done one of them
(DC).

3. ARGH: In any concrete subclass of an abstract base class, wherever 
the pure virtuals are actually defined, add those definitions to the .h 
files. This is ugly because it means updating MOST of the Event classes, 
MANY of the DC classes, ALL of the Sizer classes, and MOST of the 
classes that inherit from ControlWithItems. It''s also a pain because I 
either have to guess where each method is defined, or examine each real 
wx .h file.

Having "fixed" WindowDC, the caret sample works again. So at least
there
is hope again.
> It''s too hot in my house with the air 
> conditioner out for me to attempt to puzzle this out later.  If cooler 
> heads haven''t figured this out later I may revisit it.
You have my sympathies, especially if you''re someplace really hot like 
me (Florida). Fortunately, my A/C is working fine, so my cool brain was 
functioning up to spec tonight.

Thanks,

Kevin

Kevin Smith wrote:
> 
> Argh! You''re right. wxruby2 is pretty thoroughly broken at the
moment.
> 
> After quite a bit of research, though, and experimentation, I think I 
> see the answer:
> 
> 1. Insert the word "virtual" at the beginning of each pure
virtual
> method declaration. I don''t think swig actually cares, but
it''s the
> right thing to do.
Done (I think).
> 2. Stop generating directors for any abstract base class. Fortunately, 
> there only seem to be 8 of these, and I''ve already done one of
them (DC).
Done (I think).
> 3. ARGH: In any concrete subclass of an abstract base class, wherever 
> the pure virtuals are actually defined, add those definitions to the .h 
> files. This is ugly because it means updating MOST of the Event classes, 
> MANY of the DC classes, ALL of the Sizer classes, and MOST of the 
> classes that inherit from ControlWithItems. It''s also a pain
because I
> either have to guess where each method is defined, or examine each real 
> wx .h file.
Done (I think).

At this point, wxruby2 is again able to run the samples that I tried 
(that it ran before). So I think it is "unbroken" again. I have tagged
this as 0.0.28.

Kevin

Didn''t test against this latest checkin but just in case you missed it 
here''s a patched wxRadioBox.h

Kevin Smith wrote:
> Kevin Smith wrote:
>
>>
>> Argh! You''re right. wxruby2 is pretty thoroughly broken at the
moment.
>>
>> After quite a bit of research, though, and experimentation, I think I 
>> see the answer:
>>
>> 1. Insert the word "virtual" at the beginning of each pure
virtual
>> method declaration. I don''t think swig actually cares, but
it''s the
>> right thing to do.
>
>
> Done (I think).
>
>> 2. Stop generating directors for any abstract base class. 
>> Fortunately, there only seem to be 8 of these, and I''ve
already done
>> one of them (DC).
>
>
> Done (I think).
>
>> 3. ARGH: In any concrete subclass of an abstract base class, wherever 
>> the pure virtuals are actually defined, add those definitions to the 
>> .h files. This is ugly because it means updating MOST of the Event 
>> classes, MANY of the DC classes, ALL of the Sizer classes, and MOST 
>> of the classes that inherit from ControlWithItems. It''s also a
pain
>> because I either have to guess where each method is defined, or 
>> examine each real wx .h file.
>
>
> Done (I think).
>
> At this point, wxruby2 is again able to run the samples that I tried 
> (that it ran before). So I think it is "unbroken" again. I have
tagged
> this as 0.0.28.
>
> Kevin
> _______________________________________________
> wxruby-users mailing list
> wxruby-users@rubyforge.org
> http://rubyforge.org/mailman/listinfo/wxruby-users
>
>
>
-------------- next part --------------
--- wxruby2_new/swig/classes/include/wxRadioBox.h	2005-01-30 22:18:52.000000000
-0500
+++ wxruby2/swig/classes/include/wxRadioBox.h	2005-08-22 00:08:53.334144000
-0400
@@ -56,7 +56,7 @@
 	 * \param bool  
 	*/
 
-  void Enable(bool enable = true) ;
+  virtual bool Enable(bool enable = true) ;
 	/**
 	 * \brief Enables or disables an individual button in the radiobox.
 
