My biggest gripe with freecad isn't necessarily that (it's easy enough to work around), but that going back few steps to edit things and apply it forward is very fragile. There are ideas floating around to better handle this workflow but we'll have to see how well that goes (it's probably a very difficult problem to solve with no general solution). In the meantime this gives me a lot of frustration and rework, but, running on Linux, there's no real alternative as far as I can see.
You make a base sketch and pad it up to create a square part with a single hole in the centre. You chamfer the hole. You change the base sketch so that it is now a rectangular part with 2 holes, 1 at the left and 1 at the right.
Should the chamfer apply to the left hole, the right hole, or both holes?
Obviously there are special cases that FreeCAD doesn't currently handle optimally, but a general solution is impossible without FreeCAD reading your mind.
(You could also asy "any of those 3 options is better than trying to apply the chamfer to a random bit of un-chamferable geometry". But a general solution is still impossible.).
Any changes to the first hole only apply to the first hole, unless that second hole is a linked copy of the first.
____
Is there an argument against this general solution?
IIRC (not testing now) FreeCAD already handles this example correctly. You can add circles to the sketch and the chamfer stays put. If you delete the circle and add a new one, it appears on the new hole. But if you add a square instead, it moves to a random edge of the rectangular hole.
The edge of the rectangular hole is actually a loop of four edges, so this is understandable. Commercial software can fix some of these cases - I don't think the Freecad link branch can, but it does at least detect that the original geometry was deleted and raises an error asking you to reselect instead of allowing the chamfer to move. Not dissimilar UX to when you mess with something that Fusion can't handle, I say.
I'm not even sure FreeCAD's editing model is anything like that (on the surface it seems that it understands geometries but disregards the process to obtain them).
Changing the hole from a circle to a square on mainline will completely screw things up though.
The goal of a well-written program is to give the user the opportunity to express what is in their mind at an appropriate time, and to change their mind after seeing some output. And to be told explicitly or implicitly why their inputs don't work.
FreeCAD tends to get wedged so you can't back up, very unpredictably (for a newbie). This makes all its other issues worse because it inhibits experimentation.
It's a lot of work, but often less work than starting from scratch.
What I mean is when you start getting errors that say things are invalid and every time you try to correct it flips to another window/tab.
Even that wouldn't be so bad if you could reliably undo out of it.
When you move things around and break stuff, let's say by creating two disconnected solids, then I would expect a preview to show that.
Also, one problem with FreeCAD is that you need to open a sketch to modify it but often the model you are modifying is then hidden. So now you have to roughly guess the optimal value, close the sketch and inspect the model and repeat this process until you have something usable. If you need to use clipping planes to see inside the model then enjoy setting those up every iteration.
Anything beyond simple polyhedra and booleans runs into a problem. Fillets, chamfers, sweeps and lofts don't work well. It's a shame.