2c052b4b5fa1856dfcdf0d0673effecb4596b0df
6
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
b841eb939f |
Merge main into drawings (Nov 4)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
fd5d0e6738 |
Assemblies: mass, centre of mass and inertia per occurrence and rigid group (#9715) (#9717)
* claim: #9715 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Assemblies: mass, centre of mass and inertia per occurrence and rigid group (#9715) assembly.mass_properties (and Session::mass_properties) gives, for every occurrence and every rigidly-joined group, mass (kg), centre of mass (mm) and the inertia tensor about it in world and frame axes (kg mm2), the input the multibody dynamics work needs. Bodies are measured as Properties measures them (kernel volume and centre, mesh gyration), once per body, then moved by each placement and combined by the parallel-axis theorem. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
d01de87552 |
Merge main into drawings (Oct 30)
Brings the 1,270 commits on main since the drawings branch last met it,
including the first drawings landing (#8249), into drawings.
Merged with the drawings snapshot that #8249 landed (
|
||
|
|
413c145c56 |
Bill of materials: stable item numbers, saved with the design (#3874) (#7779)
* Bill of materials types on one occurrence (#3874) An occurrence can count in the bill of materials as another type than its component, for that occurrence only: `bom_type` on the occurrence, set with occurrence.bom_type (one undo step; it changes no geometry, so nothing is evaluated). An occurrence without it counts as its component, so older designs need no step; the format change is one pending file (occurrence-bom-type.rs). assembly.bom counts each placement by its own type: a purchased placement of a kit part is one row without the kit's contents, a phantom placement lists what it holds where it sits, and one component placed as two types gives a row for each. Paste New keeps the types; Paste Instance counts as the component. The compact list and the locales (en, de, ja, ko, zh) carry it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials rows: material, and mass when asked (#3874) Every row of assembly.bom (and every part folded into a row) names the physical materials of its definitions' own bodies, each once, with "; " between several. An unassigned body's is Default. A subassembly or an empty component has none of its own. With mass: true each row also carries mass_g, what one of its parts weighs as the row stands for it, so the rows of any list add up to the assembly once: - at the top level, all it holds; - in parts only, its own bodies, because what it holds is listed beside it; but a folded part's row is the kit (all it holds), a purchased part is all it holds, and an inseparable one all but its purchased parts; - a reference placement adds nothing, nor does anything inside it. It is null when nothing in the part has a body, a body can't be measured or measures a negative volume (inside out, such as an open mesh facing the origin), or the definitions sharing the row's part number weigh differently. Measuring runs the kernel, so it is asked for: the Bill of Materials window re-runs assembly.bom after every change and doesn't ask (0.2 ms without mass, 135 ms with it on a bolted bracket). Placements are rigid, so each definition is weighed once however often it is placed, and each body is measured once with measure_mass (the same volume as measure, without the topology counts' tessellation). A body's material and its density now follow one rule: Document::body_material (new, beside Document::density) is the material its mass is worked out with, named as the library or the design spells it: the assigned one when it is known and has a usable density, else DEFAULT_MATERIAL (new, beside DEFAULT_DENSITY). Document::density is that material's density, with the same values as before. So a damaged file's unknown or broken material reads as Default, as it weighs, and a hand-edited "steel" reads as Steel. The bill of materials uses both; the other places that read a body's material or spell "Default" (component.properties, inspect and others) move to them in a separate cleanup PR for #4783, which brings the same two functions to main. Document::body_owners returns each body's owner in order, so the rows find their bodies in one pass. It no longer repeats the owner rule: model_body_component, legacy_body_component and body_owners go through the rule written once, given how to find a feature's component (the timeline for one body; body_owners still builds its map once). The CSV columns are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Physical properties: a reference placement adds nothing (#3874) component.properties and occurrence.properties now leave out a placement that counts as a reference in the bill of materials (its occurrence's bom_type, else its component's: a fixture shown for context), and everything inside it: no body entry, mass, volume, area, centre of mass or bounding box, and it isn't in placement_count. That is the bill of materials' rule, so a