* Sketcher: Fix symmetry of slot redundancy false positive
Perturb symmetric geometries when using the 'create symmetric constraints' option to avoid numerical singularities in the solver (Jacobian Rank).
If geometry is "perfect", the solver cannot distinguish between the derivative of a Symmetry constraint and an Equal constraint, flagging one as redundant.
* BIM: Implement smart base removal for Walls
Previously, removing the Base object from an Arch Wall would cause the
wall to reset its position to the document origin and could lead to
unintended geometric changes for complex walls.
This commit introduces a "smart debasing" mechanism integrated into the
Component Task Panel's "Remove" button:
- For walls based on a single straight line, the operation now preserves
the wall's global position and parametric `Length`, making it an
independent object.
- For walls with complex bases (multi-segment, curved), a warning dialog
is now presented to the user, explaining the consequences (shape
alteration and position reset) before allowing the operation to
proceed.
This is supported by new API functions `Arch.is_debasable()` and
`Arch.debaseWall()`, which contain the core logic for the feature.
Fixes: https://github.com/FreeCAD/FreeCAD/issues/24453
* BIM: Move wall debasing logic into ArchWall proxy
The logic for handling the removal of a wall's base object was previously
implemented directly within the generic `ComponentTaskPanel` in
`ArchComponent.py`. This created a tight coupling, forcing the generic
component UI to have specific knowledge about the `ArchWall` type.
This commit refactors the implementation to follow a more object-oriented
and polymorphic design:
1. A new overridable method, `handleComponentRemoval(subobject)`, has been
added to the base `ArchComponent` proxy class. Its default implementation
maintains the standard removal behavior.
2. The `_Wall` proxy class in `ArchWall.py` now overrides this method. All
wall-specific debasing logic, including the eligibility check and the
user-facing warning dialog, now resides entirely within this override.
3. The `ComponentTaskPanel.removeElement` method has been simplified. It is
now a generic dispatcher that calls `handleComponentRemoval` on the
proxy of the object being edited, with no specific knowledge of object types.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* BIM: Add interactive creation of baseless walls
Introduce a new workflow for creating Arch Walls without a dependency on
a baseline object (e.g., a Draft Line).
- The `Arch_Wall` command is enhanced with a "No baseline" mode, controlled
by a new "Walls baseline" preference, allowing users to create
placement-driven walls directly in the 3D view.
- The existing `debaseWall` function has been refactored for correctness
and consistency with the new baseless wall geometry.
Co-authored-by: Yorik van Havre <yorik@uncreated.net>
* BIM: Refactor structure of the Arch Wall command
Refactor the `Arch_Wall` GUI command (`BimWall.py`) for improved
readability, maintainability, and architectural clarity.
- A `WallBaselineMode` Enum is introduced to replace the original
integer values, making the code self-documenting.
- The monolithic `create_wall` method is broken down into smaller,
single-responsibility helper functions for each creation mode.
- The `addDefault` method has been removed, with its logic
integrated into the new structure.
* BIM: Add Draft Stretch support for baseless walls
This commit makes the new baseless Arch Walls graphically editable using
the `Draft_Stretch` tool.
- An API for stretching (`calc_endpoints` and `set_from_endpoints`)
has been added to the `ArchWall` proxy.
- The `Draft_Stretch` tool is now aware of baseless walls and calls this
new proxy API to perform the stretch operation, enabling users to
stretch them.
Co-authored-by: Yorik van Havre <yorik@uncreated.net>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* BIM: address CodeQL warnings
* BIM: Fix wall alignment for GUI creation of baseless walls
Fixes an issue whereby creating a baseless wall from the GUI would
ignore the selected `Align` property, always resulting in a
center-aligned wall.
- The underlying geometry generation for baseless walls now correctly
honors the `Align` property passed by the GUI and API.
- To ensure predictable behavior, the implementation uses the same
geometric convention as walls built from a base object, making the
`Align` property work uniformly for all wall types.
- This also corrects the behavior of the `Arch.makeWall` function for
baseless walls.
- Update unit tests to test wall alignment.
* BIM: Refactor wall geometry generation for improved clarity and maintainability
Improves the internal logic for wall geometry creation, addressing CodeQL warnings and enhancing overall maintainability without changing external behavior.
- The `build_base_from_scratch` method is refactored to unify the separate logic paths for single- and multi-layer walls, reducing code duplication.
- A local helper function is introduced to create face geometry, for better modularity and readability.
- In the `_Wall.execute` method, the control flow that relied on implicit type checking is replaced with an explicit strategy pattern for fusing solids, making the logic more robust.
- Variable names are made more descriptive.
- A NumPy-style docstring is added to better document the function.
* Draft: fix stretching of rotated baseless walls
* BIM: add unit test for stretching baseless walls
* BIM: add regression tests for working-plane-relative coordinates and reuse of base sketches
* BIM: Fix baseless wall creation to respect the working plane
Corrects an issue where baseless walls were created using global
coordinates instead of being relative to the active Draft working plane.
The calculated local placement of the wall is now correctly transformed
into the global coordinate system by multiplying it with the working
plane's placement.
* BIM: Ensure unique baselines for subsequent wall creation
Fixes a bug where creating multiple walls with baselines would
incorrectly reuse the same underlying Sketch or Draft Line object.
The object retrieval logic after the `doCommand` call now correctly uses
`ActiveObject` to get a reliable reference to the new object instead of
relying on a hardcoded name.
* BIM: Make the wall's base object label translatable
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* BIM: use singular for consistency with other labels
Co-authored-by: Roy-043 <70520633+Roy-043@users.noreply.github.com>
* Fix typo
* BIM: address reviewer's comments about improving object reference passing between Python and FreeCAD contexts, and functions
* BIM: remove defensive programming: the callback is only executed as a result of a user's GUI action
* BIM: use the params API to define WallBaseline parameter
* BIM: add Arch Wall tests for joining wall logic
* BIM: add joining logic
* BIM: re-add ArchSketch support
* BIM: re-add multimaterial support on wall creation
* BIM: address CodeQL warning, remove module duplication
* BIM: fix check for SketchArch module when creating sketch-based walls
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Yorik van Havre <yorik@uncreated.net>
Co-authored-by: Roy-043 <70520633+Roy-043@users.noreply.github.com>
* Part: Fix Part_Extrude taper angle regression for internal faces
The taper angle for holes (inner wires) in `Part_Extrude` was not being
negated after the toponaming refactor, causing internal faces to taper
in the wrong direction compared to v0.21.2.
The original makeDraft function correctly handled inner wires by:
- Negating taper for all inner wires in Part_Extrude
- Negating taper only for multi-edge inner wires in PartDesign
This logic was controlled via an isPartDesign function parameter.
When makeElementDraft was introduced for toponaming support, it was
designed to use innerTaperAngleFwd/innerTaperAngleRev fields that were
never ported from realthunder's branch. The cleanup (commit c31ebeeee6)
removed references to these non-existent fields, leaving makeElementDraft
with no inner wire taper handling at all.
So, this fix ports the makeDraft inner wire logic to makeElementDraft by
adding the flag, and counting wires and then triggering proper angle
flip depending on the flag/wire situation.
---------
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
Arc segments with extremely large radius compared to chord length can
cause floating-point precision issues in simulation. This change
introduces ARC_LINEARIZATION_THRESHOLD to treat such arcs as linear,
improving numerical stability in the simulator.
The following g-code was causing erractic behavior in the simulator due
to large-radius arcs:
```
G2 I88775.835760 J-1936991.545892 K0.000000 X102.063107 Y100.102815 Z12.700001
```
src/Mod/CAM/PathSimulator/AppGL/MillPathSegment.cpp:
- Added ARC_LINEARIZATION_THRESHOLD constant
- Updated arc motion detection to treat large-radius arcs as linear
- Improved handling of arc vs. linear segment classification
* Add None check before constraining new_point
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Gui: Add common types to Add Property dialog
Add a preselection of commonly used types in the Add Property dialog.
The dialog supports now all properties that have an editor. This
doesn't necessarily mean that the editor is shown in the Add Property
dialog; some properties should not show their editor in the Add Property
dialog, such as vector and placement.
* Gui: Make Add Property dialog Qt 6 compatible
This change stops using a Qt 6.10 feature and makes it compatible with
all of Qt 6.
* Update src/Gui/Dialogs/DlgAddProperty.cpp
---------
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
* added a copy button which copies all the errors and traceback to clipboard
* removed unused variable
* removed space at end of Copy string
* copied text notification in status bar
* removed unwanted header
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Test: Address review comments
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
This is followup for https://github.com/FreeCAD/FreeCAD/pull/26563 which
fixed previews for a given file but broke it for others.
This approach should be more robust as it is applied in the Feature
itself and not on view provider level which should give us more precise
control and per feature transform ability.
This commit introduces PreviewUpdateScheduler class that is responsible
to schedule the true recompute of the preview. View Providers (or other
components) can use this service to ask for the preview recompute to
happend at a time that is convinent for a program and that won't impact
performance.
The provided implementation uses Queued Connections in Qt to calculate
preview essentially on next run of the event loop. It allows business
logic in FreeCAD (like property propagation) to execute fully and then
recompute preview once. This greately reduces number of recompute calls
for previews.
- The initial commit to block shortcuts in text edit fields
(particularly for MacOS, where those shortcuts would activate FreeCAD
functionality rather than text edit functionality) made a wrong
assumption.
