Better imprint for assy by adam-urbanczyk · Pull Request #2069 · CadQuery/cadquery

adam-urbanczyk · 2026-07-16T19:08:48Z

Low level fix enabling faster imprints for well behaved models.

NB: this PR also contains a typing fix related to Numba version update.

Tangentially related to this PR (I added explicit test for this use case) - it is possible to imprint subassemblies separately and on per subassy basis use different imprint glue settings in order to optimize performance:

res1, origins1 = cq.occ_impl.assembly.imprint(complex_assy.sub1, glue='partial')
res2, origins2 = cq.occ_impl.assembly.imprint(complex_assy.sub2, glue='full')

# combine the results
res: Compound = res1 | res2
origins: dict[Shape, Shape] = origins1 | origins2

codecov · 2026-07-17T10:24:36Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.70%. Comparing base (b47f7b7) to head (2da3b69).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2069   +/-   ##
=======================================
  Coverage   95.70%   95.70%           
=======================================
  Files          30       30           
  Lines        9479     9479           
  Branches     1410     1410           
=======================================
  Hits         9072     9072           
  Misses        253      253           
  Partials      154      154

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

lorenzncode

Thanks @adam-urbanczyk !

OCC's BRepMesh_IncrementalMesh simple constructor leaves IMeshTools_Parameters.AngleInterior unset, and OCC then meshes the interior grid of doubly-periodic BSpline faces (full-torus shells) with cells that violate the angular tolerance by an order of magnitude in a band next to the u-seam: 150 cm chords at a 0.1 cm deflection request, sagging several cm off the true surface. On nested thin shells (the complex-cad radial build: first wall, vacuum vessel, nuclear shield, thermal shield in contact) the seam bands of adjacent surfaces cross each other, so every shell is watertight and consistently oriented but the volumes locally interpenetrate, and DAGMC loses ~1e-3 of all particles ("No intersection found with DAGMC cell N"). Route every BRepMesh call through a new run_brepmesh() helper that sets Deflection/Angle and DeflectionInterior/AngleInterior explicitly via the IMeshTools_Parameters overload. On the radial build this removes all seam-band triangles (max edge 147 cm to 41 cm) and takes OpenMC lost particles from 191/200k (and 386/2M in a void streaming run) to 0. Also: - _imprint_assembly: use the BOPAlgo_Builder based imprint with glue="partial" when the installed cadquery has it (CadQuery/cadquery#2069, faster and lower RAM); fall back to the old single-argument imprint otherwise. - Skip imprinting for single-solid assemblies: it is a no-op there, and the BOPAlgo_Builder imprint returns a Null shape for fewer than two arguments. - Add the complex-cad radial build transport driver to the local sweep defaults (tier 3, lost=0 baseline).

adam-urbanczyk · 2026-07-24T19:59:58Z

@jmwright would you be able to take a look?

jmwright

This MRE:

import cadquery as cq
from cadquery.occ_impl.assembly import imprint
imprint(cq.Assembly().add(cq.Workplane().box(1, 1, 1)))

gives this error:

Traceback (most recent call last):
  File "<python-input-4>", line 1, in <module>
    imprint(cq.Assembly().add(cq.Workplane().box(1, 1, 1)))
    ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/assembly.py", line 885, in imprint
    res = Shape(builder.Shape())
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 445, in __init__
    self.wrapped = downcast(obj)
                   ~~~~~~~~^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 419, in downcast
    f_downcast: Any = downcast_LUT[shapetype(obj)]
                                   ~~~~~~~~~^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 409, in shapetype
    raise ValueError("Null TopoDS_Shape object")

Co-authored-by: Jeremy Wright <wrightjmf@gmail.com>

adam-urbanczyk · 2026-07-26T07:44:08Z

This MRE:

import cadquery as cq
from cadquery.occ_impl.assembly import imprint
imprint(cq.Assembly().add(cq.Workplane().box(1, 1, 1)))

gives this error:

