Python support: transpiled Python on every target, Python servers, Pillow and font-from-data by shai-almog · Pull Request #6128 · codenameone/CodenameOne

shai-almog · 2026-10-29T01:52:51Z

Python support for Codename One: Python sources are transpiled to Java and run on every target, client and server.

What is in it

maven/python-transpiler and maven/python-runtime: the transpiler, the runtime and the bundled stdlib and site packages, with ~125 conformance programs whose expected output is real CPython output.
Build integration in build-engine (Maven goal, Gradle task, project importer, PyPI resolution, C-extension stubs) and the python.virtualThreadStack build hint / @VirtualThreadStack.
Python servers on the backend: FastAPI/Starlette/Flask, WebSockets and streaming responses, BackgroundTasks after the response, SQLAlchemy on SQLite, PostgreSQL and MySQL (CI runs it against both), a GIL that cooperates with virtual threads.
Pillow: on servers over a new backend image API (com.codename1.backend.image: raster, JPEG/PNG, TrueType text); on clients over Codename One's Image/ImageIO.
Core: Font.createTrueTypeFontFromData / isTrueTypeStreamSupported on JavaSE, Android, iOS/Mac, JavaScript, Windows and Linux; Database column types.
Backend: MySQL/PostgreSQL sslmode=verify-ca with a CA file (plus the MySQL spellings).
Fix: the core inflater (io/gzip/InfCodes) rejected valid deflate streams whose last code ended a few bits into the final byte (about half of all gzip files, depending on the header mtime). Fuzzed against java.util.zip with 160k streams fed in chunks.
Developer guide chapters for all of the above.

Verification

python-transpiler/python-runtime tests (145), core-unittests, backend tests; SpotBugs/PMD/Checkstyle clean on the touched modules; cast-semantics, native-signature, C++ linkage, control-character and build-hint gates.
End-to-end server runs on the JVM and native, and SQLAlchemy against PostgreSQL 16 and MySQL 8.4.
Not verified locally: the JavaScript font path in a browser (TrueTypeFontFromDataTest in the JS suite covers it here), the Windows/Linux C font changes (no toolchain locally) and the iOS private-API scan.

🤖 Generated with Claude Code

Python 3 sources transpile to Java 17 (maven/python-transpiler) and run on maven/python-runtime on every Codename One target. Generators, async, match, dataclasses, the builtins and a stdlib wave are covered by a conformance suite diffed against CPython on the JVM and natively through ParparVM. Constructs a translated program cannot run (eval, exec, compile, __import__, frame hooks) are build errors naming file and line. ParparVM: a frameless method returning a value after several try blocks that end on one label ran a single END_TRY and left the other blocks on the thread's try stack, so a later throw longjmp'd into a returned frame. The folded return paths now unwind the stack (FramelessTryReturnIntegrationTest). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

from java.<package> import <Class> resolves at build time against the compile class path; a generated JavaInterop bridge dispatches with instanceof and direct calls (no reflection), converts values both ways, adapts Python callables to functional interfaces and exposes Java collections to len/in/iteration/indexing. cn1:import-python-project and cn1:transpile-python (Maven), the transpilePython task (Gradle), and the archetype's python profile build src/main/python. runMain returns the exit status, honouring SystemExit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…vice library from java.java.* resolves against the class path's own class library (java-runtime, JavaAPI) when it carries one, so an API the device lacks is a build error rather than a bridge that fails to link. python-project-test.sh imports scripts/python-samples/hello-ui (a Codename One UI written in Python), builds it with Maven and Gradle and stages every target; python-project.yml runs it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

PyHttp is the client the packages are written over: ConnectionRequest in an application, installable like PyFileSystem. requests (verbs, Session, auth, params/json/form bodies, cookies, exceptions) and httpx (verbs, Client, AsyncClient, URL, Headers, HTTPStatusError) are our own Python over it; http_requests and http_httpx match the real packages' output against the same test server. BaseException.__new__ ignores keyword arguments, as CPython's does, so an exception subclass whose __init__ takes keywords can be raised. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…_type_hints Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…lice.indices The HTTP client is installed by the generated registry only when the program bundles requests or httpx, so other programs do not link the network stack. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ule __dict__ Fixes found while porting attrs and pydantic: an annotation built by a call (Annotated[int, Field(gt=0)], conint(ge=2)) was stored as None; a class __delattr__ was bypassed by del; classes had no __hash__ attribute; and type(name, bases, ns) gave __module__ 'builtins'. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