- It assumed that there were no text edit fields where we would want to
actually use application shortcuts. In retrospect, this was very
wrong, and I completely missed the macro editor.
- For now, it should be fine to change the field to only cover
'LineEdit'. I cannot imagine a case where you'd want to (e.g.) save
text/document from a LineEdit, but if anyone knows of one then please
let me know.
- This does mean there are some quirks. For example in the materials
editor, the description is a TextEdit field. Some text editing
shortcuts won't work in here now (similar to how they didn't before
the original commit that I made).
- I've since learned that freecad also has a text editor functionality,
I've tested that now Cmd+S works for save as it should.
Major refactor of the Machine Editor to use QTabWidget for section
navigation. Added tabbed spindle management with add/remove
functionality, split machine configuration into Output Options, G-Code
Blocks, and Processing Options tabs. Updated preferences UI to use tabs
instead of QToolBox.
src/Mod/CAM/Gui/Resources/preferences/PathJob.ui:
- Replace QToolBox with QTabWidget for preferences tabs
src/Mod/CAM/Path/Dressup/Gui/Preferences.py:
- Use QWidget with vertical layout instead of QToolBox for dressup
preferences
src/Mod/CAM/Path/Machine/ui/editor/machine_editor.py:
- Refactor to use QTabWidget for editor sections
- Implement tabbed spindle management with add/remove
- Split configuration into Output Options, G-Code Blocks, and Processing
Options tabs
- Update post processor selection logic
src/Mod/CAM/Path/Main/Gui/PreferencesJob.py:
- Update to use tabWidget instead of toolBox
src/Mod/CAM/Path/Tool/assets/ui/preferences.py:
- Use QWidget and direct layout instead of QToolBox for asset
preferences
This recompute seemed unnecessary, and caused slowdowns when opening the file and recomputing. This decreases the time necessary to both open and recompute files that use boolean operations, and should not change anything regarding preview computation when the dialog is open.
* Adding retract annotation to drilling commands
checkpoint
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* CAM: Dropped trailing percent from fanuc post processor output.
A trailing percent cause one Fanuc controller to hang when the program
is completed when drip feeding G code, and the minicom ascii upload
was unable to complete the upload.
Had to use emergecy stop to get the machine out of the hang.
* CAM: Adjust Fanuc post processor test to no longer expect trailing percent.
* Patch for DXF without CODE30 (Z for POINT)
Time to time I use DXF file from measurement microscope without Z for POINT.
This patch help me for download this files without errors.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Removed explicit colons from all form labels.
Fixed lint warnings and replaced wildcard imports with explicit
imports for clarity and maintainability.
src/Mod/CAM/Path/Machine/ui/editor/machine_editor.py:
- Removed ':' from all form labels
- Fixed lint warnings (missing docstrings, empty except, etc.)
- Replaced wildcard imports with explicit imports
src/Mod/CAM/Path/Machine/models/__init__.py:
- Updated __all__ and imports for explicit API
src/Mod/CAM/Path/Tool/assets/ui/preferences.py:
- Updated imports to use package-level import
src/Mod/CAM/Path/Machine/models/machine.py:
- Added explanatory comments to empty except blocks
- Fixed duplicate variable assignment
- Added missing class docstrings
src/Mod/CAM/CAMTests/TestMachine.py:
- Fixed unused variable warning by using returned filepath
This PR introduces a machine object and a machine library, along with a
new machine editor dialog for creating and editing *.fcm machine asset
files. The editor is integrated into the CAM preferences panel, with new
Python modules for the dialog and minimal model validation. Machine
management (add, edit, delete) is now available in the CAM asset
preferences panel.
Key Features:
- Machines now have a type and units property. The machine type can be
used to distinguish between different classes of machines (e.g., mill,
lathe, laser).
- Machine units are stored internally and in the .fcm JSON file as
metric. Note: This differs from how toolbits work (which store units in
their native units)
- Support for 2-5 axis machines.
- Support for multiple spindles (up to 9)
- Processor defaults
- JSON Text Editor with basic validation and line numbers.
* Part: Fix orientation of internal shells of a solid
If a solid consists of multiple shells then it can happen that the
internal shells have incorrect orientation that leads to a broken solid.
This fixes issue https://github.com/FreeCAD/FreeCAD/issues/24994
* Part: Add method Feature::fixSolids
This method explicitly checks for solids with the error
'BRepCheck_EnclosedRegion' that means that internal shells of a solid
are wrong.
Using ShapeFix_Solid fixes the solid.
---------
Co-authored-by: wwmayer <wmayer@freecad.org>
* Gui: Handle exception in showValidExpression
If an exception is raised inside showValidExpression() because the
expression cannot be evaluated then handle the exception and show it as
invalid expression.
This fixes issue: https://github.com/FreeCAD/FreeCAD/issues/26501
---------
Co-authored-by: wwmayer <wmayer@freecad.org>
Currently, the selection batching optimization introduced in #26663
stores ElementItem/ConstraintItem pointers in a buffer to be processed
on the next event loop cycle via `QTimer::singleShot`.
When a constraint is applied,
`slotElementsChanged()/slotConstraintsChanged()`
rebuilds the list widget, destroying all items. However, the selection
buffer was not cleared, leaving dangling pointers. When
`processSelectionBuffer()`
executed, it attempted to call `QListWidget::row()` on deleted items,
causing a segmentation fault.
So, this patch fixes it by clearing `selectionBuffer` when the list is
rebuilt, since the buffered pointers are invalidated anyway.
This reverts commit 527b2de560.
The `bakeInTransform()` call uses `transformGeometry()` which converts
planar faces to BSplineSurfaces. This prevents the refine algorithm
(BRepBuilderAPI_RefineModel) from detecting and merging coplanar faces,
breaking the Refine option for Boolean operations.
This matches Part WB's Boolean behavior which does not use
`bakeInTransform()` and has working refine functionality.
Fixes the following issue:
```
Warning:
/Users/runner/work/FreeCAD/FreeCAD/src/3rdParty/FastSignals/libfastsignals/include/fastsignals/connection.h:76:32:
warning: macro 'ATOMIC_VAR_INIT' has been marked as deprecated
[-Wdeprecated-pragma]
76 | std::atomic<bool> m_blocked = ATOMIC_VAR_INIT(false);
| ^
```
* Sketcher: Support Bezier and Offset curves as external geometry
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Sketcher: Fix adding projection of external arc of circle
In processEdge() it's checked if first and last point of the projected arc of circle are almost equal. If yes then a full circle
is assumed but this conclusion is not necessarily correct because it's still possible that the external arc of circle is very
short. The better criterion is to check the parameter range of the arc of circle.
See also: https://github.com/FreeCAD/FreeCAD/pull/25144#issuecomment-3502916685
---------
Co-authored-by: wwmayer <wmayer@freecad.org>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* PartDesign: Enable drag and drop of shapeBinders
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Add diameter measurement support
* change author name and change icon
* correct svg size to 64x64
* Update src/Mod/Measure/App/MeasureDiameter.cpp
Co-authored-by: Kacper Donat <kadet1090@gmail.com>
* somehow to icon color is undefine in inkscape, change to black
* Revert "somehow to icon color is undefine in inkscape, change to black"
This reverts commit 2277c1b9f4a7ab7519856986e2d6aec6e37ebf3e.
modified: src/Mod/Measure/Gui/Resources/icons/Measurement-Diameter.svg
* fix black color icon
---------
Co-authored-by: Kacper Donat <kadet1090@gmail.com>
Add `SoDepthBuffer` nodes around the constraint group to disable depth
testing for constraint icons. This ensures icons render on top of
geometry lines regardless of Z position, improving visibility and
selectability.
Add Z offset for arrowheads and explicitly enable depth testing for
constraint lines in SoDatumLabel. This ensures constraint lines render
below geometry (respecting zConstr level) while text and arrowheads
render on top for better visibility and selection.
Changes include:
1. Modernize `for` loops with range whenever possible.
2. Flip `if` statements for a "flatter" flow.
3. Use internal functions for repeated identical tasks.
* [TD]fix dim text on wrong side of dim line
- this fixes the problem when the dimension line is +/- vertical
* [TD]Avoid use of CosmeticVertex for Axo Dimension
* [TD]fix CI warnings re unused variable
* Update DrawViewPyImp.cpp per review comments.
The `lupdate` tool from Qt used to extract translations from files was
not in sync with what the macros defined in the code. The mismatch came
from two places:
1. The DrawSketchHandeler used non fully-qualified name for the
context whereas lupdate assumed the FQN.
2. Deriving DrawSketchHandlers did not override the translate method
context to their own class names while lupdate assumed that they do.
While it's not fully clear if what `lupdate` does here is correct
(it's a lot of assumptions) it's way easier to fix our metadata than
fight with lupdate.
* [Surf]provide geometry to Measure
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Sometimes when user would switch from editing one sketch to edit to another sketch by double clicking on the new sketch, without living the previous sketch, it would then render the text of constraints backward. This was happening because the unsetEdit of the previous sketch was clearing the selection and adding the old sketch as selected. Then the new setEdit was failing to find the correct editing placement resulting in backward text.
* PD: fix gizmo direction when the calculated point lies outside the face
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
- Offset and None cut Pattern error
- Ensure the tool is directly above the entry point before plunging,
- Retract to SafeHeight instead of ClearanceHeight between transitions
This commit applies a simple fix to the V-bit tool definition, adding a
small offset (+0.05 um) to the tip diameter. This prevents the sketch
from generating invalid constraints due to a zero or near-zero tip size.
src/Mod/CAM/Path/Op/SurfaceSupport.py:
- Add +0.05 um to tip diameter calculation to avoid constraint errors in
sketches
Since some time we have new stylesheets based on style parameters. For
compatibility reasons however old stylesheets were left in the project -
this commit removes them by reapplying the theme if old stylesheet is
detected.
