- Cython functions now use the PEP-590 vectorcall protocol in Py3.7+. Patch by Jeroen Demeyer. (Github issue #2263)
- Unicode identifiers are supported in Cython code (PEP 3131). Patch by da-woods. (Github issue #2601)
with gil/nogil
statements can be conditional based on compile-time constants, e.g. fused type checks. Patch by Noam Hershtig. (Github issue #2579)const
can be used together with fused types. Patch by Thomas Vincent. (Github issue #1772)- Reimports of already imported modules are substantially faster. (Github issue #2854)
- Positional-only arguments are supported in Python functions. Patch by Josh Tobin. (Github issue #2915)
- The
volatile
C modifier is supported in Cython code. Patch by Jeroen Demeyer. (Github issue #1667) @cython.trashcan(True)
can be used on an extension type to enable the CPython trashcan. This allows deallocating deeply recursive objects without overflowing the stack. Patch by Jeroen Demeyer. (Github issue #2842)- Properties can be defined for external extension types. Patch by Matti Picus. (Github issue #2640)
- Multiplication of Python numbers with small constant integers is faster. (Github issue #2808)
- Extension types that do not need their own
tp_new
implementation (because they have no object attributes etc.) directly inherit the implementation of their parent type if possible. (Github issue #1555) - The attributes
gen.gi_frame
andcoro.cr_frame
of Cython compiled generators and coroutines now return an actual frame object for introspection. (Github issue #2306) - Several missing declarations in
cpython.*
were added. Patches by Jeroen Demeyer and Zackery Spytz. (Github issues #2826, #2713) - The builtin
abs()
function can now be used on C numbers in nogil code. Patch by Elliott Sales de Andrade. (Github issue #2748) - PEP-479 (
generator_stop
) is now enabled by default with language level 3. (Github issue #2580) - Code annotation accepts a new debugging argument
--annotate-fullc
that will include the complete syntax highlighted C file in the HTML output. (Github issue #2855) --no-capture
added toruntests.py
to prevent stdout/stderr capturing during srctree tests. Patch by Matti Picus.--no-docstrings
option added tocythonize
script. Original patch by mo-han. (Github issue #2889)- The
@cython.binding
decorator is available in Python code.
- The unicode methods
.upper()
,.lower()
and.title()
were incorrectly optimised for single character input values and only returned the first character if multiple characters should have been returned. They now use the original Python methods again. - Diverging from the usual behaviour,
len(memoryview)
,len(char*)
andlen(Py_UNICODE*)
returned an unsignedsize_t
value. They now return a signedPy_ssize_t
, like other usages oflen()
. - Nested dict literals in function call kwargs could incorrectly raise an error about duplicate keyword arguments, which are allowed when passing them from dict literals. (Github issue #2963)
- Item access (subscripting) with integer indices/keys always tried the Sequence protocol before the Mapping protocol, which diverged from Python semantics. It now passes through the Mapping protocol first when supported. (Github issue #1807)
- Name lookups in class bodies no longer go through an attribute lookup. Patch by Jeroen Demeyer. (Github issue #3100)
- Broadcast assignments to a multi-dimensional memory view slice could end up in the wrong places when the underlying memory view is known to be contiguous but the slice is not. (Github issue #2941)
- Pickling unbound methods of Python classes failed. Patch by Pierre Glaser. (Github issue #2972)
- The
Py_hash_t
type failed to accept arbitrary "index" values. (Github issue #2752) - The first function line number of functions with decorators pointed to the signature line and not the first decorator line, as in Python. Patch by Felix Kohlgrüber. (Github issue #2536)
- Constant integer expressions that used a negative exponent were evaluated as integer 0 instead of the expected float value. Patch by Kryštof Pilnáček. (Github issue #2133)
__doc__
was not available inside of the class body during class creation. (Github issue #1635)- Setting
language_level=2
in a file did not work iflanguage_level=3
was enabled globally before. Patch by Jeroen Demeyer. (Github issue #2791) __init__.pyx
files were not always considered as package indicators. (Github issue #2665)- Compiling package
__init__
files could fail under Windows due to an undefined export symbol. (Github issue #2968) - A C compiler cast warning was resolved. Patch by Michael Buesch. (Github issue #2775)
- Binding staticmethods of Cython functions were not behaving like Python methods. Patch by Jeroen Demeyer. (Github issue #3106, #3102)
- The default language level was changed to
3str
, i.e. Python 3 semantics, but withstr
literals (also in Python 2.7). This is a backwards incompatible change from the previous default of Python 2 semantics. The previous behaviour is available through the directivelanguage_level=2
. - Cython no longer generates
__qualname__
attributes for classes in Python 2.x since they are problematic there and not correctly maintained for subclasses. Patch by Jeroen Demeyer. (Github issue #2772) - Source file fingerprinting now uses SHA-1 instead of MD5 since the latter tends to be slower and less widely supported these days. (Github issue #2790)
- The long deprecated include files
python_*
,stdio
,stdlib
andstl
inCython/Includes/Deprecated/
were removed. Use thelibc.*
andcpython.*
pxd modules instead. Patch by Jeroen Demeyer. (Github issue #2904) - The search order for include files was changed. Previously it was
include_directories
,Cython/Includes
,sys.path
. Now it isinclude_directories
,sys.path
,Cython/Includes
. This was done to allow third-party*.pxd
files to override the ones in Cython. Patch by Matti Picus. (Github issue #2905) - The command line parser was rewritten and modernised using
argparse
. Patch by Egor Dranischnikow. (Github issue #2952, #3001) - Support for Python 2.6 was removed.