@@ -74,7 +74,7 @@
 	 * \param bool  
 	*/
 
-  void Enable(int  n , bool enable = true) ;
+  virtual bool Enable(int  n , bool enable = true) ;
 	/**
 	 * \brief Finds a button matching the given string, returning the position if
found, or
 -1 if not found. 
@@ -86,7 +86,11 @@
 	 * \brief Returns the number of items in the radiobox. 
 	*/
 
-  int GetCount() const;
+  virtual int GetCount() const;
+
+  virtual int GetColumnCount() const;
+  virtual int GetRowCount() const;
+
 	/**
 	 * \brief Returns the radiobox label. 
 	*/
@@ -102,7 +106,7 @@
 	 * \brief Returns the zero-based position of the selected button. 
 	*/
 
-  int GetSelection() const;
+  virtual int GetSelection() const;
 	/**
 	 * \brief Returns the selected string. 
 	*/
@@ -149,14 +153,14 @@
 	 * \param int  
 	*/
 
-  void SetSelection(int  n ) ;
+  virtual void SetSelection(int  n ) ;
 	/**
 	 * \brief Sets the selection to a button by passing the desired string. This
does not cause
 a wxEVT_COMMAND_RADIOBOX_SELECTED event to get emitted. 
 	 * \param const wxString&   
 	*/
 
-  void SetStringSelection(const wxString&  string ) ;
+  virtual void SetStringSelection(const wxString&  string ) ;
 	/**
 	 * \brief Shows or hides the entire radiobox. 
 	 * \param const bool  
@@ -180,13 +184,13 @@
 	 * \param const bool  
 	*/
 
-  void Show(int  item , const bool  show ) ;
+  virtual void Show(int  item , const bool  show = true ) ;
 	/**
 	 * \brief Returns the label for the button at the given position. 
 	 * \param int  
 	*/
 
-  wxString GetString(int  n ) const;
+  virtual wxString GetString(int  n ) const;
 };

Just checked it out and it compiles fine on OS X but when the samples
are run no windows are shown. I will look into it a bit to see if
anything obvious jumps out at me.

Sean

On 8/21/05, Kevin Smith <wxruby@qualitycode.com>
wrote:
> Kevin Smith wrote:
> >
> > Argh! You''re right. wxruby2 is pretty thoroughly broken at
the moment.
> >
> > After quite a bit of research, though, and experimentation, I think I
> > see the answer:
> >
> > 1. Insert the word "virtual" at the beginning of each pure
virtual
> > method declaration. I don''t think swig actually cares, but
it''s the
> > right thing to do.
> 
> Done (I think).
> 
> > 2. Stop generating directors for any abstract base class. Fortunately,
> > there only seem to be 8 of these, and I''ve already done one
of them (DC).
> 
> Done (I think).
> 
> > 3. ARGH: In any concrete subclass of an abstract base class, wherever
> > the pure virtuals are actually defined, add those definitions to the
.h
> > files. This is ugly because it means updating MOST of the Event
classes,
> > MANY of the DC classes, ALL of the Sizer classes, and MOST of the
> > classes that inherit from ControlWithItems. It''s also a pain
because I
> > either have to guess where each method is defined, or examine each
real
> > wx .h file.
> 
> Done (I think).
> 
> At this point, wxruby2 is again able to run the samples that I tried
> (that it ran before). So I think it is "unbroken" again. I have
tagged
> this as 0.0.28.
> 
> Kevin
> _______________________________________________
> wxruby-users mailing list
> wxruby-users@rubyforge.org
> http://rubyforge.org/mailman/listinfo/wxruby-users
>

I still have not gotten the OS X version working completely. I have
found that Dialog based windows work just fine it is Frame based ones
that are not displaying.

Kevin and Roy do you remember changing anything durring the
inheritance overhaul that would affect this? I only see this problem
on OS X it runs fine on Windows.

Sean

On 8/21/05, Sean Long <sean.m.long@gmail.com>
wrote:
> Just checked it out and it compiles fine on OS X but when the samples
> are run no windows are shown. I will look into it a bit to see if
> anything obvious jumps out at me.
> 
> Sean
> 
> On 8/21/05, Kevin Smith <wxruby@qualitycode.com> wrote:
> > Kevin Smith wrote:
> > >
> > > Argh! You''re right. wxruby2 is pretty thoroughly broken
at the moment.
> > >
> > > After quite a bit of research, though, and experimentation, I
think I
> > > see the answer:
> > >
> > > 1. Insert the word "virtual" at the beginning of each
pure virtual
> > > method declaration. I don''t think swig actually cares,
but it''s the
> > > right thing to do.
> >
> > Done (I think).
> >
> > > 2. Stop generating directors for any abstract base class.
Fortunately,
> > > there only seem to be 8 of these, and I''ve already done
one of them (DC).
> >
> > Done (I think).
> >
> > > 3. ARGH: In any concrete subclass of an abstract base class,
wherever
> > > the pure virtuals are actually defined, add those definitions to
the .h
> > > files. This is ugly because it means updating MOST of the Event
classes,
> > > MANY of the DC classes, ALL of the Sizer classes, and MOST of the
> > > classes that inherit from ControlWithItems. It''s also a
pain because I
> > > either have to guess where each method is defined, or examine
each real
> > > wx .h file.
> >
> > Done (I think).
> >
> > At this point, wxruby2 is again able to run the samples that I tried
> > (that it ran before). So I think it is "unbroken" again. I
have tagged
> > this as 0.0.28.
> >
> > Kevin
> > _______________________________________________
> > wxruby-users mailing list
> > wxruby-users@rubyforge.org
> > http://rubyforge.org/mailman/listinfo/wxruby-users
> >
>