Addresses problems caused by the "Allow duplicate object labels in one
document" option. Toolbit sub-objects now always get unique labels by
temporarily disabling this option during copy, preventing expression and
property binding errors. Also ensures tool controllers always use the
"TC: " prefix for consistency.
src/Mod/CAM/Path/Tool/shape/models/base.py:
- Temporarily disable DuplicateLabels during shape copy to enforce
unique labels
src/Mod/CAM/Path/Main/Gui/Job.py:
- Add "TC: " prefix to tool controller names when adding from the Job
panel
* Gui: Add hidden anchor object to the root for transparency
Image planes with transparency failed to render correctly in empty scenes
because OpenInventor's two-pass transparency rendering requires at least
one opaque object to properly initialize the depth buffer.
The fix adds a zero-scaled cube with no material node (making it use
OpenGL's default opaque material) to each image plane's scene graph.
This hidden object:
- Acts as a depth buffer anchor for transparent rendering
- Is invisible (scaled to 0,0,0)
- Has negligible performance impact
This matches the workaround already used in the rotation center indicator
and resolves the issue where image transparency only worked when the
rotation center, grid, or other opaque objects were visible.
* Gui: Exclude hidden anchor from bounding box calculations
Prevents the hidden anchor from affecting "fit all" and other bounding box
operations by wrapping it in `SoSkipBoundingGroup`.
* Part: Toposhape: fix regressions due to changes in getElementTypeAndIndex
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Update TopoShape.cpp
* Update TopoShape.cpp
* Update TopoShape.cpp
* Update TopoShape.h
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
When snapping to an axis, snapToObject returns false because we should also be able to snap to grid at the same time. Recent changes made the end return to be the initial unmodified position. Breaking snap to axis
* Gui: Prevent whole-object highlight when picked list is enabled
Removed the logic that forced `onTop=1` when `needPickedList()` is true
in View3DInventorSelection. This was causing the entire object to be
highlighted instead of just the selected sub-element when the "Picked
object list" option was enabled.
* Part: Remove `needPickedList` mat override to prevent rendering artifact
The `needPickedList` check in `SoBrepFaceSet` was triggering
unnecessary material override processing that caused face clipping
artifacts when the "Picked object list" option was enabled in Selection
View. This check has been removed as the picked list functionality works
independently of the rendering path.
* Sketcher: Remove old constraint positioning logic that was making sketcher laggy
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
The `needPickedList` check in `SoBrepFaceSet` was triggering
unnecessary material override processing that caused face clipping
artifacts when the "Picked object list" option was enabled in Selection
View. This check has been removed as the picked list functionality works
independently of the rendering path.
Removed the logic that forced `onTop=1` when `needPickedList()` is true
in View3DInventorSelection. This was causing the entire object to be
highlighted instead of just the selected sub-element when the "Picked
object list" option was enabled.
* Part: Preserve child visibility when deleting compound copies
Currently, if we have a FC document two compounds, when one of them is a
copy of the other containing same children and we delete the copy
compound to keep the children, their visibility state changes.
The cause of that was a part of code in `onDelete()` method, which was
unconditionally calling `showViewProvider()` on every child object,
without checking if other compounds still claim those children or what
the current visibility should be.
So this patch adds parent-checking logic before changing visibility. The
visibility is only updated if the child is truly orphaned. So in the
scenario where we have the children referenced STILL by some compounds,
the visibility remains unchanged. It only changes when child has truly
no parents.
* Part: Use std::ranges::find for finding parent
Change `ViewProviderPart` to inherit from `ViewProviderGeometryObject`,
giving Part containers the Selectable property. This allows recursive
selectability checks to respect the entire container hierarchy.
When using the default GitHub token, no further Actions are triggered, which means a backport PR is not checked against the normal CI runs, and it should be. This switches to using a token generated by the freecad-ci-runner GitHub account.
This could be a return of parts of commit d4feb51402
Without these, the rc1 build fails on FreeBSD with "pybind11/eigen.h not
found" - the layout of the pybind11 directories on the BSDs may differ
somewhat from your typical Linux distribution. But then, on Linux, this
additional include path will just point to an already-known location.
- Add missing <title> element in <head> (required by XHTML Strict)
- Add <body> wrapper around content (required by XHTML)
- Add Content-Type meta tag for proper character encoding
- Fix indentation of buttons to be inside body element
After discussion in several recent PR review meetings the Maintainer
team has decided to roll back some recent work on object naming in STEP
files. Inadequate automated testing has led to a situation where every
change made to address one bug introduces another someplace else. It
is difficult to determine which, if any, of these fixes represent an
appropriate path forward, versus fixing a symptom rather than the
underlying problem.
---
Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks"
This reverts commit 59715114f5.
Revert "core: fix various issues exporting step files, ie. #25540#24962#25567"
This reverts commit e226e9b06e.
Revert "export: step file preserve last feature when exporting step file, fix issue #25567"
This reverts commit 9c48c9a3e3.
Revert "core: preserve body name in step file when exporting part that contains body with feature, fixes#24962"
This reverts commit 246218b28b.
Revert "Import: STEP export preserve body names (#25042)"
This reverts commit f218c5f25c.
Revert "Import: Export tip when body is selected (#24678)"
This reverts commit 6fd6558040.
Currently if user selects object with `Selectable=false` property which
is nested inside a Part container, for example `Part->Body->Pad`, parent
Part container will get highlighted in the 3D view, even though the user
clicked on a non-selectable child object.
Cause of that is that tree view's `getTopParent()` function resolves
nested objects selections to their top-level container. When a Pad
inside Part is selected, `SoFCUnifiedSelection` was only checking Part's
`isSelectable()` status, even though the target is `Pad`.
So the fix is to resolve subname to get the final target object and
checks its `isSelectable()` status before checking for highlighting.
* Core: Fix completer popup adjustment after 'Tab' pressing
Completer popup in Expression Editor is now positioned and sized after pressing 'Tab'. This was missing when feature was added.
* Core: Extend completer popup list to up 20 elements or up to screen edge
Completer popup will now show up to 20 elements or will be limited to screen edge; if limited < 3 rows, popup will be displayed over cursor.
* Core: Fix completer wrapping on secondary screens
* Core: Fix completer positioning when includes long entries
* Core: Fix linter complaints
* sketcher: fix issue #25992 keep snapped position when releasing mosue for updated sketch points
* Update src/Mod/Sketcher/Gui/ViewProviderSketch.cpp
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
---------
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
- Any action part of the application menu will trigger based on shortcut
when a native file dialog is open
- Another way to fix this is to switch out the application menu, but
afaict it requires writing native objective-c code. I think that's too
much complexity just to get cmd+c/cmd+v in these file dialogs
- For now, just disable the actions so that select-all, rotate-left, etc
don't trigger when pressed in these dialogs
- I've implemented an RAII wrapper to disable this. It should take
pointers which should be fine because all of these dialog calls are
blocking (so in principle nothing can change underneath).
I'm quite sure this won't have any adverse effects on other platforms,
but will need help from other developers to test in case.
The current code in \src\Mod\CAM\Path\Main\Gui\Job.py (line 719) ignores selected stock selections. In case of e.g. cloned stock objects it defaults to the first item in the combolist, showing a wrong stock selection.
The cause lies in comparing different strings to find a label match, but one string object is generic (short name) while the other one has a suffix created during the clone process. The strings never match and the ridgid string comparison fails. I recommend using a partial (needle in haystack) string comparison.
This PR adds migration for toolbit units by automatically infering the
Units (Metric/Imperial) from toolbit parameter strings and sets the
Units property if missing. It adds a utility function to detect units
from JSON. This is done at the JSON level during migration to ensure
that all toolbits have the correct Units property set.
src/Mod/CAM/Path/Tool/toolbit/migration.py:
- Infer Units from parameter strings if not set during migration
- Set Units property and log inference
- Refactor migration logic for clarity and reliability
src/Mod/CAM/Path/Tool/toolbit/models/base.py:
- Use Path.Log.debug instead of print when adding Units property
src/Mod/CAM/Path/Tool/toolbit/util.py:
- Add units_from_json() to infer Metric/Imperial from parameter strings
- It seems that on MacOS (vs other platforms), shortcuts for items in
the application menu are given 'ultimate' priority, and will even take
precedence over text inputs
- There is a mechanism in QT (I believe designed with mac in mind) to
try to 'block' these shortcuts and send it to the focus instead. It's
'Shortcut Override': https://wiki.qt.io/ShortcutOverride
- Initially I was going to only apply this check when it's a Command
that overrides a known line-editing shortcut, but I figure it's
simpler to just always apply it when editing text. I can't really
imagine a user wanting to use an application shortcut while editing
text, but if there's some compelling use-case for this then let me know
and I'll add a further filter.
I'm quite optimistic that this won't have any ill-effects on other
platforms, but I'll need help from others to test this.
* Fix rotation expression handling
- Make added rotation angle unit the same as the original expression unit
- Keep rotated angle expressions within the accepted [-180, +180] range
In older versions of FreeCAD the boolean order was base + result. After
TNP mitigation the order was changed to be result + base. For the
resulting shape that does not matter, but order of edges can differ if
arguments are in a different order.
