Somehow I missed your post.chrisb wrote:I have this model for a car's door lock, which I would like to discuss here in order to learn how to improve stability in terms of parametrization. And so I would like to hear your recommendations how to achieve this task.
It has a linear dependency graph and is supposed to be fully parametric. However, it brakes very easyly, even if I make only tiny changes.
The topological naming issue remains in PartDesignNext, nothing has changed on that point, and unless many developers join ezzieyguywuf's work and combine all their energies on the problem (which is not realistic IMHO), I doubt anything will change by the time 0.17 is released sometime next year. But astute use of PartDesignNext's new stuff and possibly expressions and spreadsheet will help reduce the effects of topological naming problems. I believe it could be bypassed entirely if not for procedural features such as Chamfer, Draft and Fillet. At least in v0.17, these 3 types of features now can be edited to change edge/face selection, so if a PartDesign Fillet fails you can fix it rather than delete and redo it again.
Basically, you need to stop mapping sketches to faces - entirely! Place your sketches on standard planes, or on custom datum planes. And when creating datum geometry, do not base it on the part topology, base it on standard planes/axes and/or sketches. I think it's also worth pursuing a "master sketch" approach, since you can now create external geometry linked to an other sketch. I assume (maybe wrongly) that sketches may not suffer as much from the topological naming issue.
For really complex models, it may also be wise to work with separate bodies, that you'll combine at the end into a single part. Use of Shapebinders will allow links between them, but it may induce instability.
So, forget the nice linear dependency graph. It's what causes the models to break in the first place.
One thing that is still true: leave fillets, chamfers and drafts for the very end. In your doorlock.FCStd file, the second feature in the tree is a chamfer and toward the bottom of the tree you mixed a pocket with fillets. That's bad practice.
Also, I would not add the holes into the very first sketch. This adds a lot more faces to the first Pad, and consequently raises the chances that the face numbering could change after edits. In parametric design, it is considered good practice to first work on the basic shape of the model, then add holes, and finally add chamfers and fillets.
But all this is theoretical, it needs to be put to the test.