Sean,

Before I corrected some of the inheritance problems the windows would 
not open on Windows either.  I didn''t investigate it exactly as the 
problems were pretty widespread at first.  It could be some issue still 
related to the inheritance.  You might want to compare the Mac vs. 
Windows tree.

Roy

Sean Long wrote:
>I still have not gotten the OS X version working completely. I have
>found that Dialog based windows work just fine it is Frame based ones
>that are not displaying.
>
>Kevin and Roy do you remember changing anything durring the
>inheritance overhaul that would affect this? I only see this problem
>on OS X it runs fine on Windows.
>
>Sean
>
>On 8/21/05, Sean Long <sean.m.long@gmail.com> wrote:
>  
>
>>Just checked it out and it compiles fine on OS X but when the samples
>>are run no windows are shown. I will look into it a bit to see if
>>anything obvious jumps out at me.
>>
>>Sean
>>
>>On 8/21/05, Kevin Smith <wxruby@qualitycode.com> wrote:
>>    
>>
>>>Kevin Smith wrote:
>>>      
>>>
>>>>Argh! You''re right. wxruby2 is pretty thoroughly broken
at the moment.
>>>>
>>>>After quite a bit of research, though, and experimentation, I
think I
>>>>see the answer:
>>>>
>>>>1. Insert the word "virtual" at the beginning of each
pure virtual
>>>>method declaration. I don''t think swig actually cares,
but it''s the
>>>>right thing to do.
>>>>        
>>>>
>>>Done (I think).
>>>
>>>      
>>>
>>>>2. Stop generating directors for any abstract base class.
Fortunately,
>>>>there only seem to be 8 of these, and I''ve already done
one of them (DC).
>>>>        
>>>>
>>>Done (I think).
>>>
>>>      
>>>
>>>>3. ARGH: In any concrete subclass of an abstract base class,
wherever
>>>>the pure virtuals are actually defined, add those definitions to
the .h
>>>>files. This is ugly because it means updating MOST of the Event
classes,
>>>>MANY of the DC classes, ALL of the Sizer classes, and MOST of
the
>>>>classes that inherit from ControlWithItems. It''s also a
pain because I
>>>>either have to guess where each method is defined, or examine
each real
>>>>wx .h file.
>>>>        
>>>>
>>>Done (I think).
>>>
>>>At this point, wxruby2 is again able to run the samples that I tried
>>>(that it ran before). So I think it is "unbroken" again. I
have tagged
>>>this as 0.0.28.
>>>
>>>Kevin
>>>_______________________________________________
>>>wxruby-users mailing list
>>>wxruby-users@rubyforge.org
>>>http://rubyforge.org/mailman/listinfo/wxruby-users
>>>
>>>      
>>>
>
>_______________________________________________
>wxruby-users mailing list
>wxruby-users@rubyforge.org
>http://rubyforge.org/mailman/listinfo/wxruby-users
>
>
>
>  
>

Roy Sutton wrote:
> Didn''t test against this latest checkin but just in case you
missed it
> here''s a patched wxRadioBox.h
I had to work with this a while before it would compile for me. I ended 
up removing GetColumns and GetRows because they don''t exist on my 
system, and aren''t in the wx docs.

I also had to split Show into Show and ShowItem, like wxPython has. 
Hopefully what I did will still work on the other platforms.

By the way, my earlier pass was to catch all the pure virtual methods, 
not all the virtual methods. There are probably quite a few other 
methods that should be marked virtual but still aren''t.