This can impact refine algorithm which may pick other face as the base
one and result in a differnt shape after refining. This commit restores
previous order. For most files it should not make any difference, but it
may fix some older files.
To support all cases we introduce FuseOrder compatibility property that
will be set to FeatureFirst for files saved with 1.0 to preserve
behavior.
* Sketcher: add reverse mapping correction to Carbon Copy
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Replace M_PI with std::numbers::pi
* Replace vector initialization with Base::Vector3d::UnitX/UnitY, apply suggestions from code review
Co-authored-by: Kacper Donat <kadet1090@gmail.com>
* Fix guard clause logic, apply suggestions from code review
* Replace std::numbers::pi with pi via `using std::numbers`
* Fix issues reported by the linter
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Kacper Donat <kadet1090@gmail.com>
* part design: partdesign: pd: fix regression issue #26223 add sketch sub element names for refs in revolutions
* refractor code to remove else blocks as every condition has a return statement, make lsp happy
Using these commands by mistake can result in broken interface and the
shortcuts are way too easy to hit by mistake. The feature is rarely (if
at all) used - so there is no need to have easy to hit shortcuts.
Calling BRepMesh_IncrementalMesh repeatedly would accumulate BRep_CurveRepresentation instances attached to the BRep_TEdges of the shape. We now clean added triangulations before adding a new one, which improves performance in the long run.
Co-authored-by: Kacper Donat <kadet1090@gmail.com>
* BIM: fix deletion of BuildingPart in strict IFC mode
Add `sg = FreeCADGui.ActiveDocument.ActiveView.getSceneGraph()` to `onDelete`.
* BIM: fix deletion of BuildingPart in strict IFC mode
Explicitly call `obj.ViewObject.Proxy.attach(obj.ViewObject)`.
* BIM: fix deletion of BuildingPart in strict IFC mode
In `ifc_vp_buildingpart` call `attach` from `ArchBuildingPart.ViewProviderBuildingPart`.
* SVG export: use a set to match edges instead of a list
Lookup in edge list becomes O(1), significantly improving SVG export speed
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
The ApplicationDirectories::findHomePath() on BSD uses sysctl()
to find the path to the running executable. On FreeBSD, we need
to include sys/sysctl.h for that - and as it is not included anywhere
else, add it to the includes directly in front of this function,
with a suitable ifdef.
* Part: fix Part_EditAttachment nesting handling
Fixes#26264.
The previous 'fix' (#25887) did not consider that the object to-be-attached may not be in the global space.
The code in this PR has been tested on the forum:
https://forum.freecad.org/viewtopic.php?p=862968#p862968
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Improves the point-based TSP solver to match tunnel solver behavior
for open routes (no endpoint constraint). Now applies 2-opt and
relocation optimizations that allow reversing or relocating segments
to the end of the route, resulting in better path optimization when
the ending point is flexible. Now links tsp_solver with
Python3::Python and uses add_library for compatibility with FreeCAD
and Fedora packaging.
src/Mod/CAM/App/tsp_solver.cpp:
- Add optimization limit variables for controlled iteration
- Add 2-opt and relocation optimizations for open routes
- Use Base::Precision::Confusion() for epsilon values
- Track last improvement step for efficient loop control
src/Mod/CAM/App/CMakeLists.txt:
- Switch tsp_solver from pybind11_add_module to add_library
- Link tsp_solver with pybind11::module and Python3::Python
- Update include directories for consistency
This fixes segfault that was caused by changing settings in
FirstStartWidget that triggered change event on not fully intialized
StartView.
It is quick and dirty fix that just disables handling of the event
causing segfault until widget is fully ready. The correct solution would
be to remove the logic that changes settings from the widget (as it
should not be there) and instad move it to another place in pipeline.
This would be however much more work and segfault should be fixed ASAP.
Update TSP tunnel solver to match revised Pythonlogic, including
improved early exit, open-ended route handling, and performance tweaks.
Also ensure extra data in tunnel dictionaries is preserved through the
C++/Python interface and add a test for passthrough of extra keys.
src/Mod/CAM/App/tsp_solver.cpp:
- Refactor optimization loop to use lastImprovementAtStep and limit
variables
- Add special handling for open-ended routes (no end point)
- Change epsilon to 10e-6 for consistency with Python
- Cache distance calculations for relocation step
src/Mod/CAM/App/tsp_solver_pybind.cpp:
- Preserve extra keys from input tunnel dicts in output
- Set tunnel index for passthrough
src/Mod/CAM/CAMTests/TestTSPSolver.py:
- Add test to verify extra data in tunnel dicts is preserved through
TSP solver
- Print extra data for debugging
Ensures that all extra keys in tunnel dictionaries are preserved after
TSP solving by copying the original input dict and updating solver
results. Adds a dedicated test to verify passthrough of extra data and
prints extra keys for debugging.
src/Mod/CAM/App/tsp_solver_pybind.cpp:
- Copy original tunnel dict and update with solver results to preserve extra keys
src/Mod/CAM/CAMTests/TestTSPSolver.py:
- Add test_09_tunnels_extra_data_passthrough to verify extra data preservation
- Print extra tunnel data in print_tunnels for easier debugging
The 'old' DiffuseColor comes in 2 flavors. In v0.21.2 it stores transparency values. In v1.0.2 and v1.1 it stores alpha values. The ArchReference code should handle both. DiffuseColor is still used by BuildingParts in v1.0.2 and v1.1.
The solution is a bit hacky: if 100% transparency occurs we assume we are actually dealing with alpha values.
This is a hacky fix for https://github.com/FreeCAD/FreeCAD/issues/23607
Basically after widget is shown or polished we enforce it's minimum size to at
least cover the minimum size hint - something that QSS ignores if min-width is
specified.
This introduces much faster CenterOfMassProvider#supports method that
can be used as cheap pre-check to see if the object supplied could have
the center of mass and so delay the domputation to a moment where it is
actually needed.
Added extensive test coverage for the TSP tunnel solver including linear
tunnels, pentagram diagonals, and complex wire geometries with various
constraint combinations. Introduced a Python wrapper function for
tunnel sorting that integrates with the C++ solver.
src/Mod/CAM/CAMTests/TestTSPSolver.py:
- Renumbered existing tests to run in sequential order
(test_01_simple_tsp, test_02_start_point, etc.)
- Added print_tunnels() helper function for displaying tunnel
information with original indices
- Added test_06_tunnels_tsp: Tests 7 linear tunnels with varying
lengths, connectivity, and flipping behavior
- Added test_07_pentagram_tunnels_tsp: Tests pentagram diagonal tunnels
with no constraints, start+end constraints, and start-only constraints
- Added test_08_open_wire_end_only: Tests end-only constraint on complex
wire with 8 tunnels including crossings and diagonals
src/Mod/CAM/PathScripts/PathUtils.py:
- Added sort_tunnels_tsp() wrapper function that interfaces with the C++
tsp_solver.solveTunnels()
- Supports allowFlipping, routeStartPoint, and routeEndPoint parameters
This is a quick and dirty fix for https://github.com/FreeCAD/FreeCAD/issues/22010
Basically if we want to use a bigger marker size than available, we use the biggest one
available. This is not a good way to fix the issue - we should ensure that the marker
size that the user requests is actually available - this, however, requires more
significant changes to the code.
Introduce TSPTunnel struct and implement TSPSolver::solveTunnels
for optimizing tunnel order with support for flipping and start/end points.
Expose the new functionality to Python via pybind11, returning tunnel
dictionaries with flipped status.
src/Mod/CAM/App/tsp_solver.cpp:
- Add solveTunnels implementation for tunnel TSP with flipping and route endpoints
src/Mod/CAM/App/tsp_solver.h:
- Define TSPTunnel struct
- Declare solveTunnels static method in TSPSolver
src/Mod/CAM/App/tsp_solver_pybind.cpp:
- Add Python wrapper for solveTunnels
- Expose solveTunnels to Python with argument parsing and result conversion
* BIM: BIM Report MVP
* BIM: add Arch.selectObjects() API call and tests
* BIM: adopted ArchSchedule spreadsheet update and linking patterns
* BIM: SELECT with dynamic headers and dynamic data population
* BIM: fix deletion, dependencies, and serialization issues
* BIM: transition from individual query to multiple-query statements
* BIM: Mostly fix serialization issues
* BIM: Report editor UI/UX improvements
- Make edits directly on table (Description field only, for now)
- Sync table edits with query editor (Description field only, for now)
- Fix Property group name for spreadsheet object
- Add tooltips for all task panel widgets
- Put Description input and label on the same row
- Edit statement is now a static title
- Open report editor upon creation
- Remove overview table's redundant edit button
- Put report query options in a box
- Open spreadsheet after accepting report changes
* BIM: Aggregarion and grouping - implement GROUP BY with COUNT(*)
* BIM: Aggregation and grouping - implement SUM, MIN, MAX functions
* BIM: Aggregation and grouping - implement validation
* BIM: add reporting presets support
* BIM: align description and query preset inputs vertically
* BIM: write units on their own spreadsheet cell
* BIM: update test suite to new SQL engine behaviour
* BIM: fix various bugs: return values should not be stringized, non-grouped query with mixed extractors, handle nested properties
* BIM: expand Report test suite, fix property count vs count all bug
* BIM: add report engine SQL's IN clause support
* BIM: enable running Report presets tests from the build directory
* BIM: make spreadsheet more useful for analysis, units on headers
* BIM: update BIM Report icons
* BIM: Add key option columns to report overview table
* BIM: add syntax highlighting support to report query editor
* BIM: Add lark build dependency for SQL parser generator
* BIM: Install the generated parser
* BIM: implement TYPE() function
* BIM: simplify function registry, make it OOP
* BIM: add CHILDREN function
* BIM: improve SQL engine error descriptions
* BIM: Implement ORDER BY clause
* BIM: Implement column aliasing (AS...)
