Sketcher: Harden against flip with signed constraints

[theo-vt]

This PR adds a concept of orientation to

• point-line distance
• circle-circle distance
• circle-line distance

to make sketches harder to flip

The orientation is saved to file and handled by the App::Constraint rather than the planegcs::Constraint to better capture design intent (it allows to react to external geometry movement, which the solver is blind to)

The orientation is computed on sketch restore so older files are upgraded seamlessly

The constraints can still be made driven or deactivated and their orientation is recomputed when made driving/reactivated so the user can manually flip them

Point-Line distance

This PR makes point-line constraints signed which prevents flips when constraints values change drastically or a reference geometry moves around

Before:

Screencast.From.2025-12-16.15-49-43.mp4

After:

Screencast.From.2025-12-16.15-50-50.mp4

Circle-Circle distance

This PR adds a concept of relative size for circle-circle distance which allows to keep an inner circle and an outer circle when one of the circles is completely within the other

Before:

Screencast.From.2025-12-18.23-16-07.mp4

After:

Screencast.From.2025-12-18.23-15-23.mp4

Circle-Line distance

This PR adds an orientation concept to circle-line distances. 4 possible orientations are established with 2 flags: ccw and internal.

A ccw orientation means that the sequence [line.pt1, line.pt2, circle.center] is and should remain ccw, this is the same as the point-line distance orientation from https://github.com/FreeCAD/FreeCAD/pull/26228/changes

An internal orientation means that the line is and should remain closer to the center then the radius of the circle

Before:

Screencast.From.2025-12-29.10-04-07.mp4

After:

Screencast.From.2025-12-29.10-05-23.mp4

Issues

Fix: #26634
Fix #17579

[belinski-roman]

What about parallel lines? I noticed that the offset is often broken in the sketch. Even for simple rectangles, I'm not talking about polygons.

[hyarion]

What about parallel lines? I noticed that the offset is often broken in the sketch. Even for simple rectangles, I'm not talking about polygons.

do you mean like this?

the green dimension here is set between the two lines, but is actually a point-line-constraint.
Setting the red dimension to 40 on main branch makes this jump as @belinski-roman explains

To my understanding, that should result in a point-line distance which should be covered by this PR.

[belinski-roman]

but is actually a point-line-constraint

But when setting the size, you can specify two vertices, you can specify a vertex and a straight line, or you can specify two straight lines. Is it always considered as a constraint between a point and a straight line?

[hyarion]

A rectangle with dimensions looks like this when horizontal and vertical constraints are removed and vertices moved around:

So a normal rectangle has point-point constraints.
That is not covered as we need three points to have a winding to ensure it is the same.

@theo-vt I think there's a case here we might be able to handle.
If a line (A to B) has horizontal constraint and a point-point dimension for the endpoints, then we could try to ensure that the vector AB still points in the same direction.
The big question is if we can make this connection between constraints.

Another option would be to change so rectangles gets line-point constraints by default.

[belinski-roman]

the current offset tool https://wiki .freecad.org/Sketcher_Offset Additionally, it draws auxiliary geometry — perpendicular segments (if you specify to create parametricity by setting the offset distance by size). In fact, you can force the distance from a point to a straight line to be tracked there. It's worth checking out

[hyarion]

https://wiki.freecad.org/Sketcher_Offset

Yes, the geometries created with this tool already have point-line constraints which are covered :)

[belinski-roman]

Yes, the geometries created with this tool already have point-line constraints which are covered :)

Which does not prevent you from flipping in most cases.

[hyarion]

Yes, the geometries created with this tool already have point-line constraints which are covered :)

Which does not prevent you from flipping in most cases.

Yes, but that's what I think this PR takes care of

[theo-vt]

@theo-vt I think there's a case here we might be able to handle.
If a line (A to B) has horizontal constraint and a point-point dimension for the endpoints, then we could try to ensure that the vector AB still points in the same direction.
The big question is if we can make this connection between constraints.

This is handled in #25645 the constraints are iteratively reduced to simpler constraints and it tries to use as many ConstraintDifference (signed) as possible

Currently this only works for axis aligned lines, but I want to extend it in 2 ways

• Transform point-point distance into point-line distance if the 2 lines are perpendicular
• Try to extend line sign when their angle can be deduced before

But I am marking it as draft because after the discussion in the cad-topic, I think this algorithm should be moved to App level rather than solver level to do smarter things, handle external geometries and eventually include other algorithms

[theo-vt]

Here are builds if people want to test!
https://github.com/theo-vt/FreeCAD/releases/tag/weekly-2026.03.02

[belinski-roman]

Here are builds if people want to test!

https://github.com/theo-vt/FreeCAD/releases/tag/weekly-2026.03.02

I tested it and didn't see any difference

trim.FDA88948-1A6C-4562-B39E-5D09361AD467.MOV

[belinski-roman]

Discord suggested what's the matter, but then you need to redo the Offset tool itself.

2026-03-03.15-23-14.mp4

[hyarion]

These issues are because point-to-point constraints are created and not the constraints modified in this PR.

I think we should change how the constraints are set up for rectangles, offsets etc, but that would be best in a follow up PR

[maxwxyz]

I have played around a bit and did not find any issues

[FEA-eng]

I've tried testing it with the first example shown here, and it still flipped, but maybe I used the wrong constraints?

Flipping.FCStd.zip

OS: Windows 11 build 26200
Architecture: x86_64
Version: 1.2.0dev.46046 (Git)
Build date: 2026/03/30 15:55:23
Build type: Release
Branch: main
Hash: 8cc01fa8bd5c0ef9cbd74b4dd0d1bed7d79f7272
Python 3.11.14, Qt 6.8.3, Coin 4.0.3, Vtk 9.3.0, boost 1_84, Eigen3 3.4.0, PySide 6.8.3
shiboken 6.8.3, xerces-c 3.3.0, IfcOpenShell 0.0.0, OCC 7.8.1
Locale: Polish/Poland (pl_PL)
Stylesheet/Theme/QtStyle: FreeCAD.qss/FreeCAD Light/
Navigation Style/Orbit Style/Rotation Mode: Blender/Rounded Arcball/Window center
Logical DPI/Physical DPI/Pixel Ratio: 96/81.5973/1

[theo-vt]

This is expected! Flipping is pretty vaporous problem so it can't be solved with a single pr, the example flip still does flip and I have future plans to solve it somewhat!

[chennes]

@theo-vt this is breaking two of my integration test files. Can you look at these two cases?

Recompute fails:
https://github.com/chennes/IntegrationTests/blob/main/Data/CADFiles/FL_AP3030S8_TNutM6.FCStd

Large volume change:
https://github.com/chennes/IntegrationTests/blob/main/Data/CADFiles/MG_KidToy.FCStd

[hyarion]

I've tried testing it with the first example shown here, and it still flipped, but maybe I used the wrong constraints?

this PR fixes flipping when constraints are based on more than two points (three points or point+line)