- Double reference free in
__class__
cell handling forsuper()
calls. (Github issue #3246) - Deprecated import failed in Python 3.9. (Github issue #3266)
- The generated code failed to initialise the
tp_print
slot in CPython 3.8. Patches by Pablo Galindo and Orivej Desh (Github issues #3171, #3201). ?
forbool
was missing from the supported NumPy dtypes. Patch by Max Klein. (Github issue #2675)await
was not allowed inside of f-strings. Patch by Dmitro Getz. (Github issue #2877)- Coverage analysis failed for projects where the code resides in separate source sub-directories. Patch by Antonio Valentino. (Github issue #1985)
- An incorrect compiler warning was fixed in automatic C++ string conversions. Patch by Gerion Entrup. (Github issue #3108)
- Error reports in the Jupyter notebook showed unhelpful stack traces. Patch by Matthew Edwards (Github issue #3196).
Python.h
is now also included explicitly frompublic
header files. (Github issue #3133).- Distutils builds with
--parallel
did not work when using Cython's deprecatedbuild_ext
command. Patch by Alphadelta14 (Github issue #3187).
- The
PyMemoryView_*()
C-API is available incpython.memoryview
. Patch by Nathan Manville. (Github issue #2541)
- A reference leak for
None
was fixed when converting a memoryview to a Python object. (Github issue #3023) - The declaration of
PyGILState_STATE
incpython.pystate
was unusable. Patch by Kirill Smelkov. (Github issue #2997)
- The declarations in
posix.mman
were extended. Patches by Kirill Smelkov. (Github issues #2893, #2894, #3012)
- Fix compile error in CPython 3.8b2 regarding the
PyCode_New()
signature. (Github issue #3031) - Fix a C compiler warning about a missing
int
downcast. (Github issue #3028) - Fix reported error positions of undefined builtins and constants. Patch by Orivej Desh. (Github issue #3030)
- A 32 bit issue in the Pythran support was resolved. Patch by Serge Guelton. (Github issue #3032)
- Fix compile error in CPython 3.8b2 regarding the
PyCode_New()
signature. Patch by Nick Coghlan. (Github issue #3009) - Invalid C code generated for lambda functions in cdef methods. Patch by Josh Tobin. (Github issue #2967)
- Support slice handling in newer Pythran versions. Patch by Serge Guelton. (Github issue #2989)
- A reference leak in power-of-2 calculation was fixed. Patch by Sebastian Berg. (Github issue #3022)
- The search order for include files was changed. Previously it was
include_directories
,Cython/Includes
,sys.path
. Now it isinclude_directories
,sys.path
,Cython/Includes
. This was done to allow third-party*.pxd
files to override the ones in Cython. Original patch by Matti Picus. (Github issue #2905) - Setting
language_level=2
in a file did not work iflanguage_level=3
was enabled globally before. Patch by Jeroen Demeyer. (Github issue #2791)
- Fix compile errors in CPython 3.8b1 due to the new "tp_vectorcall" slots. (Github issue #2976)
- Fix a crash regression in 0.29.8 when creating code objects fails.
- Remove an incorrect cast when using true-division in C++ operations. (Github issue #1950)
- C compile errors with CPython 3.8 were resolved. Patch by Marcel Plch. (Github issue #2938)
- Python tuple constants that compare equal but have different item types could incorrectly be merged into a single constant. (Github issue #2919)
- Non-ASCII characters in unprefixed strings could crash the compiler when
used with language level
3str
. - Starred expressions in %-formatting tuples could fail to compile for unicode strings. (Github issue #2939)
- Passing Python class references through
cython.inline()
was broken. (Github issue #2936)
- Crash when the shared Cython config module gets unloaded and another Cython module reports an exceptions. Cython now makes sure it keeps an owned reference to the module. (Github issue #2885)
- Resolved a C89 compilation problem when enabling the fast-gil sharing feature.
- Coverage reporting did not include the signature line of
cdef
functions. (Github issue #1461) - Casting a GIL-requiring function into a nogil function now issues a warning. (Github issue #2879)
- Generators and coroutines were missing their return type annotation. (Github issue #2884)
- Fix a crash when accessing the
__kwdefaults__
special attribute of fused functions. (Github issue #1470) - Fix the parsing of buffer format strings that contain numeric sizes, which could lead to incorrect input rejections. (Github issue #2845)
- Avoid a C #pragma in old gcc versions that was only added in GCC 4.6. Patch by Michael Anselmi. (Github issue #2838)
- Auto-encoding of Unicode strings to UTF-8 C/C++ strings failed in Python 3, even though the default encoding there is UTF-8. (Github issue #2819)
- Crash when defining a Python subclass of an extension type and repeatedly calling a cpdef method on it. (Github issue #2823)
- Compiler crash when
prange()
loops appear inside of with-statements. (Github issue #2780) - Some C compiler warnings were resolved. Patches by Christoph Gohlke. (Github issues #2815, #2816, #2817, #2822)
- Python conversion of C++ enums failed in 0.29. Patch by Orivej Desh. (Github issue #2767)
- Division of numeric constants by a runtime value of 0 could fail to raise a
ZeroDivisionError
. (Github issue #2820)
- Some C code for memoryviews was generated in a non-deterministic order. Patch by Martijn van Steenbergen. (Github issue #2779)
- C89 compatibility was accidentally lost since 0.28. Patches by gastineau and true-pasky. (Github issues #2778, #2801)
- A C compiler cast warning was resolved. Patch by Michael Buesch. (Github issue #2774)
- An compilation failure with complex numbers under