_cn1server.serve(dispatch, host, port) bridges a Python dispatch callable to the backend HttpServer when the backend is on the transpile class path; scripts/python-backend/build.sh builds such an app as a native binary. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Clean-room site packages flask, werkzeug.exceptions, fastapi, starlette (responses, requests, testclient, status) and uvicorn. Every app exposes dispatch(method, path, query, headers, body) -> (status, headers, body); app.run / uvicorn.run hand it to _cn1server.serve. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ions inspect.isgeneratorfunction/iscoroutinefunction/isasyncgenfunction now answer from __code__ as CPython's do; FastAPI's yield dependencies use it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…pter Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…indings Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… helper returns CommonWorkloads ported to Python (scripts/python-bench/common) with a Java hot-spot variant, and a comparison across CPython, PyPy, GraalPy and Codename One with peak RSS per workload. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…mance chapter The developer guide's Python chapter compares CPython, PyPy, GraalPy and Codename One on startup, kernels, Flask and the CommonWorkloads suite, with peak memory in every table, two charts, and a section on moving a measured hot spot into Java. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

IntRanges runs interval analysis over each typed function: loop counters are bounded by their range, accumulators by trip count times their per-pass growth, and masked values by the mask. A local whose every value fits in 64 bits becomes a Java long, and an int expression whose result fits is computed with wrapping long arithmetic, which equals the exact result however large the intermediates are. Expression helpers such as i32/i64 are inlined, module int constants are folded, and >> of a value masked into [0, 2^64) becomes >>>. Builtins without a typed lowering (print, sorted, ...) are now called generically from typed code instead of leaving the whole function untyped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…nd //, lowered loop tests Locals that only ever hold a freshly constructed object whose __init__ copies its arguments into fields, and are only read through fields or getters, become one local per field. while tests lower their arithmetic and compares like any statement. for c in str walks code points, with ord(c) the code point itself. l.extend([v] * n) fills in place. Lone surrogates index and repr as CPython does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ped builtins over BOT, dict clear/len intrinsics Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… int results held bare, taken in the slow path Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…t stores per instance, try-free dict key hashing Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…-top arithmetic over impure calls Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ct clear; direct dict len Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… box the list Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The semantic analyzer's merge compared values with SimdAllocaValue.equals, which refuses ASM's own BasicValue constants, so a long pushed by LCONST_1 meeting the checker's long at a join fell to BasicInterpreter.merge and became a one-slot UNINITIALIZED_VALUE; the next DUP2 failed "Illegal use of DUP2" on generated Python that ASM's plain analyzer accepts. Merge now compares types. The typed float path for ** emitted Math.pow, which CLDC lacks; more stdlib functions are typed now, and the demos' compliance check caught 30 such calls. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…pinning the GC JFR on the JVM build of the Flask sample (escape analysis off) put the request path at 17.7 KB allocated per request; it is now 12.7 KB on the JVM, and the native binary's peak RSS under ab -c 8 fell from 765-816 MB to 219-220 MB with no GC tuning. - Flask shim: Headers keeps folded names beside the pairs; routing skips a rule whose literal prefix does not match before running its regex, and makes the 405 set only when needed; prebuilt JSON encoders; lazy flask.g; no copies of the header list or the blueprint chain per request. - json: unescaped strings quote without a char list; sort_keys sorts pairs without a lambda; the circular check is an identity stack, not an id() dict. - runtime: regex scratch (text, registers, stacks) reused per program via an atomic hand-off; str.startswith/endswith intrinsics and in-place affix; str < str compares without wrapping; str iteration and indexing share Latin-1 one-char strings; dict literals presized, first growth 4. - backend: a connection's deferred-JSON sink and target memo table are made on first need, so a one-request connection allocates neither. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…66% survival The collector sat at its peak footprint for 30 seconds after load stopped, because only allocation or the long idle starts a cycle. The first quiet pass after a busy one now idles 1s, so a quiet cycle runs the major sweep, page trim and malloc relief. The backend build grows the trigger only at 66% survival instead of 25%, so the heap settles near 1.5x the live set rather than 4x. Flask sample, 3 interleaved runs: footprint after load 137-191MB -> 75-98MB, 5s later 101-139MB -> 68-70MB, at 27.7-29.2k -> 25.7-25.9k req/s. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…duty cycle A fixed 66% cost the Flask server 9% of its throughput. The threshold now moves from CN1_BIBOP_TIGHT_SURVIVAL_PERCENT (backend: 66) when the collector is idle for 90% of wall time or more, to the VM's 25% when it is busy 40% or more. Default builds set both equal and are unchanged. Full speed: 29.1-29.7k req/s vs 29.6-30.1k default and 26.0-26.9k fixed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…cales with a small budget Under a per-process budget the survival threshold is the tight one regardless of the collector's duty cycle. The 64MB admission margin becomes a quarter of the budget below 256MB: under 128MB the Flask server went from 478 to 16.3k req/s at a 75MB footprint. Budgets of 256MB or more are admitted as before. compare.py runs every server for the same 30s, reports physical footprint after load, 5s later and at peak, and runs Codename One both unconstrained and under a --budget-mb budget (default 128). render_charts.py adds the throughput-against- memory chart. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…nd under a budget The comparison now runs every Codename One binary twice: as shipped, where the adaptive collector spends spare memory for throughput, and under a 128 MB per-process budget, where it keeps the heap tight. Memory is physical footprint for every runtime; Flask gets 30 s of load and a settle read. Adds the throughput-vs-footprint chart. Also routes the transpiler's float pow through Py.powd (Math.pow): MathUtil.pow had made nbody 4x slower. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…on Windows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Per-thread pacing claims lived in __thread storage, and a virtual thread waiting for budget can resume on another host, so claims leaked until no allocation was admitted and a budgeted server crawled. The claim is now one global value stamped with the GC epoch, expiring after two cycles. Retired virtual-thread stacks are released under a budget. The comparison runs the 128 MB budget only: Flask under 64 MB cannot hold the cap, since its unpaced virtual-thread stacks and malloc swing past it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts: # docs/developer-guide/languagetool-accept.txt

