sequencer with adjustable step lenght for each step

This is an R5 ens by Dieter Zobel

If you want to use speed as gate length you need a Delay or a Hold module, IMHO.

If it is too short use a multiplier.

Few more questions and @colB is going to develop The Ultimate Step Sequencer :-D

I still can't figure out how it's possible, that the whole sequencer is one huge feedback loop without a single Unit Delay. Every time I try to do something similar, Reaktor crashes.

If you look carefully, those feedback lines are connected to ports that are not immediately 'live'. The set input of the counter will not generate an output through the separator. Try setting a couple of the inputs to the step size selector to zero - those steps are now jumped, because the event goes around the loop until it hits a non zero value... if you set them all to zero, Reaktor will crash because the same event keeps going, never reaching an ending point... But because they are always set to 1 or more, this never happens - the loop always exits immediately at the separator.

The feedback that increments the main sequencer feeds to the input port of a value module, not the trigger port - so this is acting something like a latch or z^-1 in core... almost like a unit delay for events...

This is where Primary differs the most from core - with Primary, each event is processed to the end of its chain before the next event is handled, so if you duplicate an even using an order module (or just take multiple outs from the same port), then the first will be processed completely to it's ultimate destination before the next starts it's journey (that's why you should always use order modules, because you can't always guarantee what order Reaktor will process multiple connections from the same output port). And as long as you guarantee a logical end for each event's processing chain, there won't be a lockup/infinite loop.

So in primary, it's possible to use feedback directly in many situations, as long as it makes sense. An example would be a Snap Value. You can have a feedback containing a Snap Value if you turn off its Write Thru parameter. That way, it only sends an output when a snapshot is loaded. So it looks like a feedback loop, visually, but it isn't really feedback, because mostly its just storing a new value - it only outputs during initialisation, one time, so there is no implicit infinite loop.

Core uses a breadth first approach, all events that originate from the same even will be processed with 'logical simultaneity' - in other words as though they were being handled at the same time (not really possible, but virtually). This only breaks down with certain processes using OBC memory, and in those cases, there is a module and some other techniques to force any desired order.

The consequence of this in core is that it's impossible to have feedback without a unit delay, so if you don't add one, an implicit one is added for you.

Mostly core is much more logical, and easier to work with. But for some things Primary is more elegant - like this simple sequencer.

Creating a complex more feature complete sequencer is significantly easier in core IMO.