Examples and actual execution results#

Historical Node baseline only (codex/node-baseline). For the current Rust installation, font policy and commands, see installation and architecture.

This page records a local Linux run of the repository's v0.2.0 implementation on 2026-09-26. Commands run from the repository root; npm test builds the packages itself. The recorded environment was Node.js v24.13.1, npm 11.19.1, and Python 3.13.11. Values below are local observations, not inferred results from Windows or remote Ubuntu CI.

1. Inspect the image, then edit the code#

The feature gallery has 44 focused examples under images, positioning, groups, drawing, typography, and scripting. Notebook output and complete applications have separate pages.

Actual PNG-input gallery render

Source	Functions and capabilities	Actual output	Observed `validate` dimensions
hello.lay	`canvas/rect/text/add`, center anchor	PNG	160 × 100 mm; 3 top-level placements
basic.lay	Image, shapes, text, arrow, group, independent crops and anchors	PNG	180 × 120 mm; 7 placements
functions.lay	Built-ins, functions, loops, conditions, unit arithmetic	PNG	120 × 55 mm; 3 placements
scripted.lay	Imports/exports, defaults, loops, nested groups, module-relative assets	PNG	180 × 90 mm; 3 placements
vector.lay	Paths, presets, gradients, dashes, fusion	PNG	180 × 120 mm; 9 placements
outlines.lay	Single/double/triple outlines, dashes, opacity	PNG	180 × 110 mm; 13 placements
typography.lay	Text spans, fonts, wrapping, inline/display formulas	PNG · PDF	160 × 100 mm; 8 placements
showcase.lay	Combined image, shape, path, text, component, and fusion features	PNG · SVG · PDF · 1200 DPI PNG	160 × 226.262626… mm; 1 top-level group

Vector commands are exercised by the figures and tests/vector.test.mjs; exact parameters are in the language reference. Preview images are reduced for fast display. The eight complete examples put images directly below their code in the complete-applications gallery.

Actual complete showcase render

2. CLI stdout and file dimensions#

After building, validate each example:

1npm run build2for f in examples/hello.lay examples/basic.lay examples/functions.lay examples/scripted.lay examples/vector.lay examples/outlines.lay examples/typography.lay examples/showcase.lay; do3  node packages/cli/dist/main.js validate "$f"4done

npm run build
for f in examples/hello.lay examples/basic.lay examples/functions.lay examples/scripted.lay examples/vector.lay examples/outlines.lay examples/typography.lay examples/showcase.lay; do
  node packages/cli/dist/main.js validate "$f"
done

The actual CLI output is Chinese; it reports a valid path, physical size, and top-level instance count:

1有效：examples/hello.lay（160 × 100 mm，3 个顶层实例）2有效：examples/basic.lay（180 × 120 mm，7 个顶层实例）3有效：examples/functions.lay（120 × 55 mm，3 个顶层实例）4有效：examples/scripted.lay（180 × 90 mm，3 个顶层实例）5有效：examples/vector.lay（180 × 120 mm，9 个顶层实例）6有效：examples/outlines.lay（180 × 110 mm，13 个顶层实例）7有效：examples/typography.lay（160 × 100 mm，8 个顶层实例）8有效：examples/showcase.lay（160 × 226.26262626262627 mm，1 个顶层实例）

有效：examples/hello.lay（160 × 100 mm，3 个顶层实例）
有效：examples/basic.lay（180 × 120 mm，7 个顶层实例）
有效：examples/functions.lay（120 × 55 mm，3 个顶层实例）
有效：examples/scripted.lay（180 × 90 mm，3 个顶层实例）
有效：examples/vector.lay（180 × 120 mm，9 个顶层实例）
有效：examples/outlines.lay（180 × 110 mm，13 个顶层实例）
有效：examples/typography.lay（160 × 100 mm，8 个顶层实例）
有效：examples/showcase.lay（160 × 226.26262626262627 mm，1 个顶层实例）

Export the basic figure:

