[Bf-modeling] Proportional Edit Mode (Connected)

[Campbell Barton]

On Tue, Jun 18, 2013 at 12:02 AM, Howard Trickey
<howard.trickey at gmail.com> wrote:
>
>
>
> On Mon, Jun 17, 2013 at 8:52 AM, Campbell Barton <ideasman42 at gmail.com>
> wrote:
>>
>> Hi, There are cases where the current proportional edit-mode
>> 'Connected' option doesn't give a nice smooth result across a grid for
>> example.
>>
>> For details see the 2 links below, but these images sum up the problem
>> quite well:
>> *
>> http://projects.blender.org/tracker/download.php/9/498/35590/25506/ProportionalEditingBug.png
>>
>> * http://i.imgur.com/U5TZW3i.png
>>
>>
>>
>> Details... previous discussion and bug report:
>> *
>> http://projects.blender.org/tracker/index.php?func=detail&aid=35590&group_id=9&atid=498
>>
>> * http://markmail.org/message/zkon53qzx32b5xek
>>
>>
>>
>> Probably long discussion is not needed about this.
>>
>> I think there are 2 obvious options....
>>
>> 1) Accept that connected works like this, just close the bug and tell
>> users it known limitation with current design.
>>
>> 2) Add a second "Connected" option, so we have...
>> * "Connected (Topology)" --- what we have now.
>> * "Connected (Distance)" --- the real distance would work like it did
>> many years ago, so only connected geometry is effected, but that
>> geometry uses the distance between the 2 points without measuring
>> along the topology.
>>
>> {for a less obvious option}
>>
>> 3) we could keep the options as they are now, but make "Connected"
>> work without giving ugly artifacts,
>> ... For example rather them simply measuring along edges we could
>> measure across faces too... this would take longer to do and not
>> necessarily give perfectly smooth results, but could still be worth
>> looking into.
>
>
> I think this 'across-face' rule could possible work.  After all, the
> vertices
> ARE connected via the face so it makes sense that the distance across
> the face should be what is measured for the strength of the effect.

Added support for measuring across the face, r57528 (always enabled)

The falloff is still not nice and rounded, but with diagonals its a
little better with a grid.

http://www.graphicall.org/ftp/ideasman42/pet_diagonal.png

I'm less worried about this artifact then before, I tested the Sintel
model and I cant get it to show any artifacts using pet-connected so
this is not so much a problem unless you are wanting to use pet on
grids.

At least wont look like this anymore.
http://i.imgur.com/U5TZW3i.png


Bug [#35590], was in fact an unrelated error in the connectivity check
(fixed r57532)
http://projects.blender.org/tracker/download.php/9/498/35590/25506/ProportionalEditingBug.png


> When I first read this, I though that it would destroy the use case in this
> mail:
> http://lists.blender.org/pipermail/bf-committers/2013-April/039799.html
>  -- that is, a long chain of vertices that form a path that comes
> near itself distance-wise but not topologically, and you only want to
> affect the ones that are near topologically. But here there are no faces
> across which the vertices that are 'near distance-wise' are near across.
> Even if you changed the long chain of vertices into a long strip of quads,
> the same argument would hold.
>
> I think what we would really like is that proportional edit measures
> real distances that follow the surface of the mesh, as short as possible
> following that rule.  If there are no faces, it would have to follow edges,
> but measure the length along those edges.
>
> But this might be hard to calculate, especially if the mesh is non-manifold.
> There may be multiple paths between two points, complicating things even
> further.

Was thinking over possible ways to get this working and concluded that
its probably too much to calculate...
we could...

- Add more fine grained measurement when walking over the mesh - use
face centers, middle of edges, and walk across these too, the error
will still exist but be less obvious.

- Record the path to the closest vertex, then re-walk the path and
perform a `straighten` on that path based on surrounding faces.

- Fake it and take turning corners into account when walking over
manifold edges, I think this could avoid artifacts across flat
surfaces, but it could interfere with the cases where you just want
direct topology-length without any extra calculations over the top.

> If we don't want to go down the path of trying to implement that, I would be
> in favor of option 2.  Or maybe a variant of that where you provide a slider
> that goes from 0->1 to say what proportion to consider topological distance
> vs distance in 3-space.

Not sure a slider is really needed, we dont have a good place for this
in the UI at the moment either.

>> --
>> - Campbell
>> _______________________________________________
>> Bf-modeling mailing list
>> Bf-modeling at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-modeling
>
>
>
> _______________________________________________
> Bf-modeling mailing list
> Bf-modeling at blender.org
> http://lists.blender.org/mailman/listinfo/bf-modeling
>



-- 
- Campbell