[Seaside] Re: Web app Testing frameworks

[Avi Bryant]

On Jun 7, 2004, at 12:26 AM, Julian Fitzell wrote:

> Just entered the application from the main entry point.  We were using 
> it for load testing so it definitely didn't make any difference for 
> us.  You basically record tests by just using the application and then 
> play them back again.  It has a weighting system to grab the right 
> things and can use the same system (the weightings can be tweaked) to 
> validate pages if you want it to.

A nice generic approach might be to have someone perform the same 
conceptual action in the application multiple times ("training" rather 
than just "recording"), and have the testing system try to come up with 
a minimal rule that would properly capture the user's interaction.  At 
a particular step, this rule might be something like "click one of the 
links inside the <ul> inside a <div> with id='navbar'".  There would be 
two interesting aspects to this: one is that by seeing the same action 
in multiple sessions, the testing system ought to be able to figure out 
that things like session and continuation ids are highly variable and 
need to be ignored.  The other is that if the user intentionally varies 
what they do during the recording phase, the testing system can notice 
that and randomize the playback to match (in the above example, 
choosing a different link inside that <ul> every time).

Maybe this is too optimistic: the rules might end capturing the wrong 
thing all the time ("fourth link on the page" rather than "link with 
text 'Logout'").  But the domain seems narrow enough that you could 
hopefully make it pretty smart about which properties are likely to be 
stable and which aren't.  At any rate, it would be a good research 
project to find out.

Avi