shai-almog · 2026-11-02T08:58:19Z

Compared 193 screenshots: 193 matched.
✅ Native iOS Metal screenshot tests passed.

Benchmark Results

VM Translation Time: 0 seconds
Compilation Time: 2400 seconds

Build and Run Timing

Metric	Duration
Simulator Boot	55000 ms
Simulator Boot (Run)	1000 ms
App Install	17000 ms
App Launch	2000 ms
Test Execution	463000 ms

Detailed Performance Metrics

Metric	Duration
SIMD kernel backend	SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300)	java 337ms / native 4ms = 84.2x speedup
SIMD float-mul (64K x300)	java 300ms / native 3ms = 100.0x speedup
SIMD kernel correctness	PASS (native result == scalar reference)
Base64 payload size	8192 bytes
Base64 benchmark iterations	6000
Base64 SIMD byte path	active (NEON-accelerated)
Base64 CN1 encode	97.000 ms
Base64 CN1 decode	68.000 ms
Base64 native encode	1321.000 ms
Base64 encode ratio (CN1/native)	0.073x (92.7% faster)
Base64 native decode	1141.000 ms
Base64 decode ratio (CN1/native)	0.060x (94.0% faster)
Base64 SIMD encode	61.000 ms
Base64 encode ratio (SIMD/CN1)	0.629x (37.1% faster)
Base64 SIMD decode	60.000 ms
Base64 decode ratio (SIMD/CN1)	0.882x (11.8% faster)
Base64 encode ratio (SIMD/native)	0.046x (95.4% faster)
Base64 decode ratio (SIMD/native)	0.053x (94.7% faster)
Image encode benchmark iterations	100
Image createMask (SIMD off)	7.000 ms
Image createMask (SIMD on)	3.000 ms
Image createMask ratio (SIMD on/off)	0.429x (57.1% faster)
Image applyMask (SIMD off)	367.000 ms
Image applyMask (SIMD on)	38.000 ms
Image applyMask ratio (SIMD on/off)	0.104x (89.6% faster)
Image modifyAlpha (SIMD off)	312.000 ms
Image modifyAlpha (SIMD on)	57.000 ms
Image modifyAlpha ratio (SIMD on/off)	0.183x (81.7% faster)
Image modifyAlpha removeColor (SIMD off)	55.000 ms
Image modifyAlpha removeColor (SIMD on)	46.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off)	0.836x (16.4% faster)