1node packages/cli/dist/main.js render examples/basic.lay -o basic.svg2node packages/cli/dist/main.js render examples/basic.lay -o basic.pdf3node packages/cli/dist/main.js render examples/basic.lay -o basic.png --dpi 300

node packages/cli/dist/main.js render examples/basic.lay -o basic.svg
node packages/cli/dist/main.js render examples/basic.lay -o basic.pdf
node packages/cli/dist/main.js render examples/basic.lay -o basic.png --dpi 300

The three calls print 已导出：basic.svg, 已导出：basic.pdf, and 已导出：basic.png (“exported”). The local run used a temporary directory and inspected files with sharp, pdfinfo, and pdftotext:

File	Observed result	Check
Basic SVG	Root `width="180mm" height="120mm" viewBox="0 0 180 120"`	Physical page and internal coordinates
Basic PDF	1 page, `510.236 × 340.157 pt`	`180 × 120 mm × 72/25.4`
Basic PNG, 300 DPI	`2126 × 1417 px`, 300 DPI metadata	Per-axis rounding; post-arrow-fix SHA-256 `371aabf049894afd265537e6c02538025dcc43384119266fdf2be97f43d9135c`
Typography PNG, 150 DPI	`945 × 591 px`, 150 DPI	Layout remains 160 × 100 mm
Showcase PNG, 150 DPI	`945 × 1336 px`, 150 DPI	Physical aspect ratio
Showcase PNG, 1200 DPI	`7559 × 10690 px`, 1200 DPI	Print-resolution export

The regenerated print showcase has post-seam-fix SHA-256 34e191d363496f12c22933a526eedc078e5b097f5d0bd3e29ab4bcc3555f2fe5 in this run. Verify local files with:

1pdfinfo basic.pdf | rg 'Pages:|Page size:'2pdftotext examples/typography.pdf -

pdfinfo basic.pdf | rg 'Pages:|Page size:'
pdftotext examples/typography.pdf -

Other measured PDF page sizes: scripted.pdf 510.236 × 255.118 pt; vector.pdf 510.236 × 340.157 pt; outlines.pdf 510.236 × 311.811 pt; typography.pdf 453.543 × 283.465 pt. The stored typography.pdf was also regenerated from current source.

3. Observable functions and text#

functions.lay uses append([2,4],6), for, range(3), and while to reach total=18. min/max/abs retain mm units; twice(2 mm) gives 4 mm. Run:

1node packages/cli/dist/main.js render examples/functions.lay -o functions.pdf2pdftotext functions.pdf -

node packages/cli/dist/main.js render examples/functions.lay -o functions.pdf
pdftotext functions.pdf -

Actual extractable text:

1Functions and control flow2len=3 total=18 ready3min=3 mm max=8 mm abs=4 mm twice=4 mm

Functions and control flow
len=3 total=18 ready
min=3 mm max=8 mm abs=4 mm twice=4 mm

Actual function and control-flow render

The 150 DPI PNG is 709 × 325 px. Further PDF extraction confirms content from components, outlines, and typography:

1scripted.pdf:    Card 1 | Card 2 | Card 32outlines.pdf:    SINGLE | DOUBLE | TRIPLE | CUSTOM3typography.pdf:  文字排版 / Typography4                 行内公式：\frac{a+b}{\sqrt{x}} 与文字共用基线。5                 \int_0^1 x^2\,dx = \frac{1}{3}

scripted.pdf:    Card 1 | Card 2 | Card 3
outlines.pdf:    SINGLE | DOUBLE | TRIPLE | CUSTOM
typography.pdf:  文字排版 / Typography
                 行内公式：\frac{a+b}{\sqrt{x}} 与文字共用基线。
                 \int_0^1 x^2\,dx = \frac{1}{3}

pdftotext retrieves ordinary text and the original formula LaTeX. Visible SVG formulas remain paths. Open typography.pdf to inspect selection behavior in your PDF reader.