* BIM: improve error reporting to pinpoint exact token names
* BIM: implement CONCAT, LOWER, UPPER functions. Improve exception handling
* BIM: refactor to improve initialization readability and maintenance
* BIM: Improve query editor syntax highlighting
* BIM: Address CodeQL warnings
* BIM: Enable scalar functions in WHERE clause and refactor transformer
- Enable the use of scalar functions (e.g., `LOWER`, `TYPE`) in `WHERE` clause comparisons.
- Refactor the Lark transformer to correctly handle `NUMBER`, `NULL`, and `ASTERISK` terminals using dedicated methods.
- Refactor error handling to raise a custom `BimSqlSyntaxError` exception instead of returning values on failure.
- Improve syntax error messages to be more specific and user-friendly by inspecting the failing token.
- Fix regressions in `AggregateFunction` and `TypeFunction` handling that were introduced during the refactoring.
- Update the test suite to assert for specific exceptions, aligning with the new error handling API.
* BIM: Implement arithmetic operations in SELECT clause
- Update grammar with `expr`, `term`, and `factor` rules to support operator precedence.
- Introduce `ArithmeticOperation` class to represent calculations as a recursive tree.
- Add transformer methods to build the calculation tree from the grammar rules.
- Implement recursive evaluation that correctly normalizes `Quantity` and `float` types before calculation.
* BIM: Add CONVERT() function for unit conversions
- Implement a new ConvertFunction class to handle unit conversions in the SELECT clause.
- Leverage the core Quantity.getValueAs() FreeCAD API method to perform the conversion logic.
- Add a unit test to verify both successful conversions and graceful failure on incompatible units.
* BIM: add self-documenting supported SQL syntax helper
* BIM: document internal and external API functions
* BIM: Finalize and rename public ArchSql API, update all consumers
- Rename public API functions for clarity and consistency
(run_query_for_objects -> select, get_sql_keywords -> getSqlKeywords,
etc.).
- Establish Arch.py as the official API facade by exposing all public
SQL functions and exceptions via a safe wildcard import from ArchSql,
governed by __all__.
- Refactor all consumers (ArchReport.py, bimtests/TestArchReport.py) to
use the new function names and access the API exclusively through the
Arch facade (e.g., Arch.select).
- Finalize the error handling architecture:
- Arch.select now logs and re-raises exceptions for consumers to
handle.
- Arch.count remains "safe," catching exceptions and returning an
error tuple for the UI.
- Refactor the test suite to correctly assert the behavior of the new
"unsafe" (select) and "safe" (count) APIs, including verifying
specific exception messages.
* BIM: add results preview on demand on the query editor
* BIM: add documentation quick reference button
* BIM: Handle query errors gracefully in Report preview UI
Refactor the SQL engine's error handling to improve the Report preview
UI. Invalid queries from the "Preview Results" button no longer print
tracebacks to the console. Instead, errors are now caught and displayed
contextually within the preview table, providing clear feedback without
console noise.
- `ArchSql.select()` no longer logs errors to the console; it now only
raises a specific exception on failure, delegating handling to the
caller.
- The `ReportTaskPanel`'s preview button handler now wraps its call to
`Arch.select()` in a `try...except` block.
- Caught exceptions are formatted and displayed in the preview table
widget.
- Add new unit test
* BIM: consolidate internal API into a more semantically and functionally meaningful function
* BIM: Implement two-phase execution for SQL engine performance
Refactors the ArchSql engine to improve performance and internal API
semantics. The `Arch.count()` function, used for UI validation, now
executes faster by avoiding unnecessary data extraction.
This is achieved by splitting query execution into two phases. The first
phase performs fast filtering and grouping (`FROM`/`WHERE`/`GROUP BY`).
The second, slower phase processes the `SELECT` columns. The
`Arch.count()` function now only runs the first phase, while
`Arch.select()` runs both.
- Introduces `_run_query(query_string, mode)` as the mode-driven
internal entry point for the SQL engine
- The `SelectStatement` class is refactored with new internal methods:
`_get_grouped_data()` (Phase 1) and `_process_select_columns()`
(Phase 2).
- `Arch.count()` now uses a fast path that includes a "sample execution"
on a single object to correctly validate the full query without
performance loss.
- The internal `execute()` method of `SelectStatement` is now a
coordinator for the two-phase process.
* BIM: Implement autocompletion in Report SQL editor
Introduces an autocompletion feature for the SQL query editor in the BIM
Report task panel. The completer suggests SQL keywords, functions, and
property names to improve query writing speed and accuracy.
- Adds a custom `SqlQueryEditor` subclass that manages all
autocompletion logic within its `keyPressEvent`.
- The completion model is populated with static SQL keywords and
functions, plus all unique top-level property names dynamically
scanned from every object in the document.
- A blocklist is used to filter out common non-queryable properties
(e.g., `Visibility`, `Proxy`) from the suggestions.
- The editor manually calculates the completer popup's width based on
content (`sizeHintForColumn`) to resolve a Qt rendering issue and
ensure suggestions are always visible.
- The first suggestion is now pre-selected, allowing a single `Tab`
press to accept the completion.
* BIM: remove unused import
* BIM: support SQL comments in queries
* BIM: support non-ASCII characters in queries
* BIM: Allow ORDER BY to accept a comma-separated list of columns for multi-level sorting.
* BIM: fix two-way overview/editor sync
* BIM: refactor to simplify editor modification events
* BIM: add tooltips to overview table
* BIM: add tooltips to query editor, enrich syntax data with signature and snippets
* BIM: implement PARENT function
* BIM: Enable property access on SQL function results
Previously, the SQL engine could only access properties directly from
the main object in a given row. This made it impossible to query
attributes of objects returned by functions, such as getting the name of
a parent with PARENT(*). This commit evolves the SQL grammar to handle
chained member access (.) with correct operator precedence. The parser's
transformer and execution logic were updated to recursively resolve
these chains, enabling more intuitive queries like SELECT
PARENT(*).Label and standard nested property access like Shape.Volume.
* BIM: refactor function registration, internationalize API metadata
* BIM: remove outdated Report alias
* BIM: refactor selectObjects to use latest API, move implementation to ArchSql
* BIM: improve friendly token names for error reporting
* BIM: implement chained functions e.g. PARENT(*).PARENT(*)
* BIM: add further tests for property access, fix bug with non-literal AS clause argument
* BIM: Implement full expression support for GROUP BY
The SQL engine's GROUP BY clause was previously limited to simple
property names, failing on queries that used functions (e.g., `GROUP BY
TYPE(*)`). This has been fixed by allowing the SQL transformer and
validator to correctly process function expressions.
The `SelectStatement.validate()` method now uses a canonical signature
to compare SELECT columns against GROUP BY expressions, ensuring
correctness for both simple properties and complex functions.
New regression tests have also been added to validate `GROUP BY`
functionality with chained functions (PPA) and multiple columns.
* BIM: Make arithmetic engine robust against missing properties
Previously, a query containing an arithmetic expression in the `WHERE`
clause would cause a fatal error if it processed an object that was
missing one of the properties used in the calculation.
This has been fixed by making the `ArithmeticOperation` class NULL-safe.
The engine now correctly handles `None` values returned by property
lookups, propagating them through the calculation as per standard SQL
behavior.
As a result, the engine no longer crashes on such queries. It now
gracefully evaluates all objects, simply filtering out those for which
the arithmetic expression cannot be resolved to a valid number. This
significantly improves the robustness of the query engine.
* BIM: Finalize ORDER BY logic and fix query regressions
- The `ORDER BY` clause is now governed by a single, predictable rule:
it can only refer to column names or aliases that exist in the final
`SELECT` list. The engine's transformer and execution logic have been
updated to enforce and correctly implement this, fixing several test
failures related to sorting by aliases and expressions.
- Updated test suite to align with the new engine rules. Add new
dedicated unit tests that explicitly document the supported (aliased
expression) and unsupported (raw expression) syntax for the `ORDER BY`
clause.
* BIM: Implement backend for pipelined report execution
Adds the backend architecture for multi-step, pipelined queries. The
core SQL engine can now execute a statement against the results of a
previous one, enabling complex sequential filtering.
- The internal `_run_query` function was refactored to be
pipeline-aware, allowing it to operate on a pre-filtered list of
source objects.
- A new `execute_pipeline` orchestrator was added to manage the data
flow between statements, controlled by a new `is_pipelined` flag in
the `ReportStatement` data model.
- The public API was extended with `selectObjectsFromPipeline` for
scripting, and `count()` was enhanced to support contextual validation
for the UI.
Pipelines phase 2
Pipelines phase 3
* BIM: refactor to avoid circular imports
* BIM: Address key CodeQl check errors/notes
* BIM: Refactor Task Panel UI/UX with explicit editing workflow
Refactor the report editor UI to improve usability and prevent data
loss. Editing a statement is now an explicit action, triggered by a new
"Edit Selected" button. This prevents accidental changes and enables a
new transactional workflow with "Save" and "Discard" buttons for each
edit session. A checkbox for "Apply & Next" has been added to streamline
editing multiple statements.