…docs: Vale test_virtual_thread.c links cn1_virtual_thread.c without cn1_globals.m, so the new cn1ProcessHasMemoryBudget() reference failed the link. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts: # .github/workflows/pr.yml

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

viewSafeAreaInsetsDidChange calls it on Mac Catalyst and tvOS too, but it was defined only in the iPhone layout branch, so the Metal Catalyst build failed with -Wobjc-method-access. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts: # Ports/iOSPort/nativeSources/CodenameOne_GLViewController.m # scripts/developer-guide/check-guide-api-names.py

…e fairly CPython and PyPy served the same ~690 requests a second under app.run() although PyPy runs Flask twice as fast: Werkzeug's development server drains the socket for 10 ms after every response, and ab counts a request done when the socket closes. The guide keeps that chart, marks it as suspicious, tells how the cause was found, and adds the fair comparison: compare.py now also serves each interpreter behind gunicorn with 8 workers, footprint summed over every process. flask_cost.py times Flask alone, without a server. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ide's accept list Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The base row is a single run; the branch's VM runtime measures identical to master's on the self-hosted translator (interleaved A/B, reason in the file). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ture The multi-window default scenario was captured before its first paint (skin and form solid black) because the menu bar alone passes the flat-image check; single-window mode already waits that out, multi mode had no equivalent. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The iOS accessibility thread contract (on IOSNative.m changes) and the Xcode leg of the Gradle/Maven parity check queued on hosted macos-15 for every matching pull request, and were dropped unrun when that pool was short. Both now run nightly, plus on a pull request touching their own tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

shai-almog · 2026-11-04T23:08:39Z

Compared 193 screenshots: 193 matched.
✅ Native iOS Metal screenshot tests passed.

Benchmark Results

VM Translation Time: 0 seconds
Compilation Time: 2675 seconds

Build and Run Timing

Metric	Duration
Simulator Boot	97000 ms
Simulator Boot (Run)	1000 ms
App Install	14000 ms
App Launch	6000 ms
Test Execution	449000 ms

Detailed Performance Metrics

Metric	Duration
SIMD kernel backend	SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300)	java 66ms / native 3ms = 22.0x speedup
SIMD float-mul (64K x300)	java 65ms / native 4ms = 16.2x speedup
SIMD kernel correctness	PASS (native result == scalar reference)
Base64 payload size	8192 bytes
Base64 benchmark iterations	6000
Base64 SIMD byte path	active (NEON-accelerated)
Base64 CN1 encode	521.000 ms
Base64 CN1 decode	158.000 ms
Base64 native encode	859.000 ms
Base64 encode ratio (CN1/native)	0.607x (39.3% faster)
Base64 native decode	941.000 ms
Base64 decode ratio (CN1/native)	0.168x (83.2% faster)
Base64 SIMD encode	244.000 ms
Base64 encode ratio (SIMD/CN1)	0.468x (53.2% faster)
Base64 SIMD decode	111.000 ms
Base64 decode ratio (SIMD/CN1)	0.703x (29.7% faster)
Base64 encode ratio (SIMD/native)	0.284x (71.6% faster)
Base64 decode ratio (SIMD/native)	0.118x (88.2% faster)
Image encode benchmark iterations	100
Image createMask (SIMD off)	7.000 ms
Image createMask (SIMD on)	2.000 ms
Image createMask ratio (SIMD on/off)	0.286x (71.4% faster)
Image applyMask (SIMD off)	237.000 ms
Image applyMask (SIMD on)	57.000 ms
Image applyMask ratio (SIMD on/off)	0.241x (75.9% faster)
Image modifyAlpha (SIMD off)	50.000 ms
Image modifyAlpha (SIMD on)	54.000 ms
Image modifyAlpha ratio (SIMD on/off)	1.080x (8.0% slower)
Image modifyAlpha removeColor (SIMD off)	138.000 ms
Image modifyAlpha removeColor (SIMD on)	254.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off)	1.841x (84.1% slower)