4. Node API execution#

Run this from the repository root after npm run build:

1import { parse, compileFile } from '@laymesh/core';2import { renderSvg, renderPdf, renderPng } from '@laymesh/render';3import { readFile } from 'node:fs/promises';4 5const ast = parse(await readFile('examples/basic.lay', 'utf8'), 'examples/basic.lay');6const scene = await compileFile('examples/basic.lay');7const [svg, pdf, png] = await Promise.all([8  renderSvg(scene), renderPdf(scene), renderPng(scene, 300)9]);10console.log('statements:', ast.length);11console.log('scene:', scene.width, scene.height,12            scene.nodes.length, scene.warnings.length);13console.log('formats:', svg.startsWith('<svg'),14            pdf.subarray(0, 4).toString(), png.subarray(1, 4).toString());

import { parse, compileFile } from '@laymesh/core';
import { renderSvg, renderPdf, renderPng } from '@laymesh/render';
import { readFile } from 'node:fs/promises';

const ast = parse(await readFile('examples/basic.lay', 'utf8'), 'examples/basic.lay');
const scene = await compileFile('examples/basic.lay');
const [svg, pdf, png] = await Promise.all([
  renderSvg(scene), renderPdf(scene), renderPng(scene, 300)
]);
console.log('statements:', ast.length);
console.log('scene:', scene.width, scene.height,
            scene.nodes.length, scene.warnings.length);
console.log('formats:', svg.startsWith('<svg'),
            pdf.subarray(0, 4).toString(), png.subarray(1, 4).toString());

Actual output with node --input-type=module:

1statements: 172scene: 180 120 7 03formats: true %PDF PNG

statements: 17
scene: 180 120 7 0
formats: true %PDF PNG

parse returned 17 top-level statements. Compilation produced seven top-level placements and no warnings, and all renderers returned the expected format data. See the API contract.

5. Python bridge execution#

With a built CLI and importable laymesh Python package:

1from laymesh import render_source2 3r = render_source(4    r"""page=canvas(size=(20 mm,10 mm))5box=rect(size=(5 mm, 5 mm),fill="#087f8c")6page.add(box,target=page.top_left)7"""8)9print('preview:', r.preview_svg.startswith('<svg'))10print('output:', r.output)11print('saved_source:', r.saved_source)

from laymesh import render_source

r = render_source(
    r"""page=canvas(size=(20 mm,10 mm))
box=rect(size=(5 mm, 5 mm),fill="#087f8c")
page.add(box,target=page.top_left)
"""
)
print('preview:', r.preview_svg.startswith('<svg'))
print('output:', r.output)
print('saved_source:', r.saved_source)

Observed with PYTHONPATH=python python3:

1preview: True2output: None3saved_source: None

preview: True
output: None
saved_source: None

The bridge returned a preview without persisting files. Python tests cover Figure vector/raster fallback, bindings, source protection, and magics; both actual Figure outputs are stored in the repository. See the Notebook guide.

6. Automated tests and failures#

1npm test2python3 -m unittest discover -s python/tests -v

npm test
python3 -m unittest discover -s python/tests -v

The recorded run passed 41/41 Node and 9/9 Python tests. Node covers assets, units/layout, modules, groups, fusion, outlines, fonts/formulas, located diagnostics, arrow heads, and round seams. Python covers bindings, Figure SVG/PNG adaptation, saved-source protection, and magics. One Python Figure test intentionally emits W_PLOT_BOUNDS for content beyond the Figure canvas and still passes.

An example failure is node packages/cli/dist/main.js validate /tmp/no-such-layout.lay, which exits 1 and writes a located E_FILE message such as:

1/tmp/no-such-layout.lay:1:1: E_FILE: 无法读取布局文件：/tmp/no-such-layout.lay

/tmp/no-such-layout.lay:1:1: E_FILE: 无法读取布局文件：/tmp/no-such-layout.lay

Invalid units, missing assets, unsupported SVG/TIFF, loop limits, and fusion failures have nonzero exits or LayError assertions in tests/*.test.mjs. The CI workflow configures Windows and Ubuntu, but a local Linux pass is not a remote platform result.

7. Focused gallery acceptance#

node scripts/build-gallery.mjs --check runs CLI validate and 150 DPI PNG export on all 44 focused .lay files, checks physical and pixel dimensions, compares preview pixels, and checks both language pages. In the recorded run, 44/44 succeeded at 120 × 80 mm and 709 × 472 px with no compile or render warnings. Each category page presents exact source, commands, raw validation output, and actual image.

