GUI output testing

Colin Putney cputney at wiresong.ca
Thu Aug 7 07:40:17 UTC 2003


On Wednesday, August 6, 2003, at 10:50 PM, Richard A. O'Keefe wrote:

> I thought I had made it clear that this is a *testing* tool and it
> is an important feature, as a testing tool that it NOT save enough
> information for accurate reconstruction.

Yeah, I caught that message after my post. I was confused because 
serializing morphs to a text format seems like a very awkward way to 
test them. I thought perhaps you had something else in mind.

> Run diff(1) and other file comparison tools.
> Stuff the dump into an XML library that implements XPath.
> Keep the baselines for regression testing *out* of the image.
> Give me something that _can't_ crawl across the screen spinning
> and croaking while my checking code is trying to look at it.
> And so on.

Again, this seems very awkward. Doing textual comparisons will be very 
fragile. Diff will only be able to tell you if something has changed, 
not what the change was, or whether it's significant. Parsing out the 
XML and examining a DOM might be a bit better, but you still won't be 
able to test the UI's behaviour in response to user input.

Why do you want to keep the tests out of the image? As I said in my 
reply to your first post, you can create morphs and test them without 
displaying them on screen. And surely writing SUnit tests is preferable 
to writing them in Java or C?

Colin



More information about the Squeak-dev mailing list