The results preview feature has been redesigned into a self-contained,
closable pane controlled by a "Show Preview" toggle. This pane includes
its own contextual "Refresh" button, reducing UI clutter. All action
button groups have been consistently right-aligned to improve layout and
workflow.
* BIM: Integrate pipeline execution and stabilize UI workflow
Completes the implementation of the pipelined statements feature by
integrating the new backend orchestrator with the ArchReport object. It
also includes a series of critical bug fixes that were discovered during
end-to-end testing, resulting in a more stable user experience.
The primary feature integration consists of refactoring the
_ArchReport.execute method. It now uses the ArchSql.execute_pipeline
generator, enabling the report to correctly process multi-step pipelines
and honor the user-configured data flow.
Severalbugs and regressions were fixed:
- Backend: A major flaw was fixed where FROM-clause functions (like
CHILDREN) were not pipeline-aware. The engine's GROUP BY validator was
also corrected to enforce its original, safer design of not supporting
aliases.
- UI Workflow: A feedback loop that caused the editor cursor to reset
was resolved. The transactional logic for the Save, Discard, and "Save
& Add New" actions was corrected to prevent data loss and ensure
predictable behavior. The Add and Duplicate actions no longer auto-open
the editor, creating a more consistent workflow.
- UI State: Fixed regressions related to the explicit editing model,
including incorrect statement loading (Edit Selected) and state
management when a report is reloaded from a file.
* BIM: add presets manager
* BIM: Fix 'still touched after recompute' bug in some of the tests
* BIM: cleanup tests, fixed presets tests to new presets locations
* BIM: Move test model to its own module for reusability
* BIM: Move GUI tests to their own module
- Disable two of the tests: they pass, but cause a segmentation fault on
teardown. They need to be investigated before enabling them on CI.
* BIM: fix bug in interpreting CONVERT string from GROUP BY
* BIM: Migrate signal connections from lambdas to Qt slots
Refactors all signal connections in the `ReportTaskPanel` to use the
`@QtCore.Slot()` decorator instead of lambda wrappers.
- Resolves CodeQL warning "Unnecessary lambda".
- Adheres to PySide/Qt best practices for signal/slot handling.
- Improves code clarity by using explicit, named methods for
connections.
- Prevents potential runtime `TypeError` exceptions by correctly
managing slot signatures for signals that emit arguments (e.g.,
`clicked(bool)`).
- Introduces simple slot wrappers where necessary to cleanly separate UI
event handling from core logic.
* BIM: Address CodeQl warnings
* BIM: Add CHILDREN_RECURSIVE(subquery, max_depth) SQL function to find all descendants of an object set.
- Create internal _get_bim_type and _is_bim_group helpers to remove all Draft module dependencies from the SQL engine.
- Implement a new _traverse_architectural_hierarchy function using a deque-based BFS algorithm to prevent infinite loops.
- The CHILDREN and CHILDREN_RECURSIVE functions now use the new traversal engine.
- The traversal engine now transparently navigates generic groups but excludes them from the results.
- Correct ParentFunction logic to validate containment by checking the parent's .Group list instead of the child's .InList.
- Add unit tests for recursive traversal, depth limiting, and transparent group handling.
- Update test_group_by_multiple_mixed_columns to match the corrected behavior of the TYPE(*) function.
* BIM: Add __repr__ and __str__ for the ArchReport proxy
* BIM: Align report quantity headers with internal units
- Change default `Quantity` headers to use the property's internal unit
(e.g., `mm`) instead of the user's global preferred unit (e.g., `m`).
- Fixes inconsistency where the header unit (e.g., `m²`) did not match
the raw data's unit (e.g., `mm²`), making the default behavior
predictable.
- Implement by parsing `str(Quantity)` as a workaround for the C++
`Unit::getString()` method not being exposed to the Python API.
- Add a unit test that temporarily changes the global schema to verify
the fix is independent of user preferences.
* BIM: remove dual import ArchSql and from ArchSql import
* BIM: Fix IfcRole => IfcType
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* CI: add Lark as a build dependency
* BIM: Replace QRegExp with QRegularExpression for regex patterns
QRegExp is no longer available in PySide6. Make the code compatible with
both PySide2/Qt5 and PySide6/Qt6.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* CI: move lark dependency to pixi's requirements.host
* BIM: Correct numeric comparisons in SQL WHERE clause
Refactor the `BooleanComparison.evaluate` method to handle numeric and
string-based comparisons separately. Previously, all comparisons were
being performed on string-converted values, causing numerically
incorrect results for `Quantity` objects formatted with "smart" units
(e.g., `"10.0 m"` was incorrectly evaluated as less than `"8000"`).
- The `evaluate` method now normalizes `Quantity` objects to their raw
`.Value` first.
- If both operands are numeric, a direct numerical comparison is
performed for all operators (`=`, `!=`, `>`, `<`, `>=`, `<=`).
- String-based comparisons (for `like`, or as a fallback for other
types) are handled in a separate path.
- Add unit test to lock down this behavior.
* BIM: autocompleter improvements
- Add a trailing space when autocompleting the keywords that need it
- Autocomplete multi-word keywords as a unit
* BIM: Improve live validation user experience
- Do not throw syntax error when in the middle of autocompleting a
keyword
- Introduce typing state and show incomplete query status
* BIM: change double-click action to edit in statements overview
Also allow quick editing the statement description with F2
* BIM: Improve user strings for consistency
* BIM: show count of returned objects in statements table
* BIM: disable preview on invalid queries
* BIM: define slot so that tests can run headless
* BIM: Update unit test to adapt to new behaviour
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Add detailed test coverage for the new Command constructor that accepts
annotations parameter, including positional and keyword arguments,
string and numeric annotations, empty annotations, and backward
compatibility. Split existing save/restore tests for better granularity.
src/Mod/CAM/CAMTests/TestPathCommandAnnotations.py:
- Added test14-test20 for Command constructor with annotations
- Split test12 into test12 (empty annotations) and test13
(complex annotations) for focused testing
- Completely re-implemented the TSP algorithm in C++ for better path quality
- Added detailed comments and documentation to clarify each step
- Improved nearest neighbor, 2-opt, and relocation logic
- Enhanced handling of start/end point constraints
- Updated PathUtils.py docstring to accurately describe start point behavior
- Enhanced the C++ TSP solver to accept optional start and end points, so the route can begin and/or end at the closest point to specified coordinates.
- Updated the Python pybind11 wrapper and PathUtils.sort_locations_tsp to support startPoint and endPoint as named parameters.
- Added a new Python test suite (TestTSPSolver.py) to verify correct handling of start/end points and integration with PathUtils.
- Registered the new test in TestCAMApp.py and CMakeLists.txt for automatic test discovery.
This update introduces a new C++ 2-Opt TSP solver with Python bindings.
src/Mod/CAM/App/CMakeLists.txt:
- Add build and install rules for the new pybind11-based tsp_solver Python module
src/Mod/CAM/App/tsp_solver.cpp:
- Add new C++ implementation of a 2-Opt TSP solver with nearest-neighbor initialization
src/Mod/CAM/App/tsp_solver.h:
- Add TSPPoint struct and TSPSolver class with 2-Opt solve method
src/Mod/CAM/App/tsp_solver_pybind.cpp:
- Add pybind11 wrapper exposing the TSP solver to Python as tsp_solver.solve
src/Mod/CAM/PathScripts/PathUtils.py:
- Add sort_locations_tsp Python wrapper for the C++ TSP solver
- Use tsp_solver.solve for TSP-based
Fixes#25924.
The `snapToIntersection` function has been changed to an if-elif structure, and in case of face-edge intersection snap, only two sublements are checked. This ensures a reasonable speed, even if objects have many edges and faces. The change does mean you have to move the mouse over the correct face for it to be processed. But since face-edge intersection snap is new anyway, this is acceptable I think.
Used TestMach3Mach4LegacyPost.py as the starting point with input from
other test scripts too.
The test demonstrate a fix for #25723, as the Fanuc post processor has not been not updated to the latest changes in
the Path module.
This is related to #24676, where thread tapping was added to LinuxCNC, but the equivalent code in the Fanuc
postprocessor were not adjusted to cope, and #25677 where it was observed that the Fanuc postprocessor was
broken in versions 0.20 and 1.0.
Added MockTool class to provide ShapeName to fanuc_post.py, used to
convert drill cycles to threading cycles if the tool has shapeid "tap".
Several methods introduced 2025-05-04 are preferred, but the methods
available in version 1.0 are used as fallback sources
for active status and coolent enabled.
The difference according to the documentation is that M30 will rewind
the paper tape while M2 will not. The effect on a CNC from 1994 is
that M30 turn on the indicator light marking that the program has
completed, while M2 do not, and the light is wanted to know when
the machine is done.
Fanuc only understand upper case letters, no point in providing the
file name in lower case letters. The provided file name is always "-",
so drop it completely. The first comment is presented in the Fanuc user
interface, it should get more relevant content.
Dropped useless semicolon.
Enabling tool height compensation will cause the axis to move up
the length of the tool, which will cause a Z overtravel error when
the tool change take place close to the top of the machine. To
counter this move, ask the machine to move its commanded position
down the length of the tool height, which in effect causes no upward
movement after the tool change. The #4120 variable contain the
current tool number, and #2000 - #20XX contain the tool heights.