Actual path-command output from the gallery

python scripts/build-notebook-gallery.py --check runs the Python bridge. The line plot saves .svg with no fallback; the heatmap saves .png and reports Matplotlib 图 fig 已改用 240 DPI PNG (“Figure fig switched to a 240 DPI PNG”). Both final pages are 709 × 472 px at 150 DPI; the fallback plot asset uses 240 DPI. See the two actual pages.

Spot checks of SVG input, paths, display formulas, colored spans, and the Notebook vector plot found no PDF image objects with pdfimages -list. pdftotext extracted \int_0^1 x^2\,dx=\frac{1}{3} and COLOR IN ONE LINE; pdfinfo reported 340.157 × 226.772 pt for the 120 × 80 mm formula page. Their visual layouts were inspected.

python scripts/check-gallery-docs.py checks source links, inline image placement, focused entries, and extra examples across both languages. The checks reject reader-facing internal Scene version identifiers. No formal release was performed.

8. Arrow and round-seam regression#

The arrow shaft now ends at the head base; a short arrow draws only its head, while end still denotes the tip. Round fusion uses local arcs instead of asymmetric probe patches. The arrow and fusion images were regenerated. The latter aligns the rectangle midpoint and line start at 45.5 mm.

Regression tests cover horizontal, diagonal, reverse, short, dashed, compound-outline, and fused arrows; aligned, angled, and zero-gap round junctions; and preservation of remote holes. SVG, PDF, and PNG exports of the gallery cases were visually compared. pdfimages -list found no bitmap objects. Gallery checking compares actual pixels, so an outdated same-size preview cannot pass.

Additional cases joined horizontal/upward/downward lines to rectangles, ellipses, convex and concave Bézier curves. Twelve PNG cases stayed connected with no broken shaft; SVG and PDF remained vector paths. If a concave curve's bounding-box anchor misses its visible contour, round fusion returns a located E_FUSE rather than creating a thin accidental neck. Omitting points uses nearest visible-outline points.

Actual angled line joining a straight boundary

Actual angled line joining a free curve

Hollow rectangle joins were checked at the right-edge midpoint and upper-right corner for round, miter, and bevel. Preserving PathKit's even-odd fill rule keeps the inner hole open after fusion. Both examples retain their hole and a connected outline. See the edge and corner sources and commands.

Actual join at a hollow rectangle edge

Actual join at a hollow rectangle corner

9. Bilingual site and native plot verification (2026-09-27)#

The test counts in sections 6 and 7 belong to the 2026-09-26 run. After adding the native plot references, measured benchmark, and bilingual site, the current run passed 50/50 Node and 12/12 Python tests. npm run gallery:check revalidated and rerendered 44 focused examples at 150 DPI, compared committed preview pixels, and checked the Notebook SVG asset and heatmap PNG fallback. python scripts/check-gallery-docs.py checked 42 Markdown pages, 196 inline images, and both languages of all 44 focused and 10 extended examples. npm run docs:build && npm run docs:check checked 42 static HTML pages, 248 image references, local links, anchors, and language switches.

Validating the native plot source returned 有效：examples/plot/scientific.lay（180 × 82 mm，4 个顶层实例）. Its newly rendered 180 DPI PNG is 1276 × 581 px, pixel-identical and byte-identical to the committed preview. SVG and PDF were also exported. pdfinfo reported one 510.236 × 232.441 pt page; pdftotext extracted Experiment, Reference, Intensity, Value, and Time (s). The default PDF heatmap data layer is a 4 × 4 raster image; axes, labels, and the other layers remain vector. Use mode="vector" for per-cell vector heatmap output.

1npm run laymesh -- validate examples/plot/scientific.lay2npm run laymesh -- render examples/plot/scientific.lay -o scientific.png --dpi 1803npm run laymesh -- render examples/plot/scientific.lay -o scientific.pdf

npm run laymesh -- validate examples/plot/scientific.lay
npm run laymesh -- render examples/plot/scientific.lay -o scientific.png --dpi 180
npm run laymesh -- render examples/plot/scientific.lay -o scientific.pdf

Actual native scientific plot export

At a 390 px browser viewport, the Chinese and English plot pages and English benchmark loaded their images without horizontal overflow; the plot-page language switch opened its counterpart. Regression tests also cover missing data, log axes, size reflow, portable saved layouts, clipping, overlapping scatter opacity, and nearest-neighbor heatmap pixel colors. The benchmark compares native plotting, standalone Matplotlib, and the Figure bridge; its data supports reduced bridge overhead, not a general speed advantage over standalone Matplotlib. The site was built and previewed locally; no formal release was performed.

10. Fixed data area and annotations (2026-09-28)#

Added plot_area=box(offset=(left,top), size=(width,height)), placed-instance data(x=...,y=...), nine plot_ data-rectangle anchors, and anchor="start"/"end" for lines and arrows. The standalone example fixes an 88 × 55 mm data area with its top-left at (26,20) mm on the page. Its peak arrow, threshold label and title are separate editable elements. PNG, SVG and PDF were exported and checked; the PDF has one 135 × 100 mm page, extractable text and no raster image objects.

This run passed 57/57 Node and 12/12 Python tests. Seven new tests cover locked data and label geometry after frame changes, font/tick changes, automatic and log mappings, annotations after domain changes, rotated anchors and arrow endpoints, containing-group scaling, and located errors. The existing scientific.lay 180 DPI PNG remains byte-identical to its preview. All 44 gallery examples and Matplotlib bridge checks passed. The documentation site validated 42 pages and 250 image references; performance was not remeasured in this run.

A same-day update changed drawable space shortages, overlapping labels and frame overflow to W_PLOT_LAYOUT, retaining settings and continuing output. Margin warnings include the approximate additional mm needed on each side. Nonpositive data rectangles and invalid data remain errors. Node 61/61 tests passed, including cached placement warnings, successful CLI validation and three-format export, retention of crowded content, and final-layout warnings after automatic tick thinning. A Python bridge check displayed the warning while generating an SVG preview and PDF. Both scientific and annotations PNGs remain byte-identical to their previews.

11. Scientific styling and placement precision (2026-10-02)#

Added explicit minor ticks, tick directions and lengths, tick-label visibility, major/minor grids, boxed axes, separate tick/label font sizes, and multi-column legends with physical positioning and optional transparent backgrounds. See the plotting reference. Panel data rectangles and placement anchors remain explicitly authored, without automatic alignment or cross-panel resizing. Legends never reserve automatic margins; insufficient space around fixed data rectangles retains geometry and emits a warning.

The editable example exports PDF, SVG, and PNG with two exact 64 × 45 mm data rectangles at page positions (20,26) mm and (115,26) mm, without layout warnings. PDF/PNG output was visually checked and PDF text extraction verified. This also exposed and fixed an unpainted PDF path left by transparent legend backgrounds that could be stroked by subsequent nodes.

npm test passed 67/67. Six added regressions cover linear/log ticks, grid clipping and draw order, fixed panel and annotation coordinates, multi-column legends and cached placement warnings, hidden ticks, argument diagnostics, and PDF pixel equivalence with invisible shapes removed. Python unittest passed 12/12. Existing scientific and annotations SVGs remain byte-identical to the pre-change output. The documentation build and check passed 42 pages and 252 image references. Performance was not remeasured.

12. Markers, scientific labels, and colorbars (2026-10-02)#

Added diamonds, downward triangles, transparent interiors, independent outlines, and optional markers for lines/error bars. Data and legend symbols share geometry; error-bar legends show stems and caps in their actual directions. Axis, layer, and colorbar labels accept existing text/formula materials. Tick formatting supports mathematical scientific notation and shared multipliers with physical offsets. Colorbars support four sides, manual coordinates, both orientations, independent length/thickness, and explicit ticks. New compilations use Scene schema 5; schema 4 solid point batches pass renderer compatibility checks. See the plotting reference.

