" It is often tempting to shortcut this by passing in some kind of option flag to existing code, rather than enabling a full parallel implementation. It is a grey area, but I have been tending to find the extra path complexity with the flag approach often leads to messing up both versions as you work, and you usually compromise both implementations to some degree."
I don't get this particular implementation detail. If it's not a flag, how would it be implemented?
It's not like you're coding it on two separate branches, since it needs to switched at runtime. It looks like the plea is for it to be pure functional, to make it easier to switch it in and out. But I can't quite picture how he's implementing this.
He means having both implementations use the same infrastructure. This is like taking your Raytracer class, and modify its Trace() method to check some global flag to determine which algorithm to use, shunting both into the Raytracer. Instead, he argues for making two totally separate implementations: a RaytracerKD and a RaytracerBST (say), each of which have identical interfaces and which you switch between at a higher level. This way your level of coupling is lower and you hopefully have fewer problems, even though it may also be more complex.
As I read it, the alternative implementation should branch of at a single point in the code. The alternative path usually requires modifications in the functions and data it's calling and using. But instead of going through the code adding flags everywhere, the idea is to make a source code copy and modify that. It duplicates source code, but keeps the original implementation untouched.
I can imagine that it allows more freedom re-working the code. You don't have to worry about breaking the original implementation this way, freeing up mental capacity.
another way he seems to have found to free mental capacity - his text is suspiciously lacking mentioning of abstractions and design patterns. Switching between implementations? There is a pattern for that!
Yeah - if only Carmack had read the Gang of Four on http://en.wikipedia.org/wiki/Design_Patterns - imagine how much more he would have accomplished! He could have used the Flyweight pattern for texels, since the same color is used many times - and that's just one example! :)
I don't get this particular implementation detail. If it's not a flag, how would it be implemented?
It's not like you're coding it on two separate branches, since it needs to switched at runtime. It looks like the plea is for it to be pure functional, to make it easier to switch it in and out. But I can't quite picture how he's implementing this.