[Groupkit-users] Is anyone using Groupkit for GUI regression testing?

[William D. <wi...@ds...>]

Greetings,

I'm interested in using Groupkit as a part of a regression-testing
suite.  Since Groupkit is tracking screen activity for us already, it
should be easier to record and play back those activities.  Also, I
like the idea that if a script has problems, I can run it again in a
little window that several remote people can watch and discuss.

A stickier issue is making the playback more on the widget level than
on the level of mere x/y screen locations.  In other words, instead of
saying "move mouse to 323,445 and press button 1", the playback script
would say "activate button .frame1.button1".  This would go a long way
towards helping make the playback scripts less vulnerable to being
broken by minor changes to the program or running environment.  

That may seem like an obscure issue with regard to remote
collaboration, but really it ties into the idea of cutting down on the
amount of information that needs to be transmitted between multiple
remote collaborators.  We don't need to transmit every pixel color
because both the local and remote machine share a function that can
translate a shorter message into the same effective result (e.g. both
sides keep a cache of the previous screen).

Now extend that same idea a little, and imagine that both the local and
remote user were running the same application inside identical
"sandboxes" on each machine -- let's say a tcl applet inside a "safe"
interpreter.  As long as both applications are kept completely in sync,
I don't need to transmit a crudely-compressed bitmap of the screen area
affected by that application.  All I need to do is transmit enough
information to keep both copies of the application in sync.  In other
words, instead of having a button flash, and then transmitting a
compressed image of the flash, all I need to do is recognize that
button .frame1.button1 was pressed, and send that information to the
copy of the application on the remote machine.

The point is that there's a connection between remote collaboration and
script-driven regression testing.  In both cases, we want to distill
out the information that matters (this button was pressed), throwing
away information that we decide does not matter (the *exact* path the
mouse took while moving towards the button).  In both cases, it seems
like there's promise in approaching the issue from the viewpoint of 1)
creating functions that rely entirely on a well-defined state vector,
2) giving all the collaborators a copy of the functions and initial
state vectors, and then 3) keeping the state vectors in sync between
all the collaborators, so that the functions can generate the detailed
results locally.  This would even allow people who joined late to catch
up with everyone else, since all we have to do is give them the
complete state vector -- which is easy to do for a tcl safe interp.

Obviously this requires applications that cooperate to some extent with
groupkit, but we can still fall back on screen-scraping when all else
fails.  Also tcl is very good at introspection, so for any given pure
tcl/tk application, it may be pretty easy to convince it to work with
Groupkit with regard to transmitting state vectors, etc.

Has anyone else tried or thought about this sort of thing with regard
to Groupkit?  Or at the least, has anyone tried using Groupkit in its
present form for things like regression testing?

--Will
William L. Dye