perf: optimize Python and Rust serializer hot paths

.github/workflows/build-rust-wheels.yml

@@ -148,7 +148,8 @@ jobs:
   publish:
     name: Publish to PyPI
     runs-on: ubuntu-latest
-    needs: [linux, windows, macos, sdist]
+    # Never publish artifacts that have not passed the built-wheel matrix.
+    needs: [linux, windows, macos, sdist, test]
     if: startsWith(github.ref, 'refs/tags/rust-v') || (github.event_name == 'workflow_dispatch' && github.event.inputs.publish == 'true')
     environment:
       name: pypi
@@ -180,13 +181,17 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-latest, windows-latest, macos-latest]
-        python-version: ['3.10', '3.11', '3.12', '3.13']
+        python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
+        include:
+          - os: ubuntu-latest
+            python-version: '3.15.0-beta.3'
     steps:
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
 
       - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
         with:
           python-version: ${{ matrix.python-version }}
+          allow-prereleases: true
 
       - name: Download wheels
         uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
@@ -197,15 +202,16 @@ jobs:
 
       - name: Install wheel
         run: |
-          pip install --find-links dist json2xml_rs
+          pip install --no-index --only-binary=:all: --find-links dist json2xml_rs
           pip install pytest
 
       - name: Test import
         run: |
+          python -c "import importlib.metadata as m, os; ref = os.environ['GITHUB_REF_NAME']; expected = ref.removeprefix('rust-v'); actual = m.version('json2xml-rs'); assert not ref.startswith('rust-v') or actual == expected, (actual, expected)"
           python -c "from json2xml_rs import dicttoxml; print('Import successful!')"
           python -c "from json2xml_rs import dicttoxml; result = dicttoxml({'test': 'value'}); print(result.decode())"
 
       - name: Run tests
         run: |
           pip install -e .
-          pytest tests/test_rust_dicttoxml.py -v
+          pytest -o addopts='' tests/test_rust_dicttoxml.py -v

.github/workflows/publish-to-live-pypi.yml

@@ -11,6 +11,8 @@ permissions:
 jobs:
   build-n-publish:
     name: Build and publish Python 🐍 distributions 📦 to pypi
+    # Rust releases use rust-v* tags and have their own wheel workflow.
+    if: startsWith(github.event.release.tag_name, 'v')
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
@@ -35,7 +37,6 @@ jobs:
         .
 
     - name: Publish distribution 📦 to PyPI
-      if: startsWith(github.ref, 'refs/tags')
       uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b # v1.14.0
       with:
         user: __token__

.github/workflows/pythonpackage.yml

@@ -74,7 +74,7 @@ jobs:
 
     - name: Run tests
       run: |
-        pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term -xvs tests -n auto
+        pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term --cov-fail-under=100 -xvs tests -n auto
       env:
         PYTHONPATH: ${{ github.workspace }}
 

.github/workflows/rust-ci.yml

@@ -62,14 +62,18 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
-        python-version: ['3.10', '3.11', '3.12', '3.13']
+        python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
+        include:
+          - os: ubuntu-latest
+            python-version: '3.15.0-beta.3'
     steps:
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
 
       - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
         with:
           python-version: ${{ matrix.python-version }}
+          allow-prereleases: true
 
       - name: Install Rust
         uses: actions-rust-lang/setup-rust-toolchain@166cdcfd11aee3cb47222f9ddb555ce30ddb9659 # v1
@@ -99,10 +103,10 @@ jobs:
           python -c "from json2xml.dicttoxml_fast import get_backend; print(f'Backend: {get_backend()}')"
 
       - name: Run Rust-specific tests
-        run: pytest tests/test_rust_dicttoxml.py -v
+        run: pytest -o addopts='' tests/test_rust_dicttoxml.py -v
 
       - name: Run full test suite
-        run: pytest tests/ -v --ignore=tests/test_cli.py
+        run: pytest -o addopts='' tests/ -v --ignore=tests/test_cli.py
 
   benchmark:
     name: Performance Benchmark

AGENT.md

@@ -2,7 +2,7 @@
 
 ## Build/Test Commands
 - Test: `pytest -vv` (all tests) or `pytest tests/test_<name>.py -vv` (single test file)
-- Test with coverage: `pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term -xvs`
+- Test with coverage: `pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term --cov-fail-under=100 -xvs`
 - Lint: `ruff check json2xml tests`
 - Type check: `uvx ty check json2xml tests`
 - Test all Python versions: `tox`

