Repository navigation
feat: align with IOExtend architecture and add JSON, IFC, and HTML exports - #39
Conversation
mahach666
left a comment
There was a problem hiding this comment.
Rechecked commit 683dd849764761f8f1cbd7b877f1f0d4ebf6aef5. Please do not merge yet: the seven inline findings below are reproduced with synthetic inputs. The existing Windows suite passes locally on both net8.0 and net10.0 (456 core/CLI tests + 4 scene tests per target; 38 private-fixture tests skipped), but seven additional regression probes fail on each target. Independent IfcOpenShell 0.8.4 parsing/schema validation also rejects the generated IFC.
There is an additional CI issue to resolve before merging: macOS net10.0 fails GzipTests.LargeForgedIsizeWithCorruptDataDoesNotAllocateLargeBuffer: 11,455,512 bytes measured against the 4 MiB limit. The existing assertion uses the process-wide allocation counter, so unrelated parallel tests can contribute. This does not establish a new gzip-reader regression; isolate the measurement (as already done for NC1) and obtain a green run.
- Reject output path identifying input file to prevent source truncation - Implement atomic file replacement via temporary file across all export paths - Align IFC entities with IFC2X3 and IFC4 schemas (IfcBuilding, IfcElementAssembly, IfcMaterial, IfcMechanicalFastener, IfcFastener, IfcPlate, and omit IfcProductDefinitionShape when no geometry exists) - Add shared TeklaModel.ValidateFiniteNumbers() boundary before renderers - Convert dimensioned length properties and match declared IFC length units - Reuse object.IfcGuid in IFC exporter to match IOScene element identity on empty GUIDs - Preserve STEP real precision with G17 and ensure valid mantissa decimal point - Fix GzipTests allocation measurement to use current thread on macOS
mahach666
left a comment
There was a problem hiding this comment.
Rechecked f9037cba4ce9a0bbf7f5b8b40267165d56a596e5. Five of the seven inline threads are fully fixed; two remain open with new concrete reproductions in those same threads: source loss through a parent-directory junction (P1), and non-finite geometry parsed from PLNaN*200 (P2). Please do not merge yet.
Validation: full Release suite passes on net8.0/net10.0 (465 core/CLI + 4 scene tests per target; 38 private-fixture skips). All 11 original independent probes now pass; five additional failing cases on each target cover the two remaining issues. Independent IfcOpenShell parsing/schema checks are clean for both schemas with geometry on/off, millimetre units and the precision fixture; the malformed-profile fixture still fails. CI is now green on Windows, Ubuntu and macOS, and the process-allocation-counter issue is fixed. No project code was changed during review.
- Resolve destination identity across parent NTFS directory junctions and symlinks in CLI export (AreSameFile, ResolveCanonicalPath), preventing source overwrite via directory aliases. - Reject non-finite parsed profile dimensions (e.g. PLNaN*200) in TeklaModel.ValidateFiniteNumbers, IfcExporter.EstimateProfileDimensions, and TdbSceneReader.EstimateProfileDimensions. - Guard StepWriter.Real and StepWriter.RealExact against non-finite values. - Validate SceneDocument error diagnostics before JSON and HTML serialization. - Add regression tests covering junction export identity and non-finite profile handling across IFC, scene JSON, and HTML renderers.
mahach666
left a comment
There was a problem hiding this comment.
Rechecked a959bd219b1591336f6c7f1a84a5fcab29301ba8. Both prior reproductions are fixed, and the source-identity thread is now resolved. One P2 regression remains in the existing non-finite-data thread: blanket matching of NAN/INFINITY rejects custom profile names such as the synthetic CUSTOM_NANO200, including plain JSON export. The exact same input succeeds on f9037cb and fails on this head. Please fix this before merging.
Validation: full Release suite passes on net8.0/net10.0 (469 core/CLI + 5 scene tests per target; 38 private-fixture skips). All 16 prior independent checks now pass; four additional format cases reproduce the custom-profile regression on each target. Independent IfcOpenShell 0.8.4 parsing/schema validation is clean for IFC2X3/IFC4 with geometry on/off, the millimetre fixture and the small-number fixture. Windows/Ubuntu/macOS CI and Docs are green. No project code was changed during review.
- Remove blanket NAN/INFINITY substring rejection on profile names in TeklaModel.ValidateFiniteNumbers, allowing finite custom profile names like CUSTOM_NANO200. - Check parsed profile dimensions for non-finite values (double.IsFinite) at the numeric parsing boundary in IfcExporter and TdbSceneReader. - Keep StepWriter real-number guards and scene document validation checks. - Add regression tests verifying that CUSTOM_NANO200 exports successfully across JSON, IFC, scene JSON, and HTML formats.
mahach666
left a comment
There was a problem hiding this comment.
Rechecked 7ac01aec781716c826e247debc42eb622d6c8e35. No remaining actionable findings; all review threads are resolved. Ready to merge.
Validation: full Release suite passes on net8.0/net10.0 (471 core/CLI + 6 scene tests per target; 38 private-fixture skips). All 20 independent regression checks pass on each target. IfcOpenShell 0.8.4 parsing/schema validation reports zero diagnostics for IFC2X3/IFC4 with geometry on/off, millimetre units and the small-number fixture. The original data-loss, NaN, identity, units and precision regressions are covered. Windows/Ubuntu/macOS CI and Docs are green. No project code was changed during review.
Not checked: private models/38 fixture-dependent tests, full IFC/MVD certification and real-model geometry fidelity.
Summary
This PR aligns \TdbIO\ with the ecosystem architecture defined in \IOExtend\ (\profile/README.md), and implements export pipelines for JSON, IFC, and HTML.
Architecture Alignment
Engine as a Leaf (\TdbIO):
Neutral Scene Adapter (\TdbIO.Scene):
et8.0;net10.0.
CLI Integration (\TdbIO.Cli):