Okay! Finally i found the reason.
My bad.

Yes, in the new file it seems it work.
1st problem: it acts strange combined with Solidify (tuned by v-group)
2nd problem: in this particular file gradient broken.

I think, LMB waiting for drag and double click events, that is why it is slower.
Use С-brush instead or Control-click for continuous select. It is much faster

@Jacques Lucke (JacquesLucke)
I expect even distance between edges. the same way as solidify do it, for example.
even edge offset literally: edge offset from edge, not vertex from vertex as it is now.
Not a bug but wrong behaviour, as I think.

Any news?

any news?

Any news?

@Jacques Lucke (JacquesLucke)
I hope, 2D masking will be implemented too. It is very common tool.

@Jacques Lucke (JacquesLucke)
I changed videocard, drivers, crashes now is very rare, but still in the same situation:
changing shader on the fly with enabled Cycles/Evee render in the side window.

@Vladimir Spivak (cwolf3d) any news?

@Demeter Dzadik (Mets)
I have no errors inside Blender

@ronan ducluzeau (zeauro) first question was asked in 2015. It is little more than 2 days

@Brecht Van Lommel (brecht)
Very strange. Is there reasons to make full render in case when I need only Object pass?

@Sebastian Parborg (zeddb) where I can read about this limitation and reasons of artifact?

@William Reynish (billreynish) Ok, so papercut.

It is strange that glitch appear only when mesh is on the certain positions.
Only in the certain x-positions Blender tries to connect vertices wrong. Y positiom changes nothing.

@Antonio Vazquez (antoniov) Thanks, it helped with visibility.

The same. States of system do not jump and bakes properly.
Here the comparison

@dark999 (dark999)
check the post with second video.

@dark999 (dark999) updated.

@Sebastian Parborg (zeddb) I`ve encountered this problem again.
Mode of shading in Workbench change flatshading to smoothshading.
Look at that!

Also I found that GPencil object or data have influence too, but
function ignores GPencil-object transforms as you can see:

@Vladimir Spivak (cwolf3d)
Tested. Thanks for adding checkbox.

Thanks!

@Brendon Murphy (meta-androcto)
I checked it again. Still have problem.
Now it work with active annotation layer
No error, but Gstretch stop to work after cancelling/deleting annotation stroke at all until I delete data.
There should be manual checkbox to allow me to select if i want use stroke or not.
If there is no stroke in active annotation layer, Gstrech should ignore this setting. If there is stroke but switch is off, It should ignore stroke too.

@Brendon Murphy (meta-androcto) @Mikhail Rachinskiy (alm)
Fix is preferable.
Or at least workaround: second part is already selected, so exit from local and enter again.
Sadly it will reset camera

@Tibor Horvath (tibor81)
Filter 8, clamp 2 (because lower clamp looks reaaly dull), 512px cubemap
Pinched pole still there.

@Tibor Horvath (tibor81) clamp is just cutting brightness of artifacts. But wrong highlight blur is still there. And resolution of map makes em even sharper
Here is Clamp to 1.0, roughness 0.27, cubemap 2048

@Brecht Van Lommel (brecht) thanks, a lot and sorry for disturbing. Now I see how curve behave in third case.

@Aaron Carlisle (Blendify)
I found this scene before I post here.
It just crashes my 2.79 or 2.80, and append crush it too
It made with 2.49… And it is old creepy thing I do not want to touch.

@Brecht Van Lommel (brecht) what kind of clipping? I can`t find such option for Evee.

pinching pole is not depend on rotation of environment, or object`s rotation. Glitch appearance is not depends on streight of world illuminance

@ronan ducluzeau (zeauro) added two links
Sadly filter quality just blur and pieces.
And even if blocks blurred and mixed, there is too high and sharp pinch to the pole of world. Check second video.

@Sebastian Parborg (zeddb) any news about this bug?
Filtering is a great feature, but users will avoid to use it due bug…

@Sebastian Parborg (zeddb)
it seems this is still not solved. And not documented.
I tried to use attractor (goal) with force field and it changes nothing.
I think this is powerful thing, that underestimated as metaballs due to poor documentation and tricky control

@Germano Cavalcante (mano-wii) yes, Now i see how it works. A bit strange separation of shapes of influence and acatual field-shape.
Thank you.

@Germano Cavalcante (mano-wii) and what actually it intends? How can I figure out? There is no remarks in documentation.

@Sebastian Parborg (zeddb) I confirm, that scene will work in viewport fine, after clearing all caches, saving and reverting file. But baking is impossible for me.

@Sergey Sharybin (sergey) just baked similar scene with 2.79 and all work as indended. No bugs. So yes, it is regression.

@Philipp Oeser (lichtwerk) awright, thanks!
I think sphere limits should be dashed too just for consistancy

@Germano Cavalcante (mano-wii) I hope, you are just kidding and such bad interaction can not be «intended»