Files

T

earthtojakeandClaude Opus 5.5 523ae21341 Release 0.7.19: fix what day one of PostHog telemetry showed (Windows mesh export, cad_file and cad_screenshot failures, crash noise, failure reasons) (#586 )

**This PR is the 0.7.19 release** (`scripts/release/bump-version.sh
patch`): merging it runs Publish Release. Its receiver changes under
`apps/api` deploy on the same merge through Deploy API, minutes before
PyPI has 0.7.19, so schema 4 is read before any client sends it.

Fixes for what PostHog's first day of telemetry showed (2026-10-08
00:14Z to about 21:40Z: about 209 installs and 59 crash reports). It
covers three bugs people are hitting, crash reports that were not
cadgen's bugs, and gaps in what the receiver lets us see. There is one
commit per fix.

## Bugs

**1. Builds that export a mesh crashed on Windows** (7 installs, all
Windows, about 26 crashes). `mesh_export.py` ran the Node exporter with
`text=True` and no encoding, so Windows read its UTF-8 output in the
local code page. The exporter's JSON report names every output path, so
any output folder whose name the code page cannot read (for example
`Рабочий стол` under cp1252, or most Chinese text under cp936) made
CPython's Windows output reader die quietly. `proc.stdout` came back
`None`, and `.splitlines()` raised an `AttributeError`. The exporter now
reads `utf-8` with `errors="replace"`, which keeps the JSON line intact.
The same fix goes into `run_node_builder`, whose input was also silently
empty under cp1252. ffmpeg, `gz sdf` and `doctor` now read `utf-8` with
`errors="backslashreplace"`, and doctor's child process is set to
`PYTHONIOENCODING=utf-8`. The tests force subprocess's default encoding
to cp1252, and both fail without the fix.

**2. `cad_file` failed on 48 of 49 calls on Windows** (5 of 6 installs).
Codex for Windows names a file opened from its file tree as
`openai/resource.path = "/C:/Users/…"`, read from the desktop bundle.
Python 3.13's `ntpath.isabs("/C:/…")` is False, so every call answered
"not an absolute path". The `file.resourceUri` alongside it is a
`codex-resource://` handle, so the fallback never helped. A new
`local_path` drops the slash before a drive on Windows, both for file
URIs and for plain paths, for `cad_file`, `cad_open` and `cad_show`.
This most likely also explains Antigravity's `cad_show` failures on
Windows (7 of 12). The Windows CI job now passes the path the way Codex
spells it.

**3. `cad_screenshot` failed on 30% of calls** (11 of 19 installs). The
most likely cause is an agent capturing straight after build, show or
open, while the view is still loading or has not synced yet. The view
refused with "Wait for the displayed model revision to finish loading",
"That viewer is not open" or "No CAD viewer with a model is open", or a
large model ran past the fixed 10 s wait.
- The page now waits until the view shows the requested model, loaded
and drawn (`CAPTURE_SETTLE_MS`, 20 s).
- The server waits for a view it just opened to sync (`OPENING_SECONDS`,
15 s) within one budget for the whole capture (`CAPTURE_SECONDS`, 40 s).
- The capture's reply still goes on its own call (`void answer(event)`),
so no view call is held open.

## Crash reports that were not cadgen's bugs
- **Windows viewer disconnects.** `ConnectionAbortedError` (WinError
10053) made up most of the crash volume: 23 installs. The viewer caught
only `BrokenPipeError` and `ConnectionResetError`, and the header write
had no guard. Every write to the socket now treats any `ConnectionError`
as the page having left.
- **A model's own mistakes.** A build123d name that does not exist,
raised through the `cadgen.build123d` re-export, and a non-string passed
to `srgb()`. Both now raise deliberately, so the existing rule counts
them as the person's error, and `srgb` raises a `TypeError` naming what
it was given.
- **Stopped workers.** A worker stopped by SIGTERM, SIGINT or SIGHUP (a
person quitting it, a logout) now counts as cancelled, not crashed.
SIGSEGV, SIGABRT and SIGKILL are still reported.