…nutes The x64 suite allocated a fixed 12G swap file and died 'No space left on device' two minutes into the C build on a runner with little free disk; the swap now leaves 40G for the build. build-ios-tv measured 15-22 min of build plus 28-33 min of tests and was cancelled at its 60 minute limit mid-run on this and other branches. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fix the three nightly failures in #6160 dash: GameSocketTest slept out the countdown and then sent the host's jump. Two runners who never jump are both out 6.4 s into Neon Causeway, the race ends, and a later jump is dropped, so the jump had about 6 s to land; a stalled CI container missed that and the guest waited for the server's 300 s idle close (reproduced by delaying the jump). The jump is now sent during the countdown, which the server stamps for tick INPUT_DELAY, a read timeout fails a missing frame in 30 s, and the race ending first is reported by name. OnlineRaceTest now holds the stamping of a mid-race jump on a driven clock, which the socket test no longer exercises. RaceRoom.intent's doc no longer says a countdown input is dropped. The Initializr template is regenerated. flutter-bench: the JavaScript code size moved +17.5% with #6163, whose emitter reads instance fields as o.name instead of through the o[_qN] alias table (valueEscape 20% faster). The gallery's wire bytes in the same run moved +0.03%. Rebaselined with that reason. nightly-watchdog: ios-accessibility-thread joined the nightly in #6128, merged after that night's slot, and was reported as a schedule that stopped firing. A suite with no scheduled run at all whose file reached master inside the window is now listed as awaiting its first run, which keeps the issue open; one that ran before and went quiet is still a problem. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Fix the Flutter gallery transpile #6154 broke, and gate it on every transpiler PR #6154 merged a transpiler under which the Flutter gallery (the Flutter benchmark's application) no longer compiled, and nothing on its PR ran it: flutter-bench.yml runs on a PR only when its own tree changes. Every benchmark leg of every PR since then failed in javac. Fixed, each where it went wrong: - a library-private class's static tear-off (`_DialogDemoState::_route`) was taken for a bound one and handed the class name as a receiver; - a function-type witness nested in another reused its lambda parameter; - an explicit `<num>` on a call whose parameter is `Comparable<T>` was written as `<Number>`, which no Long or Double satisfies; - named arguments did not bind a generic function's type parameters, so `ease(begin: radius, end: 0)` emitted `0` as a long; - List<E>.generate's generator had no E-typed return context; - `double + (c ? x : 0)` took the Number arms of a num slot; - the restorablePush stubs declared their route builder Object, and ScrollNotification's depth getter did not follow its siblings' naming. flutter-gallery-compile.yml runs the benchmark's own path up to javac (prepare.sh, then the generated project's compile) on every change to the transpiler, the runtimes or the Flutter build steps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rebaseline the JavaScript gallery's sizes for #6154's code generation The gallery compiles again with this PR, so this is its first measurement under #6154's transpiler: code and install grow 1,208,552 bytes while wire_bytes moves +0.22%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Stop the Dart runtime calling String.split, which ParparVM lacks #6154's FunctionType and TypeWitness split type-variable lists with String.split, which is in neither vm/JavaAPI nor Ports/CLDC11: the JVM legs pass and the native Linux and Windows translations fail to compile an undeclared java_lang_String_split. DartUri and ContentType had the same call. All four now split on a literal by hand. Also rebaselines the Android gallery's dex size for #6154's code generation; the APK itself moved +0.21%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rebaseline the Linux and Windows gallery sizes for #6154's code generation Both native galleries build again now that the Dart runtime no longer calls String.split; code grows about 11% on each while the gallery's wire size moves under 1%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Give the Flutter benchmark's Maven build a 4 GB heap The gallery now transpiles and compiles inside Maven's JVM under #6154's larger output, and the macOS runner's default quarter of 7 GB ran out in jar packaging on the iOS leg. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rebaseline the iOS and macOS gallery sizes for #6154's code generation Both now build with the 4 GB heap; code grows about 11% on each while wire size moves under 1.5%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Dash e2e: one bot steers a whole stretch, never a loop of short windows CityRunTest, PartyTest, PromoVideoTest and driveToBusy called driveFor in a loop of 100-200 ms windows. Each call removed the bot's hook when it returned, so the game ticked unsteered between calls, and on a slow CI runner the city run crashed in one of those gaps before its first landmark ("still running at the landmark"). driveUntil keeps one bot on the controls until the condition holds, the run ends or time is up. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Emit only the closure-signature members the generated code calls #6154 gave the Flutter gallery 340 new classes, 320 of them in DartFnSigs: a bridge class and a static factory for every signed SAM whether or not a call site used it (105 bridges, 18 used), and an interface per signature even where one SAM was its only carrier. Each is a class in the native executable, which put Linux's code behind Flutter's. Only the members a generated file names are emitted now, and a signature with a single carrier is written into that SAM. Subclass const keys share ConstCanon.join instead of inlining the concatenation into every keyed class. Gallery: 904 -> 715 generated classes, 3.70 -> 3.45 MB of class files. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Cut the native code #6154's transpiler output added to the gallery Measured on the gallery's Linux executable, built in a container with the same toolchain on both sides: 15.30 MB before #6154, 17.00 MB after it, 16.40 MB with this change. - A statically typed num.clamp went through Dyn.call, which made every builtin's dynamic dispatch reachable (178 KB for four clamps). It calls Dyn.numClamp directly, and invoking a callable object takes a path that reaches only transpiled objects (Dyn.objectCall), not List/Map/String. - Each class of a library with renamed private names registered the whole library's spellings from a static initializer of its own. They are registered once, from FlutterRegistry.invokeMain, one packed literal per library; the behavior harness enters through invokeMain as an app does. - An enum builds its Dart-name table only when a value was renamed for Java. - $constKey() is kept only for classes constructed as constants and their superclasses (64 of 278 in the gallery). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Lower lambdas to shared carriers instead of a class per lambda A lambda becomes an instance of a carrier -- one class per functional interface and capture shape -- holding a function address and its captures. Brought over from the python-hero-ports work; on the Flutter gallery it cuts 268 KB of native code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Let a lambda factory be frameless The frameless path rejected any custom instruction, and FunctionAddress -- a constant push -- is the one every lambda factory has, so all 2,080 of the Flutter gallery's factories got a full frame. -103 KB. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Allocate out of line at NEW sites, keeping only the init guard inline CN1_FAST_NEW expanded the always_inline bump allocator at all 9,951 NEW sites of the Flutter gallery: 894 KB of a 16.0 MB executable. cn1BibopFastNew is the same path behind one direct call. Interleaved vm/benchmarks A/B against -DCN1_INLINE_ALLOC_SITES: geomean 1.004, objectAllocation 1.009, checksums identical. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rebaseline the Windows arm64 hello perf row for the lambda carriers Lowering lambdas to shared carriers made translating HelloCodenameOne cheaper on both runtimes; ParparVM got faster, JDK 25 faster still, so the ratio moved without a VM regression. Measurements are in the row's reason. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Count a referenced SVG's syndication PNG as used in the guide image check Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rebaseline the Linux gallery's cold-start ratio for #6154's code generation Every run since the gallery compiles again measured 0.76-0.85 (still faster than Flutter); the commits here do not move start-up. Evidence in the row. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Make the game e2e waits and the site search test immune to runner load The Dash and Skirmish online tests asserted wall-clock totals (360 ticks in 30 s, 100 server ticks in 30 s), so a runner shared with a dozen other jobs failed them while the game was still ticking. They now fail only when the ticks stop for 10 s, which is the stall they were written to catch. The site search test ran against wrangler pages dev, whose workerd exits when Chromium resets a connection on navigation; settling first (#6092) did not stop it. Search is static files only, so it now runs against a plain file server; the redirect checks keep the Pages runtime. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

