[innguest]

Thank you for the apt metaphor. I often struggle with explaining how I feel about this but you nailed it.

We spend too much time getting these tiny pieces to connect correctly, and more often than not they're misaligned slightly or in ways we can't see yet (until it blows up). It's way too much micro for not enough macro. And there isn't even a way to "sketch" as I can with drawing - you know, loosely define the shape of something before starting to fill in with the details. I keep trying to think of a visual way to "sketch" a program so that it would at least mock-run at first, and then I could fill in the function definitions. I don't mean TDD.

[tracker1]

Sorry, but UML provides enough definition to accurately describe any process... you generally need two graphs for any interaction though. Most people don't think in these terms. Some can keep it in their heads, others cannot.

It's easier for most people to reason about a simple task. Not everyone can also see how that simple task will interact with the larger whole. I've seen plenty of developers afraid to introduce something as fundamental as a message queue, which is external to a core application. It breaks away from what they can conceptualize in a standing application.

Different people have differing abilities, talent and skills regarding thought and structure. Less than one percent of developers are above average through most of a "full stack" for any given application of medium to high complexity (percent pulled out of my rear). It doesn't mean the rest can't contribute. But it does mean you need to orchestrate work in ways that best utilize those you have available.

[innguest]

> Sorry, but UML provides enough definition to accurately describe any process...

Which is the opposite of sketching, which means (in drawing) to loosely define the basic shapes of something before accurately describing things with "enough definition".

We're not talking about the same thing at all and I can't relate to the rest of your post.

I'm talking about how there's no way to sketch a program that mock-runs and then fill in the details in a piecemeal fashion.

[minthd]

One way to partially do this ,by using an auto generation as far as possible. Naked objects(for complex business software) takes this to the extreme - given a set of domain objects, it creates the rest, including GUI. Than if for example, you want to customize the GUI ,you can.

I think it even makes sense at start to create uncomplete objects, run the mock, as continue. Also, the authors of the framework claim that you work very close to the domain language s you can discuss code with domain experts, so in a sense you're working relatively close to the spec level.

The learning curve is a bit high though, i think it could have been much more successful if it has used GUI tools to design domain objects and maybe a higher level language,