## Telemetry: what we can now see
- **Why a tool call failed.** There is a new `tool_failure {tool,
reason, count}` event in batch schema 4, which PostHog receives as
`tool_failed`. The reason is one word from a fixed list (`no_path`,
`relative_path`, `no_file`, `not_cad`, `no_view`, `wrong_view`,
`bad_request`, `timeout`, `view_error`, `too_large`, `no_viewer`, `bug`,
`other`), chosen where the call fails and never taken from a message. A
test checks that every `ToolFailed` and `NoAnswer` names one.
- **Rollout: the receiver goes first.** The API is its own Vercel
project now (#587) and deploys on merge to `main`, so merging this PR
puts the schema 4 receiver live before any release sends schema 4. A
refused batch is dropped, as before; there is no fallback in the client.
- **Refused batches are logged.** Each 400, 403 or 415 is one
`console.warn` line naming the rule that failed and the cadgen version.
Values, install ids and service messages are never logged. Vercel's
per-status counts need Observability Plus, so this is the only way to
see a refusal. The privacy policy says so.
- **Errors are logged by name**, for example `TimeoutError` instead of
`23`. A `/v1/forget` timed out at 17:02Z, and the client retries it.
- **`$session_id`** is now set, so error tracking can count sessions.
Our ids are UUIDv4, so PostHog's sessions table leaves them out; error
tracking should still read them, which needs checking after deploy.

Privacy policy, README and `apps/api/README.md` are updated where what
is sent or logged changed.

## Not in this PR
- **Deduplicating a resent batch.** The sender rebuilds a failed window
instead of resending it, and a batch has no id, so there is nothing
stable to dedupe on yet. It needs a per-batch id from the sender.
- **Dashboard totals.** PostHog's error-tracking "occurrences" counts
events, not each event's `count`; for the mesh-export crash that is 5
against 22. That is fixed on the dashboard side (t2c-analytics).
- **5 of 15 DXF builds failed.** DXF builds don't go through Node, so
the encoding fix doesn't cover them and they still need a look.

## Needs a real host
- Windows Codex: open a `.step` from the file tree; capture from a tab
hidden behind another tab.
- Claude Desktop: capture right after `cad_show` on a large STEP, or
while the card waits on Allow.
- Antigravity on Windows: confirm the path spelling it sends.

## Tests
Full suites on this branch, in a provisioned worktree (`.venv` from
`requirements-dev.txt`, `npm ci`, `bundle.sh --check`,
`CADGEN_DAEMON=0`): all pass.
- `scripts/test/test-python.sh --keep-going`: 2,774 tests in 8 groups,
OK.
- `scripts/test/test-js.sh`: every group passes (core, ui, web, mcp).
- `scripts/test/test-docs.sh`: receiver tests 30/30 and the rest 16/16.
- `scripts/test/test-global.sh`: 210 tests, OK (1 skipped).

Each new regression test was run against the old code, and each fails
there.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-09 01:35:24 +00:00

7.6 KiB

Raw Permalink Blame History

name, description, license

name	description	license
engineering-drawing	Make an engineering drawing of a part as a PDF - orthographic views with hidden lines and centre marks, real dimensions, hole callouts, notes and a title block on ISO sheets, all projected from the part's geometry so the drawing follows the model. Use when the user asks for a drawing, a dimensioned sheet, shop or manufacturing drawings, a print, or "2D views of this part".	MIT

Engineering drawing

Provenance: maintained in earthtojake/text-to-cad. Use the installed local skill files as the runtime source of truth.

An engineering drawing is a DOCUMENT derived from a part. cadgen.eng_drawing projects views from the part's geometry, places dimensions between points ON that geometry, finds centre marks on it, and lays the result out on an ISO sheet with a frame and a title block. Change the model, rerun the drawing script, and the views, hidden lines and measured values move with it. Nothing on the sheet is drawn by hand, and nothing on it is a number you typed unless you chose to override a value.

The output is one PDF. That is what a shop receives and what every operating system opens. No DXF is written, nothing is paired or linked, and there is no CLI: the script is the interface. $dxf makes cut layouts and flat patterns, which are toolpaths, not documents; the two are different jobs.

Setup

Run cadgen through uv, so this skill's commands share one installation, and its warm build daemon, with the CAD app's server:

cadgen below means uvx --no-config --managed-python --python 3.13 --from cadgen==0.7.19 cadgen
python below means uvx --no-config --managed-python --python 3.13 --from cadgen==0.7.19 python

The first run downloads that installation; later runs reuse it.

