This issue is due to the precision of the used numbers. It is exactly the same in the edit mode convex hull.

@tempdevnova (tempdevnova) In your test file Bevel_pinching2.blend it is clearly a user problem, as the faces are non planar. Bevel mostly works with non planar surfaces, but normals behave different and happen to cause problems.

In T95541#1303232, @Philipp Oeser (lichtwerk) wrote:

This changed in rB1a7757b0bc69: Fix T91444: Edge Loop Preview fails with two Mirror Modifiers ...
... but that commit got reverted in rBe6cd4761e7c3043c8e9dc789b6eb6d0aa1b723bd

The change in behavior is still present though (as is the code -- bisecting between spits out rBaa1c44a11348e6e2dcc987bdb7bcdb541d83b631?), not sure what is going on with git.

CC @Bastien Montagne (mont29)
CC @Campbell Barton (campbellbarton)
could you check?

This bug is now there again

• consider all values in the grease pencil object for the correct thickness

There is a regression in the complex solidify modifier. This should fix it.

diff --git a/source/blender/modifiers/intern/MOD_solidify_nonmanifold.c b/source/blender/modifiers/intern/MOD_solidify_nonmanifold.c
index 658eced3c20..d032a176bf2 100644
--- a/source/blender/modifiers/intern/MOD_solidify_nonmanifold.c
+++ b/source/blender/modifiers/intern/MOD_solidify_nonmanifold.c
@@ -1973,7 +1973,7 @@ Mesh *MOD_solidify_nonmanifold_modifyMesh(ModifierData *md,
   int *origindex_edge = CustomData_get_layer(&result->edata, CD_ORIGINDEX);
   int *origindex_poly = CustomData_get_layer(&result->pdata, CD_ORIGINDEX);

apart from the one comment it looks good. I will check if my tests work later.

Yes, this is caused by me. If you take a look at the removed lines in D14548 you can see how the code was doing the smooth based on the original stroke and not on the new position data. So the result was a slight smooth and nothing else.

Looks like the correct fix. Thanks for looking at it.

I have no trouble selecting it with box select. Clicking to select seems to be broken. So maybe the bug is somewhere in the selection code. Maybe it tries to select the non selectable strokes, as they are first in the list of strokes?

The whole problem came when the eye needed the colorful backdrop. If implemented naively, the backdrop would be blue when the eye is closed and gray when the eye is open. This does not match other areas, so to solve this, a new rna property was introduced which is equivalent to the property in modifiers. The old property is called mute, the new property is called enabled. Both access the same flag, but one is inverted. Since the new eye icon has a new rna property with the inverted meaning, the old drivers don't show up.

In T94324#1331019, @Philipp Oeser (lichtwerk) wrote:

This is because they open the menu on a timer

On mouse over you mean? Not on my side, how is this done? Does not seem to beaffected by the use_mouse_over_open User Preference?

I still very much dislike Minimum Radius. From the name you really don't get any intuition that it is just a factor which gets multiplied on to the final weight.

In T94324#1299802, @Richard Antalik (ISS) wrote:

@Henrik Dick (weasel) is this still an issue? don't have tablet to check here, but since you are involved with development, you can confirm reports. If you want somebody else to check, perhaps @Philipp Oeser (lichtwerk) can have a look.

Don't know the exact commit, but this problem does no longer persist.

The file MOD_gpencilweight_curvature.c is missing right now.

• fix: add new icon to modifier as well

• fix crash in sculpt mode
• change icon to ICON_MOD_SKIN

• requested changes to the UI and defaults
• cleanup of the code

@Matias Mendiola (mendio) I think the thickness slider in deform mode is quite nice, since the usual usecase, that I am optimizing for would otherwise always require a thickness modifier before with thickness = 0. Since I am changing the thickness in the modifier anyway and since I have the slider already from the other modes, I may as well just include it and make the modifier stack more compact for the most common usecase.
The most common usecase for the deform mode that I am talking about, is using it as an alternative to the thickness from curvature modifier stack, mostly when doing handwriting.

