DOM-shaped Python ports and an optional pywebview app wrapper built on domonic.
Additional ports and experimental libraries not included with the standard domonic installation.
This package is deliberately split:
- libs for compatibility testing domonic.
- an
appwrapper for building desktop UIs from domonic trees.
but they will be here so you can use them in an app if you want by importing off the tags here.
Behaviours that trip up the ports are logged to feed back upstream: DOM ones in docs/domonic-wrinkles.md, and domonic.javascript (the JS runtime shim) ones in docs/javascript-wrinkles.md, driven by the acorn parser port.
To get all the libraries like Mermaid, preact, turndown, marked, validator, qs, dompurify, readability etc...
pip install domonic-libsThe pywebview app wrapper is an optional extra:
pip install "domonic-libs[app]"Hack on the repo and run examples:
python3 -m venv .venv
./.venv/bin/pip install -r requirements.txt # editable install + dev deps
./.venv/bin/python examples/dompurify_demo.py
make testThe app wrapper API is in docs/app.md; the dlx CLI in docs/cli.md. myjs -- a JavaScript interpreter with an ffi bridge to native C -- ships as its own package (pip install myjs), built from src/myjs/ via packaging/myjs/; see docs/myjs.md and docs/publishing.md. Ported modules keep their upstream licenses -- see THIRD_PARTY_LICENSES.md.
Installing the package puts a dlx command on your PATH (also aliased domonic-libs, or run it as python -m domonic_libs). The cleanest way to get just the CLI, isolated, is pipx:
pipx install domonic-libs
# ...or run it once without installing
pipx run --spec domonic-libs dlx --helpEvery text command reads a file argument, or stdin when the argument is - or omitted, and writes to --output / -o or stdout -- so they pipe.
# Mermaid -> SVG (flowchart / sequence / pie / timeline, auto-detected)
dlx mermaid architecture.mmd -o architecture.svg
dlx mermaid architecture.mmd --open # render + open in the browser
dlx mermaid architecture.mmd --stats # counts on stderr
cat flow.mmd | dlx mermaid --html > flow.html
# Markdown <-> HTML
echo "# Hi *there*" | dlx md
dlx html2md article.html --gfm > article.md
# Pull the readable article out of a page
dlx read https://example.com/some/post --md > post.md
dlx read page.html --json # just the metadata
# Sanitise hostile HTML (DOMPurify)
dlx sanitize comment.html --profile html --report
# One-off validator.js checks (exit 0 = true, 1 = false)
dlx validate isEmail ada@example.com
dlx validate isIBAN DE89370400440532013000
dlx validate --list
# Query strings
dlx qs parse "user[name]=ada&tags[]=a&tags[]=b"
dlx qs stringify '{"a": 1, "b": {"c": 2}}'
# Lay out and draw an arbitrary graph with the dagre port
printf 'build -> test\ntest -> deploy\nbuild -> lint\nlint -> deploy\n' \
| dlx dagre --rankdir LR -o pipeline.svgdlx <command> -h lists each command's options; full reference in docs/cli.md.
The domonic repo has tests but also small complex ports like javascript and libraries like dquery and d3 were done to make sure the DOM was behaving as it should.
The more things I can port faithfully from js, the more expectations on the DOM I can correct. And out of that will also pop useful tools.
from domonic_libs.dompurify import sanitize
clean = sanitize('<p onclick="x"><a href="javascript:bad()">bad</a>Hello</p>')
print(clean)A port of DOMPurify -- config options, hooks, namespaces, RETURN_DOM / WHOLE_DOCUMENT, and DOM-clobbering protection. It runs DOMPurify's own test/fixtures/expect.mjs corpus (tests/fixtures/dompurify/expect.json, 223 real-world XSS payloads) at 212/223 exact-match with default config; the remaining eleven are fidelity gaps (safe output, not byte-identical) from domonic's HTML5 parser or DOMPurify's deepest namespace-confusion checks, listed in tests/test_dompurify.py::KNOWN_GAPS.
from domonic_libs.qs import parse, stringify
state = parse("filters[status][]=open&filters[status][]=draft")
print(state)
print(stringify(state, {"arrayFormat": "brackets"}))qs is useful when browser-style URL state needs to round-trip through nested Python dictionaries and arrays.