reference part is in neither the parts list nor the assembly's mass, and every bill of materials list still adds up to the assembly's Properties. The placement asked for counts whatever its type: a reference fixture's own Properties still measure it and what it holds, as a component's own bill of materials lists what it holds. Only the totals change. The placements walk that face picking, sketches on placed faces and the occurrence path lookup use stays complete, so a reference fixture's faces can still be picked. The type an occurrence counts as is now written once: Document::occurrence_bom_types (new, beside BomType) gives each occurrence's type in their order, looking each component up once. The bill of materials, occurrence.bom_type's reply and the physical properties use it, in place of the two copies in the bill of materials. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: one module of typed rows in doc (#3874) The bill of materials now lives in solidcraft_doc::bom as the typed rows of a bill of materials view. assembly.bom's JSON and CSV and the Bill of Materials window present those rows; parts lists, item numbers and balloons are to read them too. - list(tree, request) walks the placements below the assembly once. One rule (classify) decides for each placement whether it is counted in a row, how what it holds is walked and what its row stands for when weighed, from the type it counts as, the view and the component's shape. "An inseparable part's purchased parts are listed beside it" is written there once, and the mass walk reads it. - A row carries its identity (RowKey: the rows it is listed under and its lowest member, a component and the type it counts as), its item, count, each, the placements it counts by occurrence path, what it stands for (Holds), its part properties, materials, mass and children. - Mass is a Mass: not asked, grams, or unknown with the reason (nothing to weigh, unmeasurable, the definitions differ). Each definition is weighed once for each kind of row and kept in the Tree. The assembly's total is weighed from the design the same way, not added up from rows, so it is known even when a row's definitions weigh differently. - Bom::row_of(tree, path) finds the row a placement is counted in (a nut inside a purchased bolt is in the bolt's), or says why there is none: no such placement, a reference, or no row of its own. - rows_deep() lists each row followed by its children (1, 1.1, 1.2, 2); Request::default() is assembly.bom {}; View and Group carry the commands' words. What changes, beside that the output is otherwise the same: - One placement budget (100,000) and one depth limit (64 levels) cover the whole list. A folded part's contents used to be walked under a budget of their own, so a design past the limit was listed with group_children; it is refused now, as it already was without it. - Kits are compared by what they hold (part, type and count), not as text: a part without a part number whose id is 12 no longer reads as a part numbered "#12", nor a part numbered "P×1\nQ" as a P and a Q. Kits holding the same parts through different definitions (two bolt definitions, each with its own nut definition of one part number) are one row, and its children count every bolt's placements, where they showed the first bolt's definitions only. - A folded part's row always stands for all it holds. It used to stand for its own bodies when nothing inside it was listed, which differs only for a phantom with bodies of its own inside it. - An inseparable part's row in parts only names the material of what it stands for (its steel blocks read Steel) where it was blank, and a subassembly whose parts share one material names it at the top level. - A damaged file's component placed inside itself is refused at the first level too, since naming a row's material walks inside it, as weighing it already did. - A row's body count saturates, so a compact assembly with more placed bodies than a count holds (a block placed 2^64 times) is weighed instead of overflowing. Materials follow Cameron's rule for a row that stands for an assembly (C3, 2026-10-23): a part's row lists its own bodies' materials, as before; a row that weighs what is placed inside it (a subassembly at the top level, a purchased or inseparable part with parts inside it, a folded part) names a material only when everything it stands for shares it, else none. A frame of steel and aluminium parts reads blank, and a bought bearing of steel parts reads steel. One walk (adding) decides what is placed inside a row's part for both its mass and its materials, and each definition's materials are read once. Components are looked up by id once per list; they were searched for each row and each kit. The window reads the rows directly instead of running assembly.bom after every change, so it no longer adds an entry to the crash journal each time it refreshes. The engine keeps the command, its parameters and the files. Tests check rows. The module's own: an empty assembly, one part, a row named by its lowest member through part-number edits, one part number of two types, a