Re: GTK program testing

["Timothy M. Shead" <tshead k-3d com>]

Glenn Hutchings wrote:

> Hi there, GTK fans!
>
> I'm considering developing a GTK testing function that would 
> "stress" a GUI by sending signals to widgets at random, to 
> try to detect combinations of events that lead to an error or 
> inconsistent state.  The signals would simulate actions that 
> could be performed by users (e.g. button presses, entering of 
> text into entry widgets, selection of clist entries, etc).  I 
> have some questions that the GTK community can probably help 
> me with:
>
> * Does such a thing already exist?  I seem to remember seeing 
> some related functions at GDK level to record/playback 
> events, but they seem to have gone away in recent versions of 
> GTK.
>
> * How difficult would it be?  I envisage having a timer/idle 
> function running which scans the widget hierarchy (getting 
> all the toplevel windows from gtk_container_get_toplevels()) 
> and selecting an "appropriate" widget at random to signal 
> ("appropriate" meaning a widget that the user could normally 
> interact with -- i.e. visible, active, etc).  Would there 
> need to be code for each widget type, or can signalling be 
> made more generic?
>
> * How useful would it be?  I think there would have to be 
> some way of recording/playing back the signals that led up to 
> a failure, or tracking down bugs might be difficult.
>
> I'm sure there are lots more problems I haven't thought of.  
> But what are they?

The tough part in creating the kind of pseudorandom input that you'd 
need is defining the constraints that make it "sort of" reasonable ... I 
can envision that a simple generation of random signals to random 
widgets would tie GTK+ up in knots, since some widgets are probably 
going to expect that events happen in a certain order.

The SDPGTK library (http://www.k-3d.com), which is a C++ binding for 
GTK+, has considerable support for synthesizing user input ... though 
primarily for the purpose of providing recordable interactive tutorials 
for K-3D, a 3D modelling and animation application.  A user can "turn 
on" recording, and user-interface events are recorded for later playback 
by the program's scripting engine; however, the recorded events are 
high-level events, such as "button X pressed", rather than GTK+ signals. 
   If nothing else, the code may give you some suggestions on how to 
proceed.

The recorded tutorials in K-3D make excellent regression-testing tools, 
so, if you can solve the problem of generating sensible input, you may 
be on to something.

Cheers,
Timothy M. Shead