On 17/08/06, Adrian Howard <[EMAIL PROTECTED]> wrote:
On 16 Aug 2006, at 19:38, A. Pagaltzis wrote: [snip] > I don't know about you, but I tend to prefer a well-defined and > simple wire formats over APIs. :-) In a sense, it's declarative, > rather than imperative. Integration of heterogenous systems is > much easier that way – even subsystems for which no explicit > format emitter is available can participate without much trouble. > You can easily scale the complexity of the participants at either > end of the wire to match your needs. At one extreme of the scale, > you might telnet to your smoke server and manually tap in (pun > intended) "ok 1", "not ok 2" responses. Then you can go through > scripts which manually `print "ok 1"` through Test::More all the > way to Test::Class. Your infrastructure can be as simple or > complex as you want or need. > > I think this is a powerful benefit. You're right it can be a bonus. But to sell that to a group of people who already have something that works very well, is already integrated with their IDE, continuous integration package, build system, etc. you've got a tough job ;-)
When I was using DUnit, I was doing lots of pointer twiddling (serious fun in Dephi). If the library I was testing was broken enough it would crash, taking the whole test GUI with it. I really missed that aspect of TAP, F
