I've been playing with Jamie Bullock's 'a_grain' lately (see http://www.puredata.org/Members/jb/a_grain%7E/view ), and in order to understand it better, I've been refactoring it.
A_grain has 14 inputs to control various parameters; my first approach to cleaning it up was to put all the inlets, in the correct order, at the top of the patch -- I then connected those inlets to 'send' objects with $0 variables, placing matching 'receive's close by where they are needed. This really cleaned up the wiring quite a bit, and made it easier to "read".
Now it occurs to me that I could eliminate the inlets entirely, and just write to send/receive pairs directly (perhaps also passing in a "prefix" as an argument that is prepended to all receives inside the abstraction, which would allow multiple instantiations of the abstraction, with independent control of each). At the UI-level patch, I could use named senders (from number boxes, sliders, whatever) just hovering nearby the a_grain abstraction; no wires, no mess.
I'm wondering what experienced PD architects consider the best practice here; if the second approach is better, I begin to question the advisability of wired inlets for more than two or three arguments. The left-to-right ordering of them, along with the rats-nest wiring caused by high numbers of inputs, seem to argue against them. The only downside I can see to the second method is that if it's not done neatly, i.e., the senders are placed indiscriminately and not necessarily near the abstraction they're sending to -- it could become very hard to understand/maintain the patch.
I'll be interested to hear other PD user's thoughts on this.
Phil Stone UC Davis