Bent plate, best way to add feature and avoid topo errors

[kisolre]

Did not read all but here is last file with datum plane attached to the sketch:
.

AttachmentExample.JPG (108.96 KiB) Viewed 286 times

[kisolre]

Sorry, did not recalculate previous file. Here is a corrected one with imported support edge for proper positioning of geometry relative to the resulting face.

[chrisb]

It is sort of reassuring to see that as time goes by modeling standards evolve: I had attached the DatumPlane in the same way as kisolre.

Freedman's arguments cannot be neglected just with a fingersnip and we probably all want to learn how to handle topological naming issues without getting too emotional. So I provoked situations where the "attached model" (shortcut for the "model having the DatumPlane attached") could break, while the unattached stays sort of stable. As base I took a slightly modified version of kisolre's model.

• I removed a coincidence ...

  

  Snip macro screenshot-c04391.png (4.34 KiB) Viewed 269 times

  ... and added an additional corner. The result is as good as it can get, no additional action was necessary:

  

  Snip macro screenshot-62bfe1.png (8.99 KiB) Viewed 269 times

• The same but due to rounding errors changed the angle:
  No additional action was necessary
• Decreasing the height before or after one of the previous steps did not let the model break. No additional actions were necessary.
• Next I deleted the upper line and remade it. This caused in sketcher a renumbering from the upper edge from 3 to 6. That breaks the slot_padded with a multiple-solids message independent from a change in the angle.
  
  While the plane itself still seems to be ok, it seems to have turned around:

  

  Snip macro screenshot-2be6f5.png (12.27 KiB) Viewed 269 times

  I could with some difficulties (they will be subject of another topic) change the attachment and after reversing the pad things were roughly in their old position.
• Decreasing after that the height did not let the model break.
• I then did the same with the detach/reattach method.
  No additional actions were necessary, the DatumPlane could stay detached.
• With changing the angle:
  - Double click DatumPlane
  - select the same attachment mode as before
  - select Deactivated
  -confirm
• For changing the height - before or after - the same additional actions as in the previous case were necessary
• Then I redid the use case from above with deleting and recreating the upper line
  Usually I had expected that no additional actions were necessary, and that the DatumPlane could stay detached. But I had rounded the dimensions so the angle had changed.
  - without angle change: no additional action was necessary
  - with angle change: the same atachment difficulties as above + detach
• A height change after a no-angle-change requires the same new attachment with the additional difficulties as above.

So there seem to be some odds against each method which can be more or less tedious or serious. With the detach/reattach method trivial actions have to be made rather often, the benefit is that in some cases true attachment changes don't have to be performed at all.

[freedman]

chrisb, I performed the same tests you did with the same results, nice write-up. In reference to using sketch attachment, I could live with an error popping up or a modeling halt but that isn't what I get, I get a contorted model that can really mess with your head to fix, often times when these breaks occur they mess up down stream features.

For comparison: The same model using re/deactivation.
With the plane deactivated, I removed a coincidence and the model threw an error as expected, with the coincidence replaced the model recomputed no problem.
I then added a cutout on the sketch and changed the angle, this of coarse did not compute correctly because the datum plane is deactivated. With a couple mouse clicks I do an automated or a manual face-datum plane reattachment. Everything is correct.

The big difference I see here is when changes need to be done and I'm in the midst of a complex model, I don't want sketches rotating or swapping sides, it just freaks me out. I fixed many of those in the beginning and got pretty good at it but when you have a 100 hours into a customer model and it breaks somewhere back at it's beginning, well the blood pressure goes up for a while.

I found by using the deactivation method I get consistant results (when changing any sketch or feature) and I always let FreeCAD do the attachment so I know the dimensions and placements are correct. When topo errors are fixed then all datum planes can be activated.

I know everyone here has there own way of doing things and I can't change that (don't want to), but what about how do we teach the new users. What do you think?

[chrisb]

Short version: I think we should teach here how to properly attach. Telling them to detach after that may be more confusing. As I hope that the situation will improve anyway, this way is future safe too - no asking which version, before xxx? please detach, after xxx? please attach.

It takes quite some effort - usually more that it seems from the rather simple example we have here - to get maximum stability. On the other hand usually the base shape is quite well known and I hardly forget edges in a sketch or have too many. So the modifications here are almost always stable.

However, with the current version things can get worse than what we saw in the example above: I had hoped for a jumping to another plane, but it didn't . In reality such jumping can cause external references to become invalid or even worse to mess up a sketch so that it has to be reworked even after reattaching. What I would like to have is a possibility to do what you do always: detach things to be defined automatically until they are reattached.
I hope we won't need such tool after realthunder's toponaming improvements.

[freedman]

this way the future is safe too - no asking which version, before xxx? please detach, after xxx? please attach.

Ya! That's exactly what would happen. So for now, instruct on how to attach and then reference the stable model documentation to follow the current state of the program.

Seeing what happened here wouldn't it be prudent to modify the text here: https://www.freecadweb.org/wiki/Feature ... ble_models

Thanks for all the testing.