Traceback (most recent call last):
  File "<python-input-4>", line 1, in <module>
    imprint(cq.Assembly().add(cq.Workplane().box(1, 1, 1)))
    ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/assembly.py", line 885, in imprint
    res = Shape(builder.Shape())
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 445, in __init__
    self.wrapped = downcast(obj)
                   ~~~~~~~~^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 419, in downcast
    f_downcast: Any = downcast_LUT[shapetype(obj)]
                                   ~~~~~~~~~^^^^^
  File "/home/jwright/repos/cadquery/cadquery/occ_impl/shapes.py", line 409, in shapetype
    raise ValueError("Null TopoDS_Shape object")

Imprint needs at least 2 shapes to do something meaningful . The previous version does not indeed throw, I'll add an explicit check for this.

adam-urbanczyk · 2026-07-26T08:02:49Z

Done, I also added a check for no solids at all.

jmwright

Thanks @adam-urbanczyk

The zoo had no model where two solids meet over part of a curved surface. Every existing multi-solid model either shares an exactly coincident face, because the enclosing body was cut by the same primitive, or does not touch at all. Partial contact on a curved surface is the case that separates OCCT's gluing options, so add two models that cover it. TwoAnnularSectors puts a shorter sector directly outside a taller one, both spanning the same 90 degrees. The contact patch is a band that reaches the angular edges of the first sector's outer face, so the imprint edges run clean across that face. Every gluing option handles this, which makes it the control case. TwoAnnularSectorsPartialArc makes the second sector narrower as well as shorter, so the patch is strictly interior to the first sector's outer face and reaches its boundary in neither the axial nor the angular direction. Imprinting then has to split the larger curved face around an island: glue solids shared faces shared area full 2 0 0 partial 2 1 23.5619 None 2 1 23.5619 The shared area matches the analytic 6 * radians(45) * 5. Full gluing returns the two solids with no shared face at all, so the DAGMC model would not be watertight at that interface. This is the geometry that shows why partial is the right default, following CadQuery/cadquery#2069. Solid volumes are analytically exact, 212.0575 and 176.7146 for the band model and 212.0575 and 88.3573 for the partial arc model, and the CSG regions agree. Both models reproduce their CSG reference under the gmsh mesher, to 0.01 percent on material 1 and 0.10 percent on material 2. Adds the models, tests for both suites, examples, preview images and README entries. Both fail under the tolerance and angular_tolerance meshing path with about 19 percent error on material 2, which is the same cad_to_dagmc defect described in the parent pull request and not a property of these models.

shimwell · 2026-07-29T09:12:22Z

I am now using this feature and it is super useful.

I have a use case where I have some CQ geometry and I don't know if I can get away with 'full' imprinting (which is fast but can be wrong) or if I need 'partial' imprinting (which is slower but always correct).

The CQ models are large and prepared upstream so without a lot of inspection I'm never going to know if I can go for full imprint in some surfaces.

Is there any possibility of having another mode 'auto' which know when 'full' can be used and 'partial' is necessary. For example can we use check the faces/edges and somehow conservatively identify fully coincident contacts or partial face overlap

lorenzncode · 2026-07-30T01:22:05Z

I am now using this feature and it is super useful.

I have a use case where I have some CQ geometry and I don't know if I can get away with 'full' imprinting (which is fast but can be wrong) or if I need 'partial' imprinting (which is slower but always correct).

The CQ models are large and prepared upstream so without a lot of inspection I'm never going to know if I can go for full imprint in some surfaces.

Is there any possibility of having another mode 'auto' which know when 'full' can be used and 'partial' is necessary. For example can we use check the faces/edges and somehow conservatively identify fully coincident contacts or partial face overlap

I'm not an OCCT expert but according to the docs isn't glue faster because it skips computations based on assumptions up front?

https://dev.opencascade.org/doc/overview/html/specification__boolean_operations.html#specification__boolean_7

The performance improvement in gluing mode is achieved by excluding the most time consuming computations and in some case can go up to 90%:

Exclude computation of FACE/FACE intersections for partial coincidence;
Exclude computation of VERTEX/FACE, EDGE/FACE and FACE/FACE intersections for full coincidence.
By setting the Gluing option for the operation user should guarantee that the arguments are really coinciding. The algorithm does not check this itself. Setting inappropriate option for the operation is likely to lead to incorrect result.

