step exporter generates incorrect geometry

[greyltc]

This is with cadquery commit gf0308ce and opencascade 7.5.0:

widget = # insert code to make a thin uniform sheet
oddly_shaped_holes = # insert code to make some oddly shaped hole geometry
bunch_of_cylinders = # insert code to make a bunch of non intersecting cylinders on a plane
widget = widget.cut(oddly_shaped_holes)
widget = widget.cut(bunch_of_cylinders)
cadquery.exporters.export(widget, "widget.stl")
cadquery.exporters.export(widget, "widget.step")

widget.stl contains the geometry I expect and the region of interest here looks like this:

but widget.step looks like this:

I made that crescent shape (as shown in the .stl) by using widget = widget.cut(bunch_of_cylinders) to cut a bunch of cylinders out of a thin sheet. Many of the resulting voids in the thin sheet step file (not shown here) are correct, but some, like this one, have somehow managed to invert the boolean operation I asked for and instead of subtracting the cutting tool from the stock, the step file shows the cutting tool shape (a cylinder) added to the stock part.

Is this a bug in opencascade, not cadquery? Is the step file exporter known to be broken? Are there workarounds/tricks I can play to make step files in a more reliable way?

[greyltc]

I should mention that I'm viewing the step file with freecad, and just to make sure that it's not a bug in freecad, I also uploaded the step to https://www.emachineshop.com/free-online-step-file-viewer/ which agrees with freecad's render.

[greyltc]

Could this be a bug in opencascade's fuzzy boolean operations?
If I make the cutting tool cylinders an extra 1 mm taller than they need to be and translate them in Z by -0.5mm, the step file is generated correctly.

I have trouble understanding how that could only apply to step file exports though.

[jmwright]

STL export uses meshes, which is a very different process than STEP export. There could be invalid geometry at the locations that the STL exporter smooths over. Have you checked the STL to make sure it doesn't have any errors? Can you produce a minimum working example so that we can test?

[greyltc]

I don't think it's related to the triangle approximation going on in the .stl files.
Seems to actually be a problem with the step file export pathway:

• cq-editor renders it correctly
  
• my dxf exports of the region are also fine
  

[fedorkotov]

I have reproduced the problem

widget = \
    cq.Workplane("XY")\
    .rect(5,5)\
    .extrude(0.5)
    
oddly_shaped_holes = \
    cq.Workplane("XY")\
    .move(0,1.2)\
    .lineTo(-0.5,1.2)\
    .lineTo(-0.5,1.5)\
    .lineTo(-1.5,1.5)\
    .lineTo(-1.5,0)\
    .mirrorX()\
    .mirrorY()\
    .extrude(2)\
    .edges("|Z")\
    .fillet(0.13)
    
bunch_of_cylinders = \
    cq.Workplane("XY")\
    .rarray(1,3,4,2)\
    .eachpoint(
        lambda loc: cq.Workplane("XY")\
               .circle(0.15)\
               .extrude(2)\
               .val()\
               .located(loc))
widget = widget.cut(oddly_shaped_holes)
widget = widget.cut(bunch_of_cylinders)
cq.exporters.export(widget, f'widget.stl')
cq.exporters.export(widget, f'widget.step')

Everything is looks ok in cq-editor

stl looks ok in https://www.viewstl.com/#!

But in www.emachineshop.com viewer something strange happens with holes

[greyltc]

I have reproduced the problem

Ha! Thanks! Even before I had a chance to distill it! 👍

[greyltc]

@fedorkotov

If you change .extrude(2)\ in oddly_shaped_holes and bunch_of_cylinders to .extrude(0.5)\ so that the cutting tools match the thickness of the stock part, I bet you'll see more along the lines of what I see.

If you change
widget = widget.cut(bunch_of_cylinders)
to
widget = widget.cut(bunch_of_cylinders.translate((0,0,-0.5)))
the problem will be fixed I think.