Workflow

Get the part. cadgen.read_step(Path(__file__).parent / "../STEP/part.step") reads the artifact a @step model wrote (run python <model>.py first); a live build123d shape works just as well. The drawing documents geometry, it does not build it.
Write <name>_drawing.py beside the model, from the template. Import the model facts the dimensions reference (hole pitch, overall size, wall) from the model module rather than retyping them, so moving a feature moves its dimension.
Place views with sheet.three_views(part, iso=True), which returns FOUR views: top, front and right in third angle with room for two rows of dimensions, plus an isometric in the free corner. sheet.three_views(part) returns three. Use sheet.view(part, name, at=(x, y)) (sheet millimetres from the bottom-left corner; A3 is 420 × 297, A4 is 297 × 210) only for a custom layout.
Dimension what a maker needs, in MODEL coordinates: overall size per view (view.overall()), feature positions and sizes (view.dim, with tol= or fit= where the design requires one), holes as callouts (view.hole: thru, depth, counterbore, countersink, thread), and notes with leaders (view.note). Leave text unset so the dimension is measured; give a value only where the model does not define it. Leave offset unset too and the dimension takes the next free row outside the view; pass one only to place a dimension deliberately, and remember it is measured from the dimension's own points, not the view's edge. Keep notes few and short; one wider than the sheet's note column is refused, and notes= takes a LIST (notes="BREAK EDGES" is a sequence of eleven characters, so it is refused too).
Run python <name>_drawing.py. It prints the PDF it wrote, and anything it noticed while drawing — read those first:
- "measures blank paper": the dimension's model points do not land on the view's geometry. Almost always the part was built from a corner and the script assumed it was centred. Check the part's bounding box.
- "annotation overlaps": two callouts print on top of each other, usually one view's outermost dimension against the label of the view above it. Pass a bigger three_views(gap=...) or place one with offset=. The text is measured as the renderer draws it, glyph by glyph, so this fires on ink and not on markup: a chain of toleranced dimensions is silent when the values clear each other.
Steps 3 and 5 stand on their own: a sheet of views with no dimensions on it yet is a drawing, and writes its PDF.

When running unattended, set MPLCONFIGDIR to a writable directory so matplotlib's font cache does not warn.
Read the PDF and fix what collides. Check that no dimension text sits on a view, that hidden lines appear where features are behind faces, that every hole has a centre mark, and that no leader crosses the part. Very small dimensions (a few mm) put their value on their own extension line; dimension the larger feature instead. Move at= or offset= and rerun.

What the sheet contains

Layers with meaning: VISIBLE (heavy outline), HIDDEN (dashed), CENTER (centre marks), DIM, NOTES, TITLE, SHEET (frame), each printed at its own weight — 0.5 mm down to 0.18 mm, so thick reads against thin on paper. An edge is drawn once: a silhouette that the kernel returns as both visible and hidden stays on VISIBLE. Hidden TANGENT transitions are not drawn at all — a fillet running into a face, or the seam where a cylinder closes on itself, marks nothing a shop can see, and drawn dashed it reads as a hidden edge that is not there (on a bore, as a line down the hole's own axis).
Dimensions measured from the geometry, with filled arrowheads and witness lines, and true-size values at any drawing scale.
A title block with title, part number, material, author, scale, units, projection, revision, SHEET n OF m and the drawing function's name; numbered notes above it (the general tolerance, when given, is note 1); a revision table top-right when revisions= is given. Text longer than its cell is set smaller, and cut with an ellipsis rather than overrun.
PDF bytes that are a function of the content: no creation date is stamped, so an unchanged drawing rebuilds to an identical file.

What fails loudly

The script raises rather than writing a wrong or missing document:

Views that run off the frame, naming a scale that has been laid out and verified to fit, or saying that no standard scale does.
A part argument the vocabulary does not define: an unknown view name, sheet size or projection, orientation= other than "h"/"v", a negative or zero diameter, a counterbore that is not a (diameter, depth) pair, a tol that is not a number or a pair, a hole that is both thru and given a depth, notes= or revisions= given a bare string, a note wider than the sheet, angle() legs that are collinear or zero-length, a non-finite at= or offset=.
A missing renderer. The PDF is the drawing, so no matplotlib is a failure, not a skipped half. A failed render leaves no file behind.
out= missing or not naming a .pdf, at import time.

Limits to state in the handoff

Tolerances come only from tol=, fit= and general_tolerance= that you write; get the values from the design requirements or the user, never invented. A dimension without one is nominal.
Views are orthographic projections with hidden lines from the kernel; no sections, details, or auxiliary views yet. Say what a view does not show.
The sheet keeps callouts clear of the views it knows about, and reports annotation that still prints over other annotation, but it does not check annotation against line work. Read the PDF.

7.6 KiB Raw Permalink Blame History Unescape Escape