If you run these computation in 'auto' mode then doesn't that move a time consuming part to a different function?

jon-proximafusion · 2026-08-13T12:27:32Z

I was thinking the CAD builder could define imprint method to sure on the assembly surfaces, perhaps as a metadata.

Then the code ingesting the CAD and doing the imprint could look for this metadata per surface as an optional clue to what imprint to use.

Perhaps not worth the additional code but could speed up the imprinting at the expense of tagging the surfaces

lorenzncode · 2026-08-14T17:32:22Z

Assembly supports user metadata, so this could be tested now with a custom imprint wrapper.

from cadquery.occ_impl.assembly import imprint as imprint_assembly

def imprint_with_metadata(assy):
    glue = assy.metadata.get("imprint_glue")

    if glue not in (None, "partial", "full"):
        raise ValueError(f"Invalid glue mode: {glue!r}")

    return imprint_assembly(assy, glue=glue)

This works at the assembly level. I'm not clear on how the proposal would work "per surface".

adam-urbanczyk · 2026-08-14T22:05:57Z

This maybe could be done calling BOPAlgo_PaveFiller directly and/or manipulating its state somehow.

Better imprint for assy

f688133

adam-urbanczyk added the assembly label Jul 16, 2026

Typing fix

a4acc30

adam-urbanczyk closed this Jul 17, 2026

adam-urbanczyk reopened this Jul 17, 2026

Add a note about glue and threads.

bd53fc8

adam-urbanczyk requested review from jmwright and lorenzncode July 17, 2026 14:21

lorenzncode approved these changes Jul 18, 2026

View reviewed changes

Comment thread cadquery/occ_impl/assembly.py Outdated

Comment thread cadquery/occ_impl/assembly.py Outdated

This was referenced Jul 21, 2026

Use the BOPAlgo_Builder imprint with glue partial when available fusion-energy/cad_to_dagmc#191

Merged

Route closed doubly-periodic b-spline faces to BRepMesh with interior deflection control fusion-neutronics/core#251

Merged

adam-urbanczyk added 3 commits July 24, 2026 13:52

Tweak comments and docstrings

f66dbfb

Add a test for subassy imprinting

4d80f4c

Changed the default option to partial

48a005a

jmwright reviewed Jul 25, 2026

View reviewed changes

Comment thread cadquery/occ_impl/assembly.py Outdated

Comment thread cadquery/occ_impl/shapes.py Outdated

Apply suggestions from code review

9065f16

Co-authored-by: Jeremy Wright <wrightjmf@gmail.com>

Handle edge cases

2da3b69

adam-urbanczyk requested a review from jmwright July 26, 2026 08:02

jmwright approved these changes Jul 26, 2026

View reviewed changes

jmwright merged commit 6ab9565 into master Jul 26, 2026
7 checks passed

shimwell mentioned this pull request Jul 26, 2026

Add two annular sector models with curved partial contact fusion-energy/model_benchmark_zoo#68

Merged

Uh oh!

Uh oh!

Conversation

adam-urbanczyk commented Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Jul 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

lorenzncode left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

adam-urbanczyk commented Jul 24, 2026

Uh oh!

jmwright left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

adam-urbanczyk commented Jul 26, 2026

Uh oh!

adam-urbanczyk commented Jul 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jmwright left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

shimwell commented Jul 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lorenzncode commented Jul 30, 2026

Uh oh!

jon-proximafusion commented Aug 13, 2026

Uh oh!

lorenzncode commented Aug 14, 2026

Uh oh!

adam-urbanczyk commented Aug 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

adam-urbanczyk commented Jul 16, 2026 •

edited

Loading

codecov Bot commented Jul 17, 2026 •

edited

Loading

adam-urbanczyk commented Jul 26, 2026 •

edited

Loading

shimwell commented Jul 29, 2026 •

edited

Loading