from domonic_libs import validator
validator.isEmail("ada@example.com") # True
validator.isIBAN("DE89370400440532013000") # True (mod-97 checksum)
validator.isCreditCard("4111111111111111") # True (Luhn)
validator.normalizeEmail("Foo.Bar+x@googlemail.com") # 'foobar@gmail.com'A file-by-file port of validator.js v13 -- ~90 string validators and sanitizers with the upstream camelCase names. Options are passed as a dict or as keyword arguments. It runs the valid / invalid case tables from validator.js's own test suite (tests/fixtures/validator/cases.json, 858 cases). Locale-table-heavy validators (isMobilePhone, isPostalCode, isTaxID, ...) and the full normalizeEmail provider lists are not ported yet.
from domonic_libs.readability import Readability
from domonic_libs.turndown import turndown, TurndownService
article = Readability(html).parse()
markdown = turndown(article["content"])
# TurndownService mirrors turndown.js: options, .use(plugin), .addRule,
# .keep, .remove. Python option names are snake_case (heading_style="atx").
service = TurndownService(heading_style="atx", code_block_style="fenced")
from domonic_libs.turndown.gfm import gfm
service.use(gfm) # tables, strikethrough, task lists (turndown-plugin-gfm port)turndown is a file-for-file port of turndown.js and runs its upstream fixture suite (see tests/test_turndown.py). Together with Readability it is useful for content extraction, reader views, local archives, and LLM-friendly page summaries.
from domonic_libs.marked import marked
marked("# Title\n\nSome **bold** text and a [link](https://x.io).")A file-for-file port of marked v18 (Markdown to HTML) -- the inverse of turndown. It passes marked's own CommonMark 0.31.2 and GFM 0.29 conformance suites in full and 98% of marked's new/ regression specs (tests/test_marked.py), compared with the same html-differ semantics marked uses. Options mirror marked's (gfm, breaks, pedantic, silent, custom renderer); hooks, async, and third-party extensions are not ported. marked's \p{P}/\p{S} regex classes are baked from unicodedata, keeping the port dependency-free.
from domonic.dom import document
from domonic_libs.preact import h, render, Component
from domonic_libs.preact.hooks import useState, useEffect
def Greeting(props):
return h("h1", None, "Hello ", props["name"])
root = document.createElement("div")
render(h(Greeting, {"name": "world"}), root)
str(root) # '<div><h1>Hello world</h1></div>'A file-for-file port of Preact 10.29.8 -- create_element/h, the vnode-diffing reconciler (keyed children with the skew algorithm, Fragments, refs, dangerouslySetInnerHTML), the Component class with the full lifecycle, createContext, and preact/hooks (as domonic_libs.preact.hooks) -- running against domonic's server-side DOM instead of a browser. There is no JSX, so h is the authoring API and lifecycle names keep their upstream camelCase. Rendering is synchronous (Preact's microtask batching is replaced by a flush at the end of each render or event handler); Suspense and preact/compat are not ported. tests/test_preact.py ports a cross-section of Preact's browser suite (render, components, keys, fragments, refs, context, hooks) to unittest.
from domonic_libs.mermaid import render
svg = render("""flowchart TD
A[Parse text] --> B{Diagram type?}
B -->|sequence / pie| C[Direct renderer]
B -->|flowchart| D[dagre layout]
C --> E[Emit SVG]
D --> E
E --> F((domonic tree))""")
# -> '<svg class="mermaid flowchart" viewBox="0 0 ..." ...>...</svg>'A port of Mermaid (mermaid@11.9.0) that turns diagram text into a domonic SVG tree -- no mermaid.js bundle, no headless browser, no Graphviz binary. render(text) auto-detects the type; the CLI is dlx mermaid.
| diagram | status |
|---|---|
| sequence | grammar + SequenceDB + renderer: participants (box and actor glyph), lifelines, the full arrow set, self-message curves, notes, activation bars, and the loop / alt / opt / par / critical / break / rect box machinery (nested boxes grow to enclose their contents). Gaps: box groups, autonumber badges, KaTeX, wrapping |
| pie | slices, percentages, legend, showData -- over domonic's d3.shape (pie/arc) and d3.scale |
| timeline | sections, periods, events, the activity line, wrapped node text |
flowchart / graph |
node shapes, the arrow zoo (labels, lengths, --x / --o), & groups, chains, subgraph; laid out by the bundled dagre port |
The sequence / pie / timeline parsers are checked against mermaid's own spec assertions (tests/test_mermaid.py); the flowchart parser is a pragmatic scanner for the common flow.jison subset (mermaid's is ~630 lines of stateful lexer), tested against hand-written cases. Text is measured through domonic's real getBBox() -- a gap this port surfaced, since implemented in domonic 1.5.0. examples/mermaid_demo.py is a live workbench.