shai-almog and others added 30 commits October 27, 2026 17:01

Python: site package root beside the stdlib

f53f018

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: runtime generic aliases and unions, string annotations in get…

abbfa76

…_type_hints Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: PyYAML package

23bdc60

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: attrs and dateutil packages

f222550

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: pydantic package

0eed8b7

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: typing_extensions, six, tomllib

66b376d

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: tabulate

a478550

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: idna and packaging site packages

6128984

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: co_flags marks generator, coroutine and async generator funct…

9c6462d

…ions inspect.isgeneratorfunction/iscoroutinefunction/isasyncgenfunction now answer from __code__ as CPython's do; FastAPI's yield dependencies use it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: Flask sample, CPython/PyPy comparison and developer guide cha…

04f235d

…pter Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: runtime and transpiler at zero SpotBugs, PMD and Checkstyle f…

1e33bef

…indings Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: drop fully qualified Set in Infer (PMD)

448ee65

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: branch/loop-test int range refinement, threshold widening, ty…

d0611f5

…ped builtins over BOT, dict clear/len intrinsics Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: long-parameter function versions from call-site ranges; typed…

537976b

… int results held bare, taken in the slow path Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: one-new construction for plain __init__, no default shape/slo…

2aa2d04

…t stores per instance, try-free dict key hashing Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: str + chains into one presized StringBuilder; hoist statement…

89021bb

…-top arithmetic over impure calls Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: int-keyed dict get/set without boxing the lookup key; bulk di…

292091f

…ct clear; direct dict len Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Python: int lists hold 32-bit items, as Java's int[] does; wider ints…

8b2f406

… box the list Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

shai-almog and others added 11 commits October 31, 2026 10:09

Python docs: re-measure the Flask row after the allocation cuts

98bf879

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

VM: the collector's duty clock reads cn1MonotonicNanos, which exists …

ba2ec51

…on Windows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge master into python-support

799e634

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/master' into python-support

5808125

# Conflicts: # docs/developer-guide/languagetool-accept.txt

shai-almog and others added 12 commits November 2, 2026 10:06

VM test: stub the budget query the virtual-thread runtime now calls; …

0c6622d

…docs: Vale test_virtual_thread.c links cn1_virtual_thread.c without cn1_globals.m, so the new cn1ProcessHasMemoryBudget() reference failed the link. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/master' into python-support

333d34d

# Conflicts: # .github/workflows/pr.yml

Docs: LanguageTool wording in the Python guide's collector trade-off

5ced6dd

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/master' into python-support

d9484e8

# Conflicts: # Ports/iOSPort/nativeSources/CodenameOne_GLViewController.m # scripts/developer-guide/check-guide-api-names.py

Merge remote-tracking branch 'origin/master' into python-support

bf64c60

Copyright header on SpinnerNodeTest (from master); gunicorn in the gu…

c95160d

…ide's accept list Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Perf baseline: Windows x64 model-106 translator row for this PR

a91159b

The base row is a single run; the branch's VM runtime measures identical to master's on the self-hosted translator (interleaved A/B, reason in the file). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/master' into python-support

d3695e6

shai-almog merged commit 5f96a6b into master Nov 5, 2026

shai-almog deleted the python-support branch November 5, 2026 06:22

shai-almog mentioned this pull request Nov 5, 2026

Fix the three nightly failures in #6160 #6172

Merged

Navigation Menu

Repository navigation

Python support: transpiled Python on every target, Python servers, Pillow and font-from-data - #6128

Python support: transpiled Python on every target, Python servers, Pillow and font-from-data#6128
shai-almog merged 146 commits into
masterfrom
python-support

shai-almog commented Oct 29, 2026

Uh oh!

shai-almog commented Nov 2, 2026 •

edited

Loading

Uh oh!

shai-almog commented Nov 4, 2026 •

edited

Loading

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

shai-almog commented Oct 29, 2026

What is in it

Verification

Uh oh!

shai-almog commented Nov 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Build and Run Timing

Detailed Performance Metrics

Uh oh!

shai-almog commented Nov 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Build and Run Timing

Detailed Performance Metrics

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

shai-almog commented Nov 2, 2026 •

edited

Loading

shai-almog commented Nov 4, 2026 •

edited

Loading