Makefile

@@ -60,7 +60,7 @@ typecheck: ## check types with ty
 	uvx ty check json2xml tests
 
 test: ## run tests quickly with the default Python
-	pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term -xvs tests
+	pytest --cov=json2xml --cov-report=xml:coverage/reports/coverage.xml --cov-report=term --cov-fail-under=100 -xvs tests
 
 test-simple: ## run tests without coverage
 	pytest -vv tests

RELEASE_NOTES.md

@@ -1,3 +1,34 @@
+# json2xml_rs 0.4.2
+
+Released 2026-07-15.
+
+## Highlights
+
+- Replaced scalar XML escape-byte scanning with `memchr` word/SIMD-optimized searches for the five XML escape characters.
+- Preserved the bounded 16 KiB streaming writer after a 4–128 KiB capacity sweep found no benefit from increasing it.
+- Kept the serialized output byte-for-byte identical at 4,093,244 bytes for the release benchmark payload.
+
+## Performance
+
+The paired CPython 3.15.0b3 release benchmark serialized the deterministic 5,000-record payload in 21 rounds of 50 conversions.
+
+| Metric | Before | After | Change |
+| --- | ---: | ---: | ---: |
+| Median conversion | 6.007 ms | 5.632 ms | 6.23% lower |
+| Mean conversion | 6.013 ms | 5.643 ms | 6.14% lower |
+| Escape scanner exclusive samples | 14.31% | 7.97% | 44.3% lower share |
+
+The committed [before](docs/flamegraphs/rust-before.svg) and [after](docs/flamegraphs/rust-after.svg) flamegraphs show the reduced escape-scanner cost. Rejected tag-building and dispatch experiments regressed the same workload by 6–38%, so this release keeps the optimization deliberately narrow.
+
+## Package Version
+
+- Rust accelerator: `json2xml-rs==0.4.2`
+
+## Verification
+
+The release is gated on Rust formatting and Clippy checks, Rust unit tests, the full Python suite, and built-wheel tests for Linux, macOS, and Windows before PyPI publication.
+
+
 # json2xml 6.4.0 and json2xml_rs 0.4.1
 
 Released 2026-07-13.

docs/flamegraphs/README.md

@@ -0,0 +1,44 @@
+# Serializer flamegraphs
+
+## Rust accelerator
+
+These profiles compare the Rust serializer before and after vectorized XML escape-byte scanning on CPython 3.15.0b3.
+
+The workload serializes a deterministic 5,000-record nested payload with type attributes enabled. Samply captured symbolized native stacks for eight seconds, and Inferno rendered the SVG flamegraphs. A separate paired release benchmark used 21 rounds of 50 conversions.
+
+| Metric | Before | After | Change |
+| --- | ---: | ---: | ---: |
+| Median conversion | 6.007 ms | 5.632 ms | 6.23% lower |
+| Mean conversion | 6.013 ms | 5.643 ms | 6.14% lower |
+| Escape scanner exclusive samples | 14.31% | 7.97% | 44.3% lower share |
+
+### Rust before
+
+[![Rust serializer flamegraph before optimization](rust-before.svg)](rust-before.svg)
+
+### Rust after
+
+[![Rust serializer flamegraph after optimization](rust-after.svg)](rust-after.svg)
+
+### Output buffer capacity
+
+A post-optimization sweep tested 4, 8, 16, 32, 64, and 128 KiB capacities. The useful range plateaued at 16–64 KiB; an ABBA-interleaved 16-vs-32 KiB confirmation measured 5.974 ms versus 6.024 ms median, so the existing 16 KiB capacity remains the best measured choice without increasing per-call memory.
+
+## Pure Python serializer
+
+These profiles compare the pure-Python serializer before and after native JSON type fast paths on CPython 3.15.0b3.
+
+The workload serializes a deterministic 5,000-record nested payload 20 times with type attributes enabled. Python 3.15's tracing profiler captured the call tree, and FlameProf rendered the SVGs.
+
+| Profile | Traced time | Function calls | `isinstance` calls |
+| --- | ---: | ---: | ---: |
+| Before | 8.311 s | 48.17 million | 11.70 million |
+| After | 5.782 s | 30.13 million | 2.80 million |
+
+### Python before
+
+[![Serializer flamegraph before optimization](python315-before.svg)](python315-before.svg)
+
+### Python after
+
+[![Serializer flamegraph after optimization](python315-after.svg)](python315-after.svg)