83f192ac2ff4bed642a8d20fbe4ff68c43aab8a0
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
|
de8dc8f813 |
fix(task-13): read revision/balloon state from the space it is drawn on, and never at 200
Three findings, one family: a guard that reads the WRONG SET silently stops
guarding and the drawing gets a duplicate or a wrong number with no refusal.
1. Layout-scoped state was read from the wrong space. add_revision called
read_revisions and add_balloon called read_balloons BEFORE enter_layout;
both read entity_list, which follows the current space, so the guard
inspected Model while the rows and balloons lived on the layout -- and the
layout is the sheet workflow (sheet_frame and titleblock_apply both take
one). Measured headlessly: two revision_add(rev="B", layout="SHEET1") calls
both returned created=True, handles 46 and 5A, both bbox
{min [230,273], max [410,287]} -- a second heading band and a second B row
drawn exactly on top of the first, twelve TEXTs on the tab instead of six.
Both reads now happen inside the entered layout, and the whole body sits in
the try/ so the caller's tab comes back on the refusal path too.
2. bom_extract truncated the parts list at 200 INSERTs and printed a wrong QTY
into a real TABLE. extract_records passed its own limit=200 default straight
to entity_list, the tool exposed no limit, and nothing in the payload said
the list had been cut: 250 bolts reported [("ISO 4014 - M12x60", 200)].
all_entities now pages entity_list (offset loop, PAGE_SIZE 1000, plus a
no-new-handles guard so a backend that ignored offset cannot spin) and
extract_records(limit=None) reads the whole drawing. bom_extract exposes
limit (0 = no cap) and reports truncated measured, by asking for one record
past the cap.
3. The balloon clash guard did not fire in two routine cases. (a) targets is
optional in the tool, so the clash test compared () != () -- False -- and
two untargeted balloons carried item 1 with no refusal. The renumber arm
above has already returned for the balloon that stands for these targets, so
anything still carrying the number is a clash, whatever its targets; the
message names the handle and says "no targets" when it has none. (b)
read_balloons and read_revisions took entity_list's 200-entity default, so a
drawing whose first 200 CIRCLEs are a hole pattern read back no balloons at
all and every guard passed blind -- and index = len(existing) restarted at
0, stacking revision bands. Both readers now go through all_entities.
Evidence, headless (ezdxf), each finding's own repro before -> after:
1 created True/True, handles 46+5A, 12 TEXTs -> created True/False, one
handle 46, 6 TEXTs; renumbered False/False -> False/True, same handle 33
2 records 200, rows [(..., 200)] -> records 250, rows [(..., 250)]
3a two balloons handles 2F+35, no refusal -> refused, naming balloon 2F
3b read_balloons() == () with 206 circles -> the balloon past the 200th
circle is found, the rerun renumbers, the item-1/BBB clash is refused
3c 200 notes + rev A twice + rev B -> created True/False/True, row 1 then
row 2 (was: every band at row 1, on top of the last)
Six regression tests added: one per finding plus the layout-scoped twin of
each, which is what would have caught all three.
The live engine was UNAVAILABLE for this round:
win32com.client.GetActiveObject("AutoCAD.Application") did not return within
90 s against the running acad.exe (PID 40068), and the ComBackend probe timed
out on its first call at both COM_CALL_TIMEOUT=60 and 180. No second AutoCAD
was launched and no document was opened. No new ActiveX call is introduced by
this fix -- it only reorders existing calls and pages entity_list(offset=),
which both engines already implement.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd
|
||
|
Umut CANandClaude Opus 5
|
499b642fbd |
feat(sheet): revisions, the ISO 7573 parts list and ISO 6433 balloons
SECTION 24 grows to six tools: revision_add, bom_extract, bom_table,
balloon_add.
bom_extract reads. It walks the INSERTs through the generic members already in
the contract -- entity_list, block_get_attributes, entity_get_xdata -- and
returns parts-list rows from the ACADMCP_MECH std_part payload, falling back to
the block's own ATTRIBs and reporting which of the two each row came from. A
block with neither is skipped rather than guessed at, and the tool never writes
to the drawing.
bom_table draws the rows as a real TABLE, 180 mm wide (ISO 7573: the item list
is as wide as the title block), growing upwards from directly above it with the
heading row against the block and item 1 the row above -- which is what ISO 7573
asks for on a drawing. A TABLE is drawn downwards from its insertion point, so
the upward list is inserted a table height above its anchor with the heading as
the last row; one table_layout feeds both the real TABLE and the pure
table_prims, so the two renderings cannot disagree.
balloon_add draws the ISO 6433 reference -- numeral, leader, dot on the item --
and writes the balloon payload naming its targets. A rerun over the same
targets renumbers that balloon instead of drawing a second one, and an item
number already used for different targets is refused by name with the other
balloon's handle.
revision_add appends a row to the revision block, is idempotent per revision
code (a repeat draws nothing at all), and optionally clouds the changed regions
and tags them with the triangle. Clouds are refused up front, carrying
capability "revcloud", on an engine that has no revcloud member -- the COM
backend -- rather than leaving half a revision behind.
Declared: ISO 7200 does not standardise the revision block's geometry and
ISO 7573 does not fix the item list's column widths or row height, so both
reuse this repository's 180 mm / 7 mm ruling and say so in their SOURCE
docstrings. ISO 6433's "numeral at least twice the dimension character height"
is met by the numeral being 1.4x the balloon radius: the 4 mm default clears
2.5 mm dimension text, and balloon_add's docstring tells a sheet dimensioned at
3.5 mm to pass radius=5.0.
Three corrections to what the plan wrote, each measured rather than assumed:
* extract_records read `block_get_attributes(...)["attributes"]`. Both engines
return the flat {tag: value} mapping itself (ezdxf_backend.py:6687 builds it
from `ent.attribs`, com_backend.py:5703 from `ref.GetAttributes()` as
{TagString: TextString}), so every attribute fallback would have been
silently empty -- measured headlessly, where a PLATE insert carrying only a
DESIGNATION ATTRIB extracted to nothing.
* draw_bom_table now reports `entity_type` / `representation` /
`child_handles`. `entity_create_table` returns EntityInfo(type="TABLE") whose
handle is the first CHILD of a composite on the headless engine, so reading
that handle back gives a rule LINE, not an ACAD_TABLE -- the contract
tests/test_table_leader.py already pins. A caller that means to edit the
table afterwards has to be told which of the two it got.
* add_revision reports one `tags` entry per tag asked for ({at, handles}), not
one per entity: a tag is two primitives, so a flat handle list claimed twice
as many tags as the caller requested.
Discovery: `bom_extract` / `bom_table` / `balloon_add` each entered the index
heavy with "parts"/"list" tokens and took top-1 for the bare query "parts list"
away from `entity_create_table` (a BM25 tie broken alphabetically). Narrowed on
my side only -- the reader speaks of the bill of materials, the drawer names a
table, the balloon names an item -- never by touching the golden case.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd
|