Cythonize (de)serialization methods and enable errors customization (#…

…235)

Closures have been converted into classes with one method. They have been regrouped in modules called methods.py. Only methods modules are cythonized.

Cythonization is made by generating a pyx file from the Python file. This approach doesn't require much adaptation to the Python code. Also, it allows optimizations like dynamic dispatch conversion to switch (Python code would have been dirty to with a lot of if-chains).

As constraints validation has been completely refactored to be cythonized too, error customization has been added at the same time as it was simpler.

.github/workflows/ci.yml

@@ -8,12 +8,16 @@ on:
 
 jobs:
   test:
-    name: Test ${{ matrix.python-version }}
+    name: Test ${{ matrix.python-version }}${{ matrix.compile && ' compiled' || '' }}
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
       matrix:
         python-version: ['3.6', '3.7', '3.8', '3.9', '3.10', 'pypy3']
+        compile: [true, false]
+        exclude:
+        - python-version: pypy3
+          compile: true
     steps:
     - uses: actions/cache@v2
       with:
@@ -24,6 +28,12 @@ jobs:
       uses: actions/setup-python@v2
       with:
         python-version: ${{ matrix.python-version }}
+    - name: cythonize
+      if: matrix.compile
+      run: |
+        python -m pip install cython ${{ matrix.compile && (matrix.python-version == '3.6' || matrix.python-version == 'pypy3') && 'dataclasses' || '' }}
+        python scripts/cythonize.py
+        python setup.py build_ext --inplace
     - name: Install tox
       run: |
         python -m pip install --upgrade pip

.gitignore

@@ -106,3 +106,6 @@ venv.bak/
 .idea
 __generated__
 cov-*
+*.c
+*.pyx
+*.pxd

README.md

@@ -24,7 +24,7 @@ This library fulfills the following goals:
 
 - stay as close as possible to the standard library (dataclasses, typing, etc.) — as a consequence we do not need plugins for editors/linters/etc.;
 - be adaptable, provide tools to support any types (ORM, etc.);
-- avoid dynamic things like using raw strings for attributes name - play nicely with your IDE.
+- avoid dynamic things like using raw strings for attributes name — play nicely with your IDE.
 
 No known alternative achieves all of this, and apischema is also [faster](https://wyfo.github.io/apischema/performance_and_benchmark) than all of them.
 

apischema/conversions/visitor.py

@@ -72,20 +72,20 @@ def annotated(self, tp: AnyType, annotations: Sequence[Any]) -> Result:
                     return super().annotated(tp, annotations)
         return super().annotated(tp, annotations)
 
-    def _union_results(self, alternatives: Iterable[AnyType]) -> Sequence[Result]:
+    def _union_results(self, types: Iterable[AnyType]) -> Sequence[Result]:
         results = []
-        for alt in alternatives:
+        for alt in types:
             with suppress(Unsupported):
                 results.append(self.visit(alt))
         if not results:
-            raise Unsupported(Union[tuple(alternatives)])
+            raise Unsupported(Union[tuple(types)])
         return results
 
     def _visited_union(self, results: Sequence[Result]) -> Result:
         raise NotImplementedError
 
-    def union(self, alternatives: Sequence[AnyType]) -> Result:
-        return self._visited_union(self._union_results(alternatives))
+    def union(self, types: Sequence[AnyType]) -> Result:
+        return self._visited_union(self._union_results(types))
 
     @contextmanager
     def _replace_conversion(self, conversion: Optional[AnyConversion]):
@@ -130,7 +130,7 @@ def visit(self, tp: AnyType) -> Result:
                 tp, self.default_conversion(get_origin_or_type(tp))  # type: ignore
             )
         next_conversion = None
-        if not dynamic and is_subclass(tp, Collection):
+        if not dynamic and is_subclass(tp, Collection) and not is_subclass(tp, str):
             next_conversion = self._conversion
         return self.visit_conversion(tp, conversion, dynamic, next_conversion)