From: Phlip <phlip_cpp@...>
Date: 2004-03-17T08:24:34+09:00
Subject: Re: GUI testing on unix

> I am curious to if there exists any GUI testing packages for unix.
> What about fxruby, gnome, qt, fltk ?

Write tests in Ruby that create your forms and manipulate the controls thru
Tk's native interface.

    http://www.rubygarden.org/ruby?SvgCanvas

This is the most efficient way, except one cannot use event_generate unless
the window displays. To get around this, use bind_info to drill down into
the Proc object bound to an event, and call that directly:

            oval.bind('Button-1') { |event| doc(event) }
            puts oval.bindinfo('Button-1')[0][0].inspect()
            puts oval.bindinfo('Button-1')[0][1].inspect()
            maybeMainloop()

The bindinfo() returns an array of arrays. inspecting inside them reveals:

    #<Proc:0x4019c804>
    "%# %a %b %c %d %f %h %k %m %o %p %s %t %w %x %y %A %B %D
                                     %E %K %N %R %S %T %W %X %Y"

That tells us one leaf element of the binding info is a Ruby Proc - a handle
to a block - and the next element is a big string representing its 27
arguments. Those match the 27 members of the event object - in the order
that TkComm::initialize() used them.

So, configure the bound block again, so it prints out the value of every
member of the event - using the ordering we researched. Then click on it one
more time:

            assert_equal(oval.class.name(), 'TkcOval')

            oval.bind('Button-1') { |event|
                p event.serial
                p event.above
                p event.num
                p event.count
                p event.detail
..
                p event.type
                p event.widget
                p event.x_root
                p event.y_root
                }
#            puts oval.bindinfo('Button-1')[0][0].inspect()
#            puts oval.bindinfo('Button-1')[0][1].inspect()
            maybeMainloop(true)

All those p calls (which are secretly puts *.inspect() calls) give us enough
data to format this list of arguments:

    58, "??", 1, 0, "??", false, 0, 1, "NotifyNormal", false,
    "PlaceOnTop", 8, 114773445, 0, 80, 26, "??", 209, 1, false,
    "??", 1, "0x0", "0x0", 4, @canvas, 209, 207

Now replace the p calls inside the event handler with something testable,
and call the Proc from the program, instead of by clicking on the oval:

            assert_equal('TkcOval', oval.class.name())
            iGotClicked = false

            oval.bind('Button-1') { |event|
                iGotClicked = true
                }
            oval.bindinfo('Button-1')[0][0].call(
                58, "??", 1, 0, "??", false, 0, 1,
                "NotifyNormal", false, "PlaceOnTop", 8,
                114773445, 0, 80, 26, "??", 209, 1, false,
                "??", 1, "0x0", "0x0", 4, @canvas, 209, 207
                )
            assert(iGotClicked)
            maybeMainloop()

And that test passes. We can now write Test Fixtures that simulate user
events as close as possible to the real thing, but without going outside Tk,
or taking shortcuts. But most projects can take those shortcuts.

-- 
  Phlip
    http://www.xpsd.org/cgi-bin/wiki?TestFirstUserInterfaces