Build: CPython 3.15.0 final (python-build-standalone 20261013)

[b-re-w]

Closes #196

1. Defaults. gradle.properties: pythonVersion 3.14.7 → 3.15.0, pythonBuildStandaloneRelease 20260807 → 20261013. The fallbacks in python-multiplatform/build.gradle.kts (pbsRelease) and python-multiplatform-gradle-plugin/build.gradle.kts (pythonVersion, pythonBuildStandaloneRelease) follow.

2. Checksums. python-checksums.properties adds 11 entries and removes the 10 3.15.0rc1 entries; 3.14.7 stays. Every hash was computed from the downloaded archive:

• PBS desktop, 8 archives (macOS aarch64/x86_64, Linux x86_64, Windows x86_64; GIL and freethreaded): all match the release's SHA256SUMS.
• python.org, 3 archives (android aarch64/x86_64, iOS XCframework): all verify against their .sigstore bundles with the identity the build pins (hugo@python.org, https://github.com/login/oauth), using uvx sigstore (nothing linked into ~/.local/bin).

3. rc-era branches, checked by reading, unchanged. For a final release pythonOrgReleaseDir yields 3.15.0, the directory the archives come from; iosFromPythonOrg is true for 3.15. The 3.14 GIL path is untouched. The build itself then downloaded the 3.15.0 archives and passed its own checksum verification.

Vendored headers. VendoredHeaderVersionTest failed once the default was 3.15.0, which is its job. src/nativeInterop/cinterop/include is refreshed from the build's verified 3.15.0 extraction, as 8e40577 did for 3.14.7: 17 added, 8 removed (cpython/lock.h became pylock.h), 138 changed. EmbedApiSurfaceTest and JniCallConventionClassificationTest pass on it, so the C API surface did not move.

5. Verification (local, macOS arm64, JDK 21, each its own invocation, results counted from XML after --rerun):

Task	Result
:python-multiplatform:compileKotlinAndroidNativeArm64	EXIT 0
:python-multiplatform:compileKotlinIosSimulatorArm64	EXIT 0
:python-multiplatform:desktopTest	608 tests, 0 failures, 0 errors, 1 skipped (first run 1 failure: the header check, fixed above)
:python-multiplatform-ksp:fixtures:app:desktopTest	79 tests, 0 failures, 0 errors, 4 skipped
:python-multiplatform-compose:fixtures:compose:desktopTest	123 tests, 0 failures, 0 errors, 0 skipped
:python-multiplatform-gradle-plugin-fixtures:artifact:desktopTest	37 tests, 0 failures, 0 errors, 0 skipped

4. WASM (research only, no change). A PEP 783 ABI for 3.15 exists only as a draft:

• Pyodide's 315.0.0a1 and 315.0.0a2 prereleases (2026-08-04, branch abi_2026_5) define PYODIDE_ABI_VERSION 2026_5, i.e. pyemscripten_2026_5, on CPython 3.15.0rc1 and Emscripten 6.0.5.
• CPYTHON_ABI_FLAGS is only d: it is a GIL build without pthreads. It doesn't match the free-threaded 3.15t + pthreads WASM decision either.
• No final 315.0.0 yet; Pyodide's main is still 3.14.2 / 2026_0 (Pyodide RFC #6084: one line per Python minor, "315.0.0 this year").
• I found no pyemscripten_2026_5 wheels on PyPI for the packages sampled (pydantic-core has 5 for 2026_0; numpy and markupsafe have none for either).
• Conclusion: our wasm stays on 3.14.2 / pyemscripten_2026_0 until our own 3.15t pthreads build lands (the user's 2026-10-04 decision). Its ABI won't be a Pyodide one either way.