Three new editable examples export SVG, 180 DPI PNG, and PDF without layout warnings: black-and-white markers, scientific labels, and four colorbar sides. SVG/PNG and rasterized PDF were visually inspected for hollow symbols, formula scripts, mixed labels, horizontal/vertical strips, and combined legends. PDF extraction retains ordinary text and formula sources including I-I_0, E(t), and multipliers.

Visual inspection also exposed an existing MathJax adapter bug: inline output could contain multiple SVGs, but only the first was retained, truncating formulas such as I-I_0 and E=mc^2 after an operator. Formulas now remain complete measured assets. A regression test and the inline-formula gallery preview were updated.

Final tests passed 82/82 Node and 13/13 Python. Coverage includes exact fixed rectangles and adjacent-panel coordinates, original-value data anchors, repeated placement, rotations and group scaling, PNG/PDF marker bounds and opacity, legend order and directions, module-relative fonts, material widths/wrapping, negative/zero values, exponent rounding, very large/small numbers, explicit multipliers, color normalization, and insufficient-space warnings. Existing scientific/annotations/publication SVG and 180 DPI PNG outputs remain byte-identical to the baseline; PDF pixels rendered at 180 DPI also match exactly.

All 44 focused gallery examples and Notebook bridge checks passed. Markdown validation covers 42 pages and 206 inline images; the built site validates 42 pages and 258 image references. Panels remain manually positioned; insufficient fixed layout space warns without adjustment. Colorbars still bind one heatmap in their own plot, with linear normalization. Performance was not remeasured, and the previous benchmark is not used to infer performance of these additions.

13. Independent axis title and multiplier offsets (2026-10-02)#

Added axis(label_offset=(dx,dy)) alongside exponent_offset to control titles and multipliers independently. A short centered x-axis title shares the row below the ticks with the right-aligned multiplier. Only overlapping measured default bounds move the title below the multiplier, with a 1.2 mm gap. Offsets apply after default positioning; manual overlaps and insufficient space retain settings with warnings. Displacements use chart-local physical coordinates and follow chart rotation and containing-group scaling. Colorbar rules and manual panel positioning retain their existing behavior. The plotting reference also explains page size, plot frame size, and margins/fixed data rectangles.

npm test passed 89/89 and Python unittest passed 13/13. Seven new Node checks cover short, long and multiline titles, formula scripts, independent positive/negative offsets for both axes, cached placement overlap warnings, automatic/fixed margins, fixed data rectangles, adjacent panels, original-value/log data anchors, rotation, group scaling, and three-format export. Python coverage saves and rerenders native plot source containing both offsets.

The scientific-label example was regenerated as SVG, 180 DPI PNG and PDF; the SVG-derived PNG and rasterized PDF were visually inspected. Its right panel shifts the title down 0.5 mm and multiplier up 0.5 mm. Both data rectangles remain 76 × 52 mm at page positions (30,32) mm and (134,32) mm, with no layout warnings. PDF text and formula sources remain extractable. The other five examples, scientific/annotations/publication/markers/colorbars, retain byte-identical SVG and 180 DPI PNG output against this change's baseline; PDF pixels rendered at 180 DPI match exactly. Performance was not remeasured.

All 44 focused gallery examples and Notebook bridge checks passed. Markdown checking covers 42 pages and 206 inline images; the site build and check passed 42 pages and 258 image references.

14. Named axes, independent breaks, statistics and shared decorations (2026-10-02)#

Implemented named axes/anchors, independent break intervals and physical segments, reversed/symlog/category ticks, bars/steps/areas/histograms/boxes/violins/ECDF, nonuniform heatmaps and gridded contours, per-point styles, five shared color norms, standalone legends/colorbars and inspect --json. Fixed data rectangles and manual panel placement remain intact.

Node 114/114, Python 14/14, with 48 gallery examples. Six default plots retain identical SVG/PNG bytes and PDF raster pixels. Five new examples export all three formats with visual checks. See this run for geometry, numerical and compatibility evidence plus fresh timing/RSS/file sizes, and the reference for APIs.