from domonic_libs.dagre import Graph, layout
g = Graph({"compound": True})
g.setGraph({"rankdir": "LR", "nodesep": 40, "ranksep": 60})
g.setDefaultEdgeLabel(lambda *a: {})
for v in "abc":
g.setNode(v, {"width": 80, "height": 30})
g.setEdge("a", "b", {}); g.setEdge("a", "c", {})
layout(g)
g.node("b") # {'x': ..., 'y': ..., 'rank': ..., ...}
g.edge("a", "b")["points"]A faithful file-for-file port of dagre and the slice of graphlib it needs -- directed-graph hierarchical layout, no C or JS dependency. The whole pipeline is ported: greedy-FAS acyclic → Sander nesting graph → network-simplex rank → barycenter + Barth bilayer crossing minimisation → Brandes-Köpf x-coordinates → edge routing. Used by the Mermaid flowchart renderer, and usable on its own for any layered-graph drawing (dlx dagre takes an edge list). Per-cluster rankdir recursion is the one omission; tests/test_dagre.py does structural/invariant checks.
from domonic_libs.acorn import parse
from domonic_libs.acorn.jsx.transform import jsx_to_python
from domonic_libs.acorn.interpret import run_js
parse("const f = x => x * 2", {"ecmaVersion": 2022}).to_dict()
# {'type': 'Program', 'body': [{'type': 'VariableDeclaration', ...}], ...}
jsx_to_python("<div className='box'>{items}</div>")
# "div(items, _class='box')"
doc, log = run_js("const b = document.createElement('button');"
"b.textContent = 'Go'; document.body.appendChild(b);")
str(doc.body) # '<body><button>Go</button></body>'A faithful port of acorn 8.18.0 (tokenizer, the regexp.js grammar validator, the recursive-descent parser) plus the acorn-jsx plugin -- a pure-Python ECMAScript / JSX front end producing an ESTree tree. Its main job is stress-testing domonic.javascript; every wrinkle it surfaced was fixed in domonic 1.6 (docs/javascript-wrinkles.md).
jsx_to_pythonrewrites the JSX markup layer todomonic.htmlfactories (orh(...)withmode="h"), passing JS expressions inside{ ... }through verbatim.interpret.run_jsis a tree-walking evaluator for the practical subset of ECMAScript -- expressions, functions + closures, control flow, objects / arrays,this,new,class(withextends/super/ fields), and the JS coercion rules. It runs against the whole domonic runtime: the global object exposes ~190 constructors auto-collected fromdomonic.javascript/domonic.webapi.*/domonic.dom(URL,Headers,Request/Response,Blob/FileReader,XMLHttpRequest,EventSource,Event/MouseEvent,MutationObserver/ResizeObserver/IntersectionObserver,Range/TreeWalker,DOMRect/DOMMatrix,XPathEvaluator,Path2D,FontFace,Notification,Worker, …), forwards to domonic's realwindowfor the rest (location,navigator,atob,getComputedStyle,setTimeout), and reaches the element surface directly --el.style/getComputedStyle(CSSOM instyle.py),el.classList/el.dataset,el.addEventListener+dispatchEvent.consoleanddocumentare isolated per run;documentalso forwards to a real backingDocument(createComment,createEvent,evaluate, …).document.createElement(...).appendChild(...)builds a real Python object tree. There is a pragmatic event loop --Promise,async/await,setTimeout-- with microtasks ahead of timers, and ES moduleimport/export. Not covered: generators, real prototype chains,Proxy/Symbol/ accessors. Running real DOM scripts this way is a live stress test ofstyle.py,domonic.events,domonic.webapi, and the DOM.
Thrown errors carry a js_line and a js_trace (call stack), and the js_demo workbench surfaces both.