phantom in a phantom, a reference at the root and its own list, the 64-level limit, the budget hit exactly with and without folding, kits told apart by what they hold and the types they hold it as, a folded part's contents not folded again, row_of for every type, every placement in one row or none, a damaged file's missing component and cycle, the mass agreement rule (moved from the engine with row_mass) and what adds to a row. In the engine: the material rule for every kind of row, the JSON against the rows for every option, why a mass is unknown, the total against component.properties in every view, what a row stands for and its placements, and the overflow. The assembly.bom tests from #1888 on pass unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: structured, all levels (#3874) P4: `assembly.bom {levels: "all"}` lists the assembly's direct contents, each subassembly a row with its own rows listed under it, to any depth (items 1, 1.1, 1.1.1), each child counted `each` per one of its parent and `count` in all. - `doc::bom::View::AllLevels` goes through the one rule (`classify`) the other views use: a normal subassembly is a row holding what it holds, listed under it as all levels again; a phantom has no row and what it holds joins the level it sits in, merged with the same parts there; a purchased part is a row with nothing under it unless `show_purchased_contents` (Request::purchased_contents) asks for its contents; an inseparable part is a row with nothing under it; a reference counts nowhere. - The rows under a row are the folded-part rows already built (kits), so subassemblies with one part number and different contents are different rows, and no second nested builder is written. All levels gathers every level as it gathers the first; a folded part's contents are still gathered by part number in the other views. - The JSON and CSV already nest to any depth; `show_purchased_contents` is echoed with the other options. Each row's mass still adds up to the assembly at the top. Tests, worked by hand: four bolts and a two-bolt subassembly placed twice (8 bolts in parts only; in all levels the subassembly twice with 2 bolts each, 4 in all, beside the top's 4); three levels deep (2 x 3 x 2: counts 2, 6, 12, items 1, 1.1, 1.1.1, the CSV in that order); each type (a phantom's bolt merged into its subassembly's, purchased with and without its contents, inseparable, reference); two subassemblies of one part number with different contents as two rows; `levels: "deep"`, a flag that isn't one, and an assembly over 100,000 placements refused. D's test that a folded part's contents are listed as parts only lists them now names the two views it holds for, and checks all levels' nested rows (the subassembly inside the folded part a row with its own rows). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: parts with one part number are one row only when they hold the same to any depth (#3874) The kit compare looked one level down: two parts with one part number whose contents differed two levels below merged into one row, and the children that only one of them held came out at each 0. Each placement now has a kit, a number interned from what it holds (each child's part, type and own kit), built in one pass from the last entry back: an entry's children come after it in the walk, so their kits are known first. Rows are told apart by that number. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: a phantom's own bodies are a row of their own (#3874) A phantom had no row, and its own bodies were weighed in the assembly's total but in no row, so the rows didn't add up. Its own bodies, when it has any, are now a row of the phantom's type that stands for them alone, as a subassembly's own bodies are in parts only; what it holds still counts where it is placed. Inside an inseparable part in parts only, nothing changes: the part's row weighs them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: the assembly's total mass in assembly.bom (#3874) `list` weighed the assembly as a whole (from the design, not added up from rows) and the command dropped it. `assembly.bom` now gives it as `total_mass_g`, and so does a JSON file it writes, only when `mass: true` (null when it isn't known). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: all levels as JSON lists its rows one after another (#3874) All levels nested each level's rows in the row above it: two JSON containers a level, so the 64 levels a list may hold came to 129, past the 128 that serde_json (and so the MCP server and the control channel) reads. In all levels `assembly.bom` now gives the rows one after another, each followed by the rows under it (items 1, 1.1, 1.1.1), as its CSV does; a row listed under another still reads its item as text and gives `each`. A folded part's contents, one level, stay inside its row as `children`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: a row whose definitions weigh something and nothing differs (#3874) A row gathering a definition with bodies and one with none (one part number on a steel block and on an empty assembly) said there was nothing to weigh. Their masses differ, so it says so; only a row where none of them has a body has nothing to weigh. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: definitions agree on a mass within the measurement accuracy (#3874) A row's definitions had to weigh the same to 1e-9, but the kernel measures a volume from tessellations, and one part modelled another way is tessellated differently: a pin made as a cylinder and as a revolved rectangle measured 1.05e-5 apart, so it read as two masses. The bill of materials now has its own `bom::MEASURE_ACCURACY` (1e-4: ten times that gap, with its doc on what it follows: the kernel's `measure` gives one solid's volume to about that, relative, however the solid was modelled), and a row's definitions agree within it. A pin 0.1% wider (0.2% heavier) still differs. The kernel is unchanged: its work is paused while it is being replaced. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials guide: all levels, purchased contents and this round's changes (#3874) The guide had no `levels: "all"` or `show_purchased_contents`. It now has both, with how all levels tells subassemblies apart and lists its rows as JSON, a phantom's own bodies as a row, the assembly's `total_mass_g`, and definitions that agree on a mass within the measurement. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: materials are bodies', as the code says (#3874) The rule stays as built (a row that stands for an assembly names a material only when everything in it shares one), but its comment and the guide said a material set on a component is set on all its bodies. `material.assign` takes bodies only: a component has no material of its own yet. The module notes now say so, and where component materials will join the rule. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: a body's volume measured once with the evaluated model (#3874) `assembly.bom` measured each definition's bodies anew for every list, and `component.properties` measured each placed body again. A body's measure depends on nothing but the body, so the evaluated model keeps its volume (`ModelBody::volume`, shared by clones, as its outline box is), and both commands weigh that one measurement: the same grams to the bit. A body offset scales it by the size of its determinant (`mat_volume_scale`) instead of copying the B-rep to measure it. The physical properties still measure the placed body for its area, centre, box and counts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: row keys, what rows stand for, members and row_of stay in doc (#3874) Nothing outside `doc` reads them yet: `RowKey`, `Unlisted`, `Holds`, a row's `key`, `stands_for` and `members`, and `Bom::row_of` are crate-private until item numbers and balloons need them, so they can still change without breaking another crate. The engine test that read them is covered by doc's own (`row_of_finds_the_row_a_placement_is_counted_in`, `every_placement_is_in_one_row_or_none`). `Unknown` stays public: `Mass::Unknown` carries it, and a public enum can't carry a private type. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Assemblies: one fallible placement walk for totals that cover the whole assembly (#3874) The physical properties walked the placements twice in their own code, with the 64 levels and 100,000 placements written as literals, and the bill of materials gave the same refusal in words of its own. `Document::placements_below` is that walk once, from any placement, refusing what is past `MAX_PLACEMENT_DEPTH` or `MAX_PLACEMENTS` (never walking it in part), and both use its refusal: "the assembly is deeper than 64 levels or has more than 100000 placements". The physical properties now allow 100,000 placements below the start, as the bill of materials does (they stopped at 99,999). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: the same rows for less work (#3874) - Part numbers are borrowed from their definitions (`Part<'a>::Number(&'a str)`) instead of copied for every placement grouped and every kit. - A row's members, the placements' paths, are built only when the list is asked for them (`list_with`, for item numbers, balloons and `row_of`): the walk then keeps paths and moves each into its one row instead of copying it. A list made for its rows alone builds no paths at all. - `row_of` reads an index of member paths, built once with the list, instead of scanning every row's members for each placement asked about. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * occurrence.bom_type: input and output schemas (#3874, #4035) `OccurrenceBomTypeParams` takes the occurrence (an id or a name), its path when given and the type (one of the five, `same_as_component` or null), each read by the handler's own reader: the path through `cmd::path_ids`, split out of `occurrence_path`, the type through `cmd::occurrence_bom_type`, split out of the handler. The handler builds its reply from `OccurrenceBomTypeResult` ({occurrence, bom_type, effective, changed}). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials tests: all levels presented at every depth, and its caps (#3874) - `assembly_bom_presents_the_typed_rows` now covers all levels (with mass, purchased contents and by definition) and the total mass, on a three-level design with a purchased gearbox holding a gear, comparing every field of every row down to 1.1.1.1. - All levels lists 100,000 placements and 64 levels, and refuses 100,001 and 65. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: static materials, typed in for a component (#3874) A value the bill of materials works out can now be typed in by hand for a component (a static value, Cameron's choice of word). Material comes first: `component.metadata.set {component, material}` takes any text up to MAX_TEXT_BYTES, and `component.metadata` reads it back as `material`. It is shown wherever the component is listed, at every level and in every view, over the materials worked out from its own bodies or an assembly's blank, and the row carries `static: ["material"]`. Clearing it (empty text or null) brings the worked-out material back. It is kept the way `bom_type` is: a part property in the component's metadata (`material`), so there is no format change and older releases keep it. `PART_ONLY` names both keys in one place: Design Properties doesn't list them and `document.metadata.set` refuses them. Another tool's non-text value is kept and reads as no static material. The bill of materials says one thing about the part everywhere: a static material also stands for the part in the one material an assembly around it shares (two parts typed in as S355 make their frame read S355; S355 beside plain steel reads blank). A phantom's row is its own bodies, so its static material names those only, and what it holds keeps its own. A row gathering several definitions takes the first one's that has a static material, as its description does. It only names: mass still comes from the bodies. The CSV gets a `material` column at the end, so the other columns keep their places: what the row shows, the static material or the worked-out ones joined with "; ". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bill of materials: stable item numbers, saved with the design (#3874, P5a) A numbered parts list keeps each row's item number under the row's key in `assembly.item_numbers`, so a row keeps its number when others are added, removed or reordered. New rows take the next number after the highest at their level; Renumber restarts a view in list or item order and leaves locked rows alone; duplicates are marked; numbers change no geometry and are part properties in the undo history. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
5e0e168ae9 |
Properties: moment of inertia, principal axes and first moments of a part or assembly (#8095) (#8103)
* claim: #8095 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Properties: moment of inertia, principal axes and first moments of a part or assembly (#8095) component.properties and occurrence.properties report each closed body's inertia tensor about its centre of mass, and in total the tensor about the total centre of mass (world axes, g·mm²), its principal moments and axes, the first moments, and with `about: [x, y, z]` the tensor about any point. The tensor comes from each body's welded export mesh, tessellated once per body, scaled to the measured mass and turned by each placement; bodies add by parallel axes. The maths is a new crates/geom inertia module. The Properties window shows the first moments, the tensor, the principal moments and axes, and the view marks the centre of mass with the principal axes while it is open. Inspect → Measure → Properties now opens that window (it ran the command and showed nothing). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Properties: Copy matrix, and the whole design's card titled just Properties (#8095) Copy matrix, under the tensor, puts it on the clipboard as three lines of tab-separated values at full precision, in the units shown. The card for the whole design is titled Properties; one opened on a component, placement or bodies keeps its name. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Properties: measure beside the window, so a large assembly doesn't freeze it (#8095) Properties ran on the window thread: 18 s on a 144-body design, with the window answering nothing meanwhile (#5203). In the desktop app the card now measures on a copy of the session on a worker, the way #4614 made agents' measures work, says "Measuring…" until the reply is in, and Close drops it. A quick measure still shows at once; headless, tests and the web measure in place. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
d71bdb78ec |
feat: strengthen modeling, configuration, and assembly workflows
Preserve linked sketch identities through deterministic face projection and resizing. Fix origin-corner picking, persistent inline navigation, appearance editing, configuration styles, component properties, contact sweeps, sheet exports, analytic STEP offsets, and bounded rolling-ball fit accuracy. Validated with full cargo xtask ci, native workflow screenshots, repeated enclosure replays, and analytic geometry regressions. |