[fedorkotov]

Indeed. Your suggestion fixed it :)

This is the result of reducing extrude depth to 0.5

And this is the result of moving both subtracted features down 0.5 with extrusion depth 2

This is the working variant

widget = \
    cq.Workplane("XY")\
    .rect(5,5)\
    .extrude(0.5)
    
oddly_shaped_holes = \
    cq.Workplane("XY")\
    .workplane(-0.5)\
    .move(0,1.2)\
    .lineTo(-0.5,1.2)\
    .lineTo(-0.5,1.5)\
    .lineTo(-1.5,1.5)\
    .lineTo(-1.5,0)\
    .mirrorX()\
    .mirrorY()\
    .extrude(2)\
    .edges("|Z")\
    .fillet(0.13)
    
bunch_of_cylinders = \
    cq.Workplane("XY")\
    .workplane(-0.5)\
    .rarray(1,3,4,2)\
    .eachpoint(
        lambda loc: cq.Workplane("XY")\
               .circle(0.15)\
               .extrude(2)\
               .val()\
               .located(loc))
widget = widget.cut(oddly_shaped_holes)
widget = widget.cut(bunch_of_cylinders)
cq.exporters.export(widget, f'widget.stl')
cq.exporters.export(widget, f'widget.step')

[fedorkotov]

@greyltc Please try the same trick on your original model. Will it fix the problem there too?

[greyltc]

Will it fix the problem there too?

Yep, it does. It's like the fuzzy boolean operation feature (see my comment above) is somehow lost in the step export pathway.
Edit: I don't know...I'm not really sure that could explain how a boolean subtraction could turn into a boolean addition.

[fedorkotov]

I thought that BREP means all the history of unions, intersections, etc is lost after each operation and only the resulting trimmed surfaces with common edges and vertexes remain. How is it possible that after successfully cutting the holes (as can be seen in cq-viewer or in STL models), STEP export result depends on how exactly the solid was made?
Maybe one of the experts here can explain..

[marcus7070]

I thought that BREP means all the history of unions, intersections, etc is lost after each operation and only the resulting trimmed surfaces with common edges and vertexes remain. How is it possible that after successfully cutting the holes (as can be seen in cq-viewer or in STL models), STEP export result depends on how exactly the solid was made?

OCCT stores that edge as something along the lines of a "trimmed circle". It looks like OOCT is inverting which side of it is trimmed when converting it to a STEP.

Edit: that sounds more confident than I intended. This is just my best guess.

[adam-urbanczyk]

It seems to be STEP related - if I export in the brep format the issue is not there.

[greyltc]

Possibly related: https://tracker.dev.opencascade.org/view.php?id=32132

[marcus7070]

Two step files attached, one made with CQ-editor export and OCP/OCCT 7.4, the other with 7.5.1. I can see the error in 7.5.1, but not in 7.4.

STEP files renamed to TXT files because github... 😑
trouble_7_5_1.step.txt
trouble_7_4.step.txt

Made with @fedorkotov's code as written here.

[jmwright]

Looks like that OCCT issue hasn't been triaged yet, and the last comment was the middle of Feb. It looks to me like it's related to this issue.

[greyltc]

To anyone else who is also being bothered by this bug, I have a workaround:
The latest freecad appimage release was built with opencascade version 7.4.0. If I save my cadquery-generated geometry as .brep like so:

from OCP.BRepTools import BRepTools
BRepTools.Write_s(widget.toOCC(), "good_geometry.brep")

I then get bugfree .brep files which I can manually import with the appimage version of FreeCAD and reexport as step files that are also bugless (since the problem doesn't exist in occt 7.4).

[adam-urbanczyk]

Also cq.Shape.exportBrep, if you don't want to get your hands dirty with OCP.

[greyltc]

Upstream bug report: https://tracker.dev.opencascade.org/view.php?id=32264

[adam-urbanczyk]

Great, thanks for reporting! They do like draw.exe scripts.

[greyltc]

The opencascade devs have quickly come up with a fix for this issue.
corrupted_step_fix.patch.txt
I guess this will appear in the upcoming occt 7.6.0 release.

Edit: Actully, looks like they've decided to roll this fix into a 7.5.2 release.