Nothing here is exported from import domonic_libs (it stays lean). examples/acorn_demo.py (JS → ESTree / tokens), examples/jsx_demo.py (JSX → domonic Python, live-rendered), and examples/js_demo.py (JS run against the DOM) are workbenches.
Two scorecards measure how close the JS + DOM layer is to spec, each a curated battery run end to end through the interpreter with a CI gate against a baseline:
python -m domonic_libs.conformance— CSSOM / DOM assertions (el.style/CSSStyleDeclaration,classList,dataset, attribute reflection) modelled on Web Platform Tests → docs/conformance.md (46/49).python -m domonic_libs.js262— a test262-style battery over language expressions / statements (closures, classes, generators,async/await, destructuring, modules) and theArray/String/Object/Number/Math/JSON/Promise/RegExpbuilt-ins → docs/js-compliance.md (100%).
A failing check is an interpreter gap or a domonic gap; the message says which, and docs/domonic-wrinkles.md / docs/javascript-wrinkles.md track the domonic ones.
myjs is a JavaScript interpreter packaged on top of the acorn port and its evaluator. It ships as its own PyPI distribution — pip install myjs (built from src/myjs/ via packaging/myjs/, released after domonic-libs; see docs/publishing.md). myjs with no arguments starts a REPL; myjs script.js executes a file; myjs -e "<code>" evaluates a snippet.
import myjs
myjs.eval("1 + 2 * 3") # -> 7
myjs.run("script.js")
s = myjs.Session() # isolated global scope
s.eval("const x = 21;")
s.eval("x * 2") # -> 42Scripts get the whole domonic DOM (document, window, ~190 constructors), an event loop (Promise, async / await, setTimeout), ES modules (import / export, with bare specifiers resolving to Python modules), a WebSocket client, a set of host bindings -- fs, path, sh, http, an asynchronous fetch, os (Node-flavoured), process, say / notify / open, and py for reaching into the entire Python ecosystem (py.import("numpy")) -- plus an ffi global that calls native C libraries through Python's ctypes -- no node-gyp, no C compiler:
const libc = ffi.loadLibrary("c");
libc.abs.argtypes = [ffi.types.int];
libc.abs.restype = ffi.types.int;
console.log(libc.abs(-42)); // 42
const buf = ffi.createStringBuffer(64);
libc.strcpy(buf, "written into C memory");
console.log(buf.value);ffi.loadLibrary takes a short name ("c", "m"), a path (./libfoo.so), or a macOS framework name; ffi.createStringBuffer, ffi.callback (wrap a JS function as a C function pointer), ffi.cast / ffi.sizeof / ffi.string, and the ffi.types table cover the rest. Thrown errors surface as myjs.JSError with js_name / js_line / js_trace. myjs --gui app.js renders the DOM the script builds in a native window (needs [app]).
Runnable tours: myjs_wow.js (the whole surface), myjs_async.js (event loop, concurrent Promise.all(fetch...)), myjs_realtime.js (WebSocket + setInterval), myjs_gui.js (--gui window), myjs_server.js (a web server in JS whose HTML is built with document.createElement), myjs_ffi.js (native C). Full reference: docs/myjs.md.
Install .[app] or .[examples] before using this part.
from domonic.events import Event
from domonic.html import button, h1, main
from domonic_libs import App, on
app = App("Hello")
def clicked(event):
print(event.type)
@app.route("/")
def index():
return main(h1("Hello"), on(button("Click me"), Event.CLICK, clicked))
app.run()App is a thin bridge for domonic trees: routes return domonic HTML, callbacks receive domonic events, and examples can use OS file dialogs, menus, drag/drop, timers, storage, and transparent windows.
The render loop lives in a host-agnostic BaseApp. App is an alias for DesktopApp, which hosts it in a pywebview window. BrowserApp serves the same application to an ordinary browser over a small stdlib HTTP server -- the same route/handler code, no changes:
from domonic_libs.app import BrowserApp
app = BrowserApp("Hello")
@app.route("/")
def index():
return main(h1("Hello"), on(button("Click me"), Event.CLICK, clicked))
app.run(port=8000) # serves http://127.0.0.1:8000/Only the transport differs (pywebview's js_api vs fetch) and the native host (menus, dialogs, native drag/drop paths are desktop-only). Because the browser has no server-to-client push, evaluate_js / refresh on BrowserApp queue a command that rides back on the next event or timer response.