The Fanuc post processor add a trailing space to all M6 lines.
This make it problematic to write test for the generated output,
when compiled with the automatic policy enforcer on github
removing trialing space from the expected output. Avoid the
problem by removing the trailing space from the generated
output.
The thread tapping implementation in the Fanuc post processor change
behaviour of M3, G81 and G82 when the tool ShapeID matches "tap".
but the code not not expect that the parse() method will be
called with two different classes as arguments.
Rewrite the M3 handling to handle ToolController arguments
instead of crashing with an AttributeError.
Issues:
Fixes#14016Fixes#25723
Some Fanuc machines do not understand the 'M6 T0' instructions in the
preamble. Move it out of the preamble and controlled by a new
command line argument --no-end-spindle-empty for the machines
where running to cause a "Tool Number Alarm" and the program to crash
as tool zero is not a valid tool.
Fixes: #25677
Reset line number when using --line-numbers. Restore all default values when a
command line argument is not used.
This helps the test scripts ensure that the arguments passed during one test is the only
one taking effect during this test.
This PR allows each Toolbit to have its own Units property (Metric/Imperial) for better unit management.
Short Summary:
- Adds a Units property (Metric/Imperial) to ToolBit objects for better unit management.
- Ensures ToolBit schema is set and restored properly in the editor and utilities.
- Updates formatting and property handling to respect the selected units.
- Improves the ToolBit editor widget to refresh and sync schema/UI when units change.
- Uses FreeCAD.Units.setSchema and getSchema to switch between unit schemas.
NOTE: Toolbit dimensions are read from JSON in their native units (as specified by the Units property),
converted to metric for all internal calculations, and displayed in the UI using the
toolbit's selected units. This ensures both accurate internal computation and user-friendly
display, while storing the correct units in the JSON. This can cause some rounding differences
when switching units. Example: 2 mm becomes 0.0787 inches. If you save that as imperial and then
switch back to metric, it will show 1.9999 mm
src/Mod/CAM/Path/Tool/toolbit/models/base.py:
- Add Units property to ToolBit and ensure it's set and synced with toolbit shape parameters.
- Update property creation and value formatting to use Units.
src/Mod/CAM/Path/Tool/toolbit/ui/editor.py:
- Use setToolBitSchema to set schema based on toolbit units.
- Add logic to refresh property editor widget when units change.
- Restore original schema on close.
- Improve docstrings and signal handling.
src/Mod/CAM/Path/Tool/toolbit/util.py:
- Add setToolBitSchema function for robust schema switching.
- Update format_value to use schema and units.
- Add docstrings and clarify formatting logic.
src/Mod/CAM/Path/Tool/library/ui/browser.py:
- Restore original schema after editing toolbit.
src/Mod/CAM/Path/Tool/library/ui/editor.py:
- Ensure correct schema is set for new toolbits.
* Part: TopoShape make getElementTypeAndIndex more robust
* Part: Add unit tests for new regex
---------
Co-authored-by: Chris Hennes <chennes@gmail.com>
* Gui: Fix preferences search popup positioning on Wayland
Change window flag from `Qt::Tool` to `Qt::ToolTip` to fix popup
appearing in center of screen and closing immediately on Wayland desktop
envs. `Qt::ToolTip` uses xdg_popup protocol which provides proper
positioning relative to parent window on Wayland.
* Gui: Initialize ok check flag to default in prefs
* Gui: Use ranged for loop where possible in prefs
* Gui: Remove unused variable in preferences controller
* Gui: Use const where possible in prefs search dialog
* Gui: Use static where possible in search prefs dialog
* Gui: Use braced initializer list when returning type in search prefs
* Gui: Use auto in search prefs dialog where possible
* Gui: Do not use else if after return in search prefs dialog
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Match the string that gets sent to setCommand
Correction of a not matching string that gets sent to setCommand. This results in a not translated menu item in the Sketcher menu.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit makes boundary stock objects distinct from regular stock by
setting a boundary flag and label, and prevents editing of boundary
objects in the job view provider.
src/Mod/CAM/Path/Dressup/Boundary.py:
- Add promoteStockToBoundary method, set boundary properties, handle
missing stock/shape
src/Mod/CAM/Path/Main/Gui/Job.py:
- Block editing for objects flagged as boundary stock
This commit introduces support for passing an 'annotations' dictionary
to the Command class constructor and its Python bindings. The annotations
dictionary can contain string or numeric values for each key, allowing for
richer metadata on CAM commands. Each annotation must be provided as a
key-value pair within a dictionary.
Examples:
cmd = Command("G1", {"X": 10.0, "Y": 5.0}, {"note": "Rapid move"})
cmd = Command("G2", {"X": 20.0, "Y": 15.0}, {"priority": 1})
cmd = Command("G3", {"X": 30.0, "Y": 25.0}, {"note": "Arc move", "speed": 1500})
cmd = Command("G0", {"X": 0.0, "Y": 0.0}, {})
cmd = Command("G0", {"X": 0.0, "Y": 0.0})
cmd = Command("G1", {"X": 10.0, "Y": 5.0}, annotations={"note": "Rapid move"})
src/Mod/CAM/App/Command.cpp:
- Added new Command constructor accepting annotations
src/Mod/CAM/App/Command.h:
- Declared new Command constructor with annotations parameter
src/Mod/CAM/App/Command.pyi:
- Updated docstring to describe annotations argument
src/Mod/CAM/App/CommandPyImp.cpp:
- Extended Python init to parse and set annotations dictionary
This PR fixes a bug introduced with PR #25800
It updates the ObjectOp class to correctly assign the job, model,
and stock properties from the parentJob object. Previously, only
PathUtils.addToJob was called, which did not set these attributes directly.
This change ensures that when creating a new operation, the job, model,
and stock fields are properly initialized from the parent job. This
fixes issues where operations did not inherit the correct job context,
leading to missing or incorrect references to the model and stock, and
potential errors in downstream processing.
- Added new migration system to handle legacy parameter conversion for ToolBit assets and objects.
- Implemented ParameterAccessor abstraction for consistent access to dicts and FreeCAD objects.
- Centralized migration logic for CornerRadius from TorusRadius or FlatRadius/Diameter, restricted to Bullnose shape-type.
- Updated asset and object initialization to use migration logic and filter Bullnose parameters.
src/Mod/CAM/Path/Tool/camassets.py:
- Integrate new migration logic for ToolBit assets using ParameterAccessor and migrate_parameters.
- Ensure shape-type is set before parameter migration; only write changes if migration occurred.
src/Mod/CAM/Path/Tool/toolbit/migration.py:
- Add ParameterAccessor class and migrate_parameters function for unified migration.
- Handle legacy parameter conversion for Bullnose tools.
src/Mod/CAM/Path/Tool/toolbit/models/base.py:
- Apply migration logic to ToolBit objects on restore.
- Filter Bullnose parameters to remove FlatRadius after migration.
src/Mod/CAM/Path/Tool/shape/models/bullnose.py:
- Add filter_parameters method to remove FlatRadius for Bullnose.
src/Mod/CAM/Path/Op/SurfaceSupport.py:
- Derive FlatRadius from CornerRadius and Diameter if both are present.
src/Mod/CAM/Path/Op/Surface.py:
- Remove obsolete setOclCutter method.
* Add MoveWithHost check for material assignment
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* BIM: fix Arch_Reference update on doc restored
Fixes: #24943
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
`SketcherTransformationExpressionHelper` was using `geoId` as the map
key for storing expressions during geometry transformations. This is
incorrect since single geometry may have multiple constraints with
different expressions.
So, using `geoId` as the key causes `std::map` overwrites, resulting in
only the last expression being preserved per geometry.
So this patch changes key to be a constraint index, and adds
`ConstraintExpressionInfo` struct to store both expression and `geoId`.
Co-authored-by: PaddleStroke <pierrelouis.boyer@gmail.com>
* [TechDraw] fix dialog to be ontop, clear the keyLst and save mouse movement between messagebox and dialog
* [TechDraw] change from setWindowFlags to setWindowFlag
* [TechDraw] handle BIM and Draft as first available views...
.... also increase the possible scale values such as 5 : 2 or 7 : 3
* [TechDraw] fix 1:1 scale regression
* [TechDraw] remove unnecessary line of code
* TD: Simplify code
Address comments from review.
Co-authored-by: WandererFan <wandererfan@gmail.com>
* [TechDraw] enable use of Fraction for scale...
... conversion
---------
Co-authored-by: Chris Hennes <chennes@pioneerlibrarysystem.org>
Co-authored-by: WandererFan <wandererfan@gmail.com>
This PR adds rigid tapping back to the legacy LinuxCNC postprocessor, which
was removed in commit cd1c29e3f23fde0a90980ffcf3e70406769f2d43 in PR 24771.
The changes include:
- Modifying the linuxcnc_legacy_post.py script to include logic for Rigid tapping
- Updating comments and documentation to reflect the addition of Rigid tapping
* part design: fix issue #25639 use c++ exception to prevent crash
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This method can be used to bake in transform directly into geometry.
Normally the setTransform or setPlacement methods only store additional
positional data within the shape without changing the actual geometry.
So if we have some geometry placed at 0,0,0 and we want to have few
copies of that shape the goemetry stays intact but additional transform
is stored. This method can be used to alter the gometry and reset the
transform to identity. It helps with model stability, as not every
method correctly handles that additional transform metadata.