Kevin

Sean Long wrote:
> I still have not gotten the OS X version working completely. I have
> found that Dialog based windows work just fine it is Frame based ones
> that are not displaying.
> 
> Kevin and Roy do you remember changing anything durring the
> inheritance overhaul that would affect this? I only see this problem
> on OS X it runs fine on Windows.
Ugh. No, I don''t remember anything specific. I did have a somewhat 
similar problem though, where the new bigdemo sample just would not show 
its main window, even though the code looked fine. Minimal and nothing 
did come up, though. So I copied their app code over piece by piece 
until the new bigdemo code started working. I still couldn''t see what 
was different between the working code and the non-working code.

To try to debug it, I inserted several printfs directly into App.cpp, to 
try to figure out what was happening. In my case, the window was being 
created, but for some reason was not being recognized by wx as being 
"the" top window, and therefore as soon as the mainloop was entered,
wx
exited because it thought the top window didn''t exist.

Best guess: the inheritance changes have caused some virtual method to 
not get called that used to be called, or vice versa. Or (as Roy hinted) 
that the parentage under Mac is different enough from Linux and Windows 
to cause problems. One would think that the issue would be with Frame.

Just to confirm: Even samples/minimal/nothing.rb doesn''t come up?

Kevin

> To try to debug it, I inserted several printfs directly into App.cpp, to
> try to figure out what was happening. In my case, the window was being
> created, but for some reason was not being recognized by wx as being
> "the" top window, and therefore as soon as the mainloop was
entered, wx
> exited because it thought the top window didn''t exist.
Will have to try that, thanks.

> Just to confirm: Even samples/minimal/nothing.rb doesn''t come up?
Nope, does not show a Frame.

Sean

I finally got Frames working again in OS X, I am not sure why it
worked but it does. I had to remove

%import "include/wxTopLevelWindow.h"

from Frame.i

I am not sure if this should have #ifndef __WXMAC__ around it or not,
can someone test it on another platform? I can test it on Windows
tomorrow.

Sean


On 8/24/05, Sean Long <sean.m.long@gmail.com>
wrote:
> > To try to debug it, I inserted several printfs directly into App.cpp,
to
> > try to figure out what was happening. In my case, the window was being
> > created, but for some reason was not being recognized by wx as being
> > "the" top window, and therefore as soon as the mainloop was
entered, wx
> > exited because it thought the top window didn''t exist.
> 
> Will have to try that, thanks.
> 
> 
> > Just to confirm: Even samples/minimal/nothing.rb doesn''t come
up?
> 
> Nope, does not show a Frame.
> 
> Sean
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: frame.patch
Type: application/octet-stream
Size: 515 bytes
Desc: not available
Url :
http://rubyforge.org/pipermail/wxruby-users/attachments/20050828/e3cd40ec/frame.obj

Sean Long wrote:
> I finally got Frames working again in OS X, I am not sure why it
> worked but it does. I had to remove
That''s great news.
> 
> %import "include/wxTopLevelWindow.h"
> 
> from Frame.i
But I don''t like that solution. It not only removes virtual-ness from 
all the TopLevelWindow methods in Frame, but it also cuts out all the 
Window, EvtHandler, and Object virtuals. If we are going to cut out 
TopLevel entirely, the proper way to do it would be to change wxFrame.h 
to declare:

class wxFrame : public wxWindow

With that change, it should still work on Mac. Can you test that? If 
not, then we know the problem isn''t actually in TopLevelWindow, but in 
one of the classes higher in the inheritance chain.

If we confirm that the problem is in TopLevelWindow, it''s quite
possible
that we wouldn''t have to remove it entirely. The two most likely
virtual
methods in TopLevelWindow that might be causing problems are IsActive 
and IsTopLevel. So the next thing to try would be to restore the 
wxTopLevelWindow parentage in wxFrame.h, and then #if out just those two 
methods from wxTopLevelWindow.h.

You can check src/Frame.h to confirm that those methods no longer appear 
as virtuals.

Thanks!

Kevin

> class wxFrame : public wxWindow
> 
> With that change, it should still work on Mac. Can you test that? If
> not, then we know the problem isn''t actually in TopLevelWindow,
but in
> one of the classes higher in the inheritance chain.
Just tried inheriting from wxWindow and it does not work. Which makes
sense because this problem existed before I added wxTopLevelWindow to
wxruby2.

Any ideas on where to go from here?

Sean

Sean Long wrote:
>>class wxFrame : public wxWindow
>>
>>With that change, it should still work on Mac. Can you test that? If
>>not, then we know the problem isn''t actually in TopLevelWindow,
but in
>>one of the classes higher in the inheritance chain.
> 
> 
> Just tried inheriting from wxWindow and it does not work. Which makes
> sense because this problem existed before I added wxTopLevelWindow to
> wxruby2.
> 
> Any ideas on where to go from here?
Oh. Try inheriting from wxEvtHandler then. If that doesn''t work, try 
inheriting from wxObject. We want to try to isolate which exact 
method(s) are causing problems when virtualized on the Mac.

