Fred Brooks makes a convincing argument about the importance of having a single architect (although he mentions important exceptions for having two) to design a system. I share this opinion, but having a single architect doesn't preclude incorporating feedback from many people as long as the single vision is maintained.

I think the process pg chose is fine - there's nothing like having an actual product to use and bang on to get information to refine the language. And at some point, it's important to do exactly that, so it may be a question of timing i.e. how soon to implement the language.

How do folks feel about a process where a language spec is developed (by the single visionary, not a committee) in an open forum for a period of time before a line of code is written? I realize pg wrote many essays and received much feedback, etc., but no spec was created. I wonder whether having a detailed spec would allow various experts to shoot holes in aspects of it before it is implemented. This would have the advantage of having an actual spec as a by product of the process instead of having the implementation being the spec.

I agree that you can only go so far before a prototypical compiler would have to be implemented, but I think you can go pretty far. For application development one must strike a balance between big design up front (BDUF) and seat of the pants highly iterative processes. Some domains require more BDUF and some less. So I guess my question is how much BDUF is advantageous for designing a programming language?

pg has stated backward compatibility is not necessary yet, so he's free to make major refinements to the language, but I would think it would be easier to do that before a compiler is written and code, such as news.ycombinator exists. In other words, in theory, he's free to break existing code, but I expect there is at least a little hesitancy to do that as more application code is written with the language.

Thoughts?