This commit removes a ton of dead code from FeatureBoolean. It might
been ported here from the Link branch but it is not used and it is
confusing. The reason for having that code here is also not really
obvious so there is no reason to keep it.
Thanks for taking the time to fill out this problem report! Please [search](https://github.com/FreeCAD/FreeCAD/issues) if a similar issue already exists and check out [how to report issues](https://github.com/FreeCAD/FreeCAD?tab=readme-ov-file#reporting-issues). By submitting this issue, you agree to follow our [Code of Conduct](https://github.com/FreeCAD/FreeCAD/blob/main/CODE_OF_CONDUCT.md).
- type:textarea
id:description
attributes:
label:Problem description
description:Describe the problem and how it impacts user experience, workflow, maintainability or performance. You can attach images or log files by clicking this area to highlight it and then dragging files in. To attach a FCStd file, ZIP it first.
placeholder:Describe your problem briefly.
validations:
required:true
- type:dropdown
id:wb
attributes:
@@ -28,15 +37,6 @@ body:
- TechDraw
- Other (specify in description)
- type:textarea
id:description
attributes:
label:Problem description
description:Describe the problem and how it impacts user experience, workflow, maintainability or performance. You can attach images or log files by clicking this area to highlight it and then dragging files in. To attach a FCStd file, ZIP it first.
description:Report problems related to code structure, maintainability, performance, correctness, or technical debt that do not directly affect end-user behavior.
# your codebase is analyzed, see https://docs.github.com/en/code-security/code-scanning/creating-an-advanced-setup-for-code-scanning/codeql-code-scanning-for-compiled-languages
steps:
- name:Harden the runner (Audit all outbound calls)
# your codebase is analyzed, see https://docs.github.com/en/code-security/code-scanning/creating-an-advanced-setup-for-code-scanning/codeql-code-scanning-for-compiled-languages
steps:
- name:Harden the runner (Audit all outbound calls)
> Bleeding edge FreeCAD development builds for testing bugfixes, regressions, and recently implemented features. Do not use in a production environment.
\pard\nowidctlpar\hyphpar0\sl276\slmult1\fs18 The different libraries used in FreeCAD and their respective licenses are described on the\line\fs24{\field{\*\fldinst{HYPERLINK "https://www.freecadweb.org/wiki/Third_Party_Libraries" }}{\fldrslt{\cf2\ul\fs18 Third Party Libraries wiki page}}}\cf0\ulnone\f0\fs18 .\fs24\par
\pard\nowidctlpar\hyphpar0\sl276\slmult1\fs18 The different libraries used in FreeCAD and their respective licenses are described on the\line\fs24{\field{\*\fldinst{HYPERLINK "https://www.freecad.org/wiki/Third_Party_Libraries" }}{\fldrslt{\cf2\ul\fs18 Third Party Libraries wiki page}}}\cf0\ulnone\f0\fs18 .\fs24\par
\par
\pard\noline\nowidctlpar\hyphpar0\sa283\fs12\par
\pard\nowidctlpar\hyphpar0\sl276\slmult1\b\fs18 GNU LIBRARY GENERAL PUBLIC LICENSE\b0\fs24\par
@@ -81,4 +81,5 @@
\pard\nowidctlpar\hyphpar0\sb240\sa180\sl276\slmult1\b\fs24 3D Mouse Support\b0\par
\pard\nowidctlpar\hyphpar0\sb180\sa180\sl276\slmult1\fs18 Development tools and related technology provided under license from 3Dconnexion.(c) 1992 - 2012 3Dconnexion. All rights reserved\fs12\par
* Use `fastsignals::bind_weak` instead of `std::bind` to ensure that nothing happens if method called when binded object already destroyed
* Pass pointer to T class method as first argument, `shared_ptr<T>` or `weak_ptr<T>` as second argument
* Example: `bind_weak(&Document::save(), document, std::placeholders::_1)`, where `document` is a `weak_ptr<Document>` or `shared_ptr<Document>`
## Weak this idiom
The `fastsignals::bind_weak(...)` function implements "weak this" idiom. This idiom helps to avoid dangling pointers and memory access wiolations in asynchronous and/or multithreaded programs.
In the following example, we use weak this idiom to avoid using dangling pointer wehn calling `print()` method of the `Enityt`:
```cpp
structEntity:std::enable_shared_from_this<Entity>
{
intvalue=42;
voidprint()
{
std::cout<<"print called, num = "<<value<<std::endl;
}
std::function<void()>print_later()
{
// ! weak this idiom here !
autoweak_this=weak_from_this();
return[weak_this]{
if(autoshared_this=weak_this.lock())
{
shared_this->print();
}
};
}
};
intmain()
{
autoentity=std::make_shared<Entity>();
autoprint=entity->print_later();
// Prints OK.
print();
// Prints nothing - last shared_ptr to the Entity destroyed, so `weak_this.lock()` will return nullptr.
entity=nullptr;
print();
}
```
## Using bind_weak to avoid signal receiver lifetime issues
In the following example, `Entity::print()` method connected to the signal. Signal emmited once before and once after the `Entity` instance destroyed. However, no memory access violation happens: once `Entity` destoryed, no slot will be called because `bind_weak` doesn't call binded method if it cannot lock `std::weak_ptr` to binded object. The second `event()` expression just does nothing.
```cpp
#include<fastsignals/signal.h>
#include<fastsignals/bind_weak.h>
#include<iostream>
usingVoidSignal=fastsignals::signal<void()>;
usingVoidSlot=VoidSignal::slot_type;
structEntity:std::enable_shared_from_this<Entity>
{
intvalue=42;
VoidSlotget_print_slot()
{
// Here fastsignals::bind_weak() used instead of std::bind.
This guide helps to migrate large codebase from Boost.Signals2 to `FastSignals` signals/slots library. It helps to solve known migration issues in the right way.
During migrations, you will probably face with following things:
* You code uses `boost::signals2::` namespace and `<boost/signals2.hpp>` header directly
* You code uses third-party headers included implicitly by the `<boost/signals2.hpp>` header
## Reasons migrate from Boost.Signals2 to FastSignals
FastSignals API mostly compatible with Boost.Signals2 - there are differences, and all differences has their reasons explained below.
Comparing to Boost.Signals2, FastSignals has following pros:
* FastSignals is not header-only - so binary code will be more compact
* FastSignals implemented using C++17 with variadic templates, `constexpr if` and other modern metaprogramming techniques - so it compiles faster and, again, binary code will be more compact
* FastSignals probably will faster than Boost.Signals2 for your codebase because with FastSignals you don't pay for things that you don't use, with one exception: you always pay for the multithreading support
## Step 1: Create header with aliases
## Step 2: Rebuild and fix compile errors
### 2.1 Add missing includes
Boost.Signals2 is header-only library. It includes a lot of STL/Boost stuff while FastSignals does not:
```cpp
#include<boost/signals2.hpp>
// Also includes std::map, boost::variant, boost::optional, etc.
// Compiled OK even without `#include <map>`!
std::mapCreateMyMap();
```
With FastSignals, you must include headers like `<map>` manually. The following table shows which files should be included explicitly if you see compile erros after migration.
If you just want to compile you code, you can add following includes in you `signals.h` header:
```cpp
// WARNING: [libfastsignals] we do not recommend to include following extra headers.
#include<map>
#include<boost/variant/variant.hpp>
#include<boost/optional/optional.hpp>
#include<boost/scoped_ptr.hpp>
#include<boost/noncopyable.hpp>
#include<boost/bind.hpp>
#include<boost/function.hpp>
```
### 2.2 Remove redundant returns for void signals
With Boost.Signals2, following code compiled without any warning:
```cpp
boost::signals2::signal<void()>event;
event.connect([]{
returntrue;
});
```
With FastSignals, slot cannot return non-void value when for `signal<void(...)>`. You must fix your code: just remove returns from your slots or add lambdas to wrap slot and ignore it result.
### 2.3 Replace track() and track_foreign() with bind_weak_ptr()
Boost.Signals2 [can track connected objects lifetype](https://www.boost.org/doc/libs/1_55_0/doc/html/signals2/tutorial.html#idp204830936) using `track(...)` and `track_foreign(...)` methods. In the following example `Entity` created with `make_shared`, and `Entity::get_print_slot()` creates slot function which tracks weak pointer to Entity:
TEST_CASE("Can connect a few slots and emit","[signal]")
{
signal<void(int)>valueChanged;
intvalue1=0;
intvalue2=0;
valueChanged.connect([&value1](intvalue){
value1=value;
});
valueChanged.connect([&value2](intvalue){
value2=value;
});
REQUIRE(value1==0);
REQUIRE(value2==0);
valueChanged(10);
REQUIRE(value1==10);
REQUIRE(value2==10);
}
TEST_CASE("Can safely pass rvalues","[signal]")
{
conststd::stringexpected="If the type T is a reference type, provides the member typedef type which is the type referred to by T. Otherwise type is T.";
std::stringpassedValue=expected;
signal<void(std::string)>valueChanged;
std::stringvalue1;
std::stringvalue2;
valueChanged.connect([&value1](std::stringvalue){
value1=value;
});
valueChanged.connect([&value2](std::stringvalue){
value2=value;
});
valueChanged(std::move(passedValue));
REQUIRE(value1==expected);
REQUIRE(value2==expected);
}
TEST_CASE("Can pass mutable ref","[signal]")
{
conststd::stringexpected="If the type T is a reference type, provides the member typedef type which is the type referred to by T. Otherwise type is T.";
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
PaddleStroke