Thanks,

Kevin

I got it to work. I had to add:
bool Show(bool show = true);

to wxFrame.h

the code for Show(bool show) in src/wxFrame.cpp looks like this
without the Show signature added to wxFrame:

bool SwigDirector_wxFrame::Show(bool show) {
    VALUE obj0 = Qnil ;
    bool c_result ;
    VALUE result;
    
    if (swig_get_up()) {
        return wxWindow::Show(show);
    }
    obj0 = show ? Qtrue : Qfalse;
    result = rb_funcall(swig_get_self(), rb_intern("show"), 1,obj0);
    c_result = (bool) RTEST(result);
    return (bool) c_result;
}

and like this with the signature added:

bool SwigDirector_wxFrame::Show(bool show) {
    VALUE obj0 = Qnil ;
    bool c_result ;
    VALUE result;
    
    if (swig_get_up()) {
        return wxFrame::Show(show);
    }
    obj0 = show ? Qtrue : Qfalse;
    result = rb_funcall(swig_get_self(), rb_intern("show"), 1,obj0);
    c_result = (bool) RTEST(result);
    return (bool) c_result;
}

Comparing the wxWidgets headers for MSW and Mac I can not see where
this would make a difference, but it obviously does in the case of the
Mac.

I am not going to submit a patch, if you think it is an appropriate
fix I will let you make the change.

Sean

On 8/29/05, Kevin Smith <wxruby@qualitycode.com>
wrote:
> Sean Long wrote:
> >>class wxFrame : public wxWindow
> >>
> >>With that change, it should still work on Mac. Can you test that?
If
> >>not, then we know the problem isn''t actually in
TopLevelWindow, but in
> >>one of the classes higher in the inheritance chain.
> >
> >
> > Just tried inheriting from wxWindow and it does not work. Which makes
> > sense because this problem existed before I added wxTopLevelWindow to
> > wxruby2.
> >
> > Any ideas on where to go from here?
> 
> Oh. Try inheriting from wxEvtHandler then. If that doesn''t work,
try
> inheriting from wxObject. We want to try to isolate which exact
> method(s) are causing problems when virtualized on the Mac.
> 
> Thanks,
> 
> Kevin
> _______________________________________________
> wxruby-users mailing list
> wxruby-users@rubyforge.org
> http://rubyforge.org/mailman/listinfo/wxruby-users
>

Ah! Good catch! We will probably find a few other cases like this, where 
the wx docs don''t mention that classes are actually overriding methods 
defined in their superclasses.

Most likely, wxFrame::Show is what sets the wx top window, so it makes 
sense that if we are directly invoking wxWindow::Show (and therefore 
bypassing wxFrame::Show), it wouldn''t work properly.

Committed. Thanks!

Kevin



Sean Long wrote:
> I got it to work. I had to add:
> bool Show(bool show = true);
> 
> to wxFrame.h
> 
> the code for Show(bool show) in src/wxFrame.cpp looks like this
> without the Show signature added to wxFrame:
> 
> bool SwigDirector_wxFrame::Show(bool show) {
>     VALUE obj0 = Qnil ;
>     bool c_result ;
>     VALUE result;
>     
>     if (swig_get_up()) {
>         return wxWindow::Show(show);
>     }
>     obj0 = show ? Qtrue : Qfalse;
>     result = rb_funcall(swig_get_self(), rb_intern("show"),
1,obj0);
>     c_result = (bool) RTEST(result);
>     return (bool) c_result;
> }
> 
> and like this with the signature added:
> 
> bool SwigDirector_wxFrame::Show(bool show) {
>     VALUE obj0 = Qnil ;
>     bool c_result ;
>     VALUE result;
>     
>     if (swig_get_up()) {
>         return wxFrame::Show(show);
>     }
>     obj0 = show ? Qtrue : Qfalse;
>     result = rb_funcall(swig_get_self(), rb_intern("show"),
1,obj0);
>     c_result = (bool) RTEST(result);
>     return (bool) c_result;
> }
> 
> Comparing the wxWidgets headers for MSW and Mac I can not see where
> this would make a difference, but it obviously does in the case of the
> Mac.
> 
> I am not going to submit a patch, if you think it is an appropriate
> fix I will let you make the change.
> 
> Sean