• fix interpf bug, where interpf works the other way round from interp_v3_v3v3
• cleanup UI to only show the relevant options

• add thickness and strength slider for new geometry in segments/fill mode

• add mode option instead of checkbox
• add material option for newly generated strokes

• added a function for general purpose smooth operations.

@Falk David (filedescriptor) so you want me to just move all of the code of the smooth modifier with all features into one function in BKE_gpencil_geom.cc? I would still need the per point functions for sculpt mode though.

@Falk David (filedescriptor) The problem is that the function is called for every point, so allocating inside is no option. Then also, not always can the function be used to skip iteration completely, see gpencil_edit.c. There it's very useful to allocate ouside the loop and then copy multiple times to the allocated memory. Sometimes if there is only one smooth iteration, it is not even really worth doing the allocation as the result is basically the same without.

• add versioning

I just did some good performance testing. The setup was a single grease pencil object with a noise modifier to make sure it gets updated every frame and a smooth modifier after that. The object has 23130 points accross 58 strokes. Then frame times are recorded for the old version (most recent commit) and the new version (next revision of this patch). The results where stunning.

• fully support cyclic strokes
• improve no segments mode

If you are on KDE you can save your computer from going down by pressing Ctrl+Alt+Esc and then click on blender to kill it.

@Antonio Vazquez (antoniov) I am not changing the active smooth in the function gpencil_smooth_buffer. That one works really well and gives a good feeling while drawing and I don't wanna mess with that. The change here is most beneficial for really strong global smoothing operations so highly doubt that it could improve the active smooth. For the editmode smooth I even had to use the old approach of using multiple iterations with less reach in order to account for unselected points in an easy way.

• cleanup patch files

Adding versioning here would mean changing the iterations option. So if you would load the updated file in an old version you would then get too strong smoothing. Is that fine? If so, I would like to add versioning code.

Cannot reproduce on linux.
Can you reproduce it with a factory settings?

• cleanup code
• fix editmode smooth defaults and ranges

Currently there is a few things left to do:

• Smoothing for the grease pencil sculpt mode is not yet improved. I could add two options to the smooth brush to keep_shape and smooth_caps. I would also wish to make it smooth symmetrically, but I haven't figured that out yet.
• Check the places where smooth was used, where I put in default values to try to match the old smooth.
• Find all places where the smooth factor is set to a default of 0.5 and replace it with 1.0 because that works much better with the new algorithm.

The patch works, but it's not yet in the state I would like it. Using the code from the curvature length modifier like in the previous revision is probably not the best fit for purpose.

Currently the patch is breaking the length modifier (as said). Make a stroke with 4 points and use the length modifier on it. Only one side will make a helix.

@Campbell Barton (campbellbarton) can you take a look here and comment on this?

The usage of Vertex Bevel Weight for the rim is currently only implemented for complex solidify. The same feature will be missing for Vertex Creases when D14170 lands.

@Bastien Montagne (mont29) Even if undo of vertex group can be considered a known issue, the silent corruption of all the other vertex groups should be considered a bug and should get fixed (maybe by disabling the undo behavior here completely?).

The patch currently breaks the length modifier. It does not apply the curvature thing on the start extention.

@Germano Cavalcante (mano-wii) The behavior is very different for vertices as vertices don't just get the selection behavior of the vertex group, but also different sizes for different weight values. To resolve this problem, either the vertex only behavior would need to be changed to use weight purely as selection and use bevel weights for this fine control or the threshold for a vertex to be bevels should be set to > 0 to avoid the jump (see previous comment).

It actually happens not only with the new operators, but also with the old ones like bpy.ops.constraint.delete(constraint="Transform Cache").

@Simon Thommes (simonthommes) I tried your new file in T95963 and there is still one little spike left on the bar. The artifact is clearly new and came from D14143. The problem with D14143 is that is does not handle non planar faces very well. I will have to think about a better solution for non planar faces, but that is out of the scope of this patch. I could increase the thresholds higher to hide the artifacts, but that will cause more and more inprecision in clean models with only planar faces. For your particular model the artifact dissappears if you triangulate your mesh using a triangulate modifier before solidify or if you use the even thickness mode.