e6f685643d768bf75944cfa95c7ef0efb113ec2e
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Umut CANandClaude Opus 5.5
|
898e775f93 |
bench: matrix v5 adds mech_assembly; the correctness suite grows with the mechanical and sheet checks
The evidence for tracks B and G, in the shape tracks A and E used.
- benchmarks/tasks_v5.py: 18 tasks, DEFAULT_MATRIX = "v5"; v4, v3 and v2 stay
addressable so an earlier report reproduces exactly rather than being only
described. mech_assembly is roadmap criterion 3 - a flange-coupling sheet
built from the part model and the sheet standard, through the code the
tools call (apply_titleblock, draw_part/add_view/dimension_part,
insert_std_part, extract_records -> draw_bom_table, add_balloon), gated on
two numbers that cannot be negotiated: drawing_critique(focus=None) returns
zero issues and the finalize score is at least 90 (it scores 100, 18/18 on
the v5 matrix). The adapter returns the focuses that fired, so a
regression names itself, and a test drops one balloon to prove the gate
fails and names mech_bom_balloon_mismatch.
- Seven correctness checks (32 -> 39). mech_section_hatch_area is the one
that cannot be faked: a sleeve of length 60, outside diameter 40 and bore
20 cuts two faces totalling 1200 mm^2 exactly, read back out of the drawing
through entity_measure. sheet_frame_iso5457_a3 pins the A3 frame (20,10 to
410,287 inside the 420 x 297 trimmed sheet); mech_iso286_on_dimension pins
40 H7 at +0.025/0 through the layout; std_part_iso4014_m12 pins s = 18 and
k = 7.5 by value; mech_thread_unrepresented_is_caught is the seventh: a
quality gate that never fires is not a gate.
- scripts/smoke_mech_com.py executed every new COM path once on AutoCAD 2026
(25.1s): frame, title block, a sectioned shaft with 12 dimensions, an ISO
4014 block with 1 ACADMCP_MECH chunk, a weld symbol, the parts list read off
the drawing, a linked balloon, an xref attached / listed / detached, a
31687-byte .dwg (AC1032) through ActiveX SaveAs, critique 0 issues, exit 0,
every scratch document closed. It retries RPC_E_CALL_REJECTED (measured:
the first call after Documents.Add is refused) and tracks its documents by
the document list, since SaveAs renames them.
- Defects the evidence found, each fixed with a test that fails on the old
code:
* ezdxf dimension_diameter put the text beyond the SECOND chord point;
ActiveX AddDimDiametric puts it beyond the first (measured live: a d50
from (x,170) to (x,220), leader 10, text at (x,160)). Headlessly every
front-view diameter stood on its own chain dimension. ezdxf now follows
ActiveX.
* draw_part's anchor goes on MECH; ActiveX refuses entity.Layer on a
missing layer (measured -0x7ffdfff7 'Key not found'). draw.py ensures
every target layer before the first entity.
* bom.extract_records filtered on INSERT; ActiveX names it BLOCKREFERENCE
(ObjectName AcDbBlockReference) - the live parts list was empty.
* COM _entity_info reported no dimension text_position: ModelSpace.Item
narrows to IAcadEntity, which has no TextPosition (measured
AttributeError); read through _com_unnarrow (TextPosition measured on
AcDbDiametricDimension, WCS). dim_overlap compared bbox centres and a
diameter collided with its own bore.
* pid graph raised KeyError 'start' mid-critique on a LINE whose point read
AutoCAD refused while busy; such a line is left out.
* the validator read "ISO 4014 - M12x60" as fake dimension text.
- The published A/B report is re-run over the grown suite rather than
re-typed (26/39 -> 39/39, 13 fixed, 0 regressed), check_doc_numbers derives
the task count from the v5 matrix, the token table is remeasured with a
TOOL_PACKS=core,mech row (193 tools / 61,673 tokens), and the README's
collected-tests figure is retaken (3990).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd
|
||
|
Umut CANandClaude Opus 5
|
bd73b3a854 |
fix(task-6): a reachable ISO 286 route, a tolerance schema that refuses, a reference dimension
Three review findings, all root causes in `engineering/mech/dimension.py`.
1. The ISO 286 fit path was unreachable from the module's own pipeline.
`resolve_tolerance` resolved `intent.fit` correctly, but nothing produced a
DimIntent carrying one: no feature and no segment has a fit, and
`dimension_intents` constructed every intent without `fit=`. Only a
hand-built DimIntent - which is what the tests used - reached the code, so
the spec's `mech_iso286_on_dimension` and the `mech_assembly` benchmark's
"ISO 286 fits on the bearing seats" had no route.
`dimension_intents(..., fits={key: code})` now stamps the code onto the
intent its key names, and the part's own diameters became addressable one
at a time through the new `segment_key(i, bore=...)` -> "segment[1]",
"segment[1].bore"; a feature's stay addressable by id. A key that names
nothing is refused with the names that exist, a malformed code is refused
where it is attached rather than at draw time, a fit on a reference
dimension or on one that already carries a tol is refused, and a fit on a
hole whose diameter moved into the hole table (which has no tolerance
column) is refused by name instead of vanishing. Task 7's planned
`dimension_part(..., fits=None)` passes straight through - the argument is
additive and keyword-only, so every call written against the pinned
signature keeps working.
The fit also now reaches the *text*: `intent_text` expands the `<>`
placeholder and appends the code, so `text_rects` measures the string that
is actually drawn and a hole that already names itself ("⌀6") keeps its own
text and gains the code.
2. An explicit tolerance was silently discarded when `tol` omitted "mode".
`mode = tol.get("mode", "none")` and `build_dim_override(0.05, 0.05,
"none", None)` returns `({}, None)` without raising, so a ±0.05 resolved,
echoed its own numbers back in the payload, and drew a dimension with no
DIMTOL/DIMTP/DIMTM at all. `DimIntent.tol` now has a schema - {"mode",
"upper", "lower"} and nothing else - validated by `_explicit_tol`: an
unknown key, a missing "mode", a non-dict, and deviations under a mode that
writes none are each refused by name. The mutual-exclusion guard reads
`tol is not None` rather than truthiness, so `tol={}` is refused too.
3. The ISO 129-1 "no measurement twice" test was vacuous and the property was
false. The old test built the key `_dedupe` uniquifies as the last step of
`dimension_intents`, so it could not fail; meanwhile `style="chain"` emitted
20, 40, 15 *and* 75 as four plain DIMENSIONs. `redundant_linear` now finds
the span that shorter ones tile end to end on the same axis and level, and
`dimension_intents` marks it as an auxiliary (reference) dimension - the new
`REFERENCE_FORMAT` "(<>)", which renders "(75)". The plan's own test still
holds (the 75 is present); it is no longer a fourth plain dimension. The
error is one-sided on purpose: a tiling that is found is real, a tiling the
greedy walk misses leaves the dimension plain, which is the direction the
repository's critique rule wants. Baseline and ordinate parenthesise
nothing, and two horizontal chains at different heights are not each other's
pieces.
Tests: the vacuous gate is replaced by four that fail against the pre-fix code
- `redundant_linear` on the shaft's plain intents returns (3,) there and ()
here; the parametrised "a tolerance is either refused or reaches the dimension
variables" gate resolves the mode-less shape to an empty DIMVAR override
there; and `fits=` is a TypeError there. `_dedupe` is now exercised through a
duplicate fed in via the view's own dims, which the collector does not
generate itself. 46 tests in tests/test_mech_dimension.py (was 23).
Live AutoCAD: no finding was measured live and no new ActiveX call was
written - this module is pure Python with no backend, no async and no DXF. A
DIMTOL probe against the running seat was attempted twice and
`GetActiveObject("AutoCAD.Application")` blocked past the ~90 s contention
window both times, so the live engine was unavailable; no scratch document was
created (the probe never got the application object) and none was left open.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd
|
||
|
Umut CANandClaude Opus 5
|
01213fc5f5 |
feat(task-6): dimension intents, ISO 286 fits and an overlap-free layout
engineering/mech/dimension.py collects what the features asked to have dimensioned, adds the part's own sizes (chain, baseline or ordinate for the axial lengths, a diameter per segment and per bore, the two overall sizes of a plate), and removes the duplicates - ISO 129-1 says each dimension appears once, and a test asserts no measurement is placed twice. Fits and tolerances are resolved through the modules that already own them: engineering.fits for an ISO 286 code and engineering.tolerances for an explicit ISO 129 tolerance, including build_dim_override's sign convention (the lower deviation is passed as the magnitude of the minus side, exactly as server.py::_fit_to_tolerances does). A H7 bore at 40 mm resolves to +0.025/0 - the correctness check the spec names. A fit together with an explicit tolerance, a fit outside the authored ISO 286 range, and a fit on an intent with no measurable nominal are each refused by name rather than guessed. The layout is the part that had to be measured rather than asserted: every intent's text is turned into a rectangle by the same helper the test uses, and an intent is placed on the lowest level whose rectangle is free, with the level pitch clamped to clear the text whatever the caller passed. The test then asserts the placed rectangles are pairwise disjoint - including on a crowded plate and with a deliberately too-small step. That is the same property the repository's dim_overlap critique focus checks on the finished drawing. The plan's layout clamped that pitch to the text HEIGHT on every axis, which is only the stacking direction for the row of dimensions under the view. Beside the view the levels march in x, so the extent they have to clear is the text WIDTH: measured here, three diameter labels are 9.45 mm wide (9.45, 5.6) at a 8.0 mm pitch, so a shaft's three segment diameters - all on the y = 0 centreline, hence all on different levels - overlapped by 1.45 mm. The pitch and the within-level span are now both taken per axis from the same _text_size helper text_rects uses, which makes the disjointness a property of the algorithm rather than of the label lengths that happened to be in the test. Over hole_table_threshold holes the per-hole diameters leave the view and become a table instead, grouped by diameter, thread, counterbore and depth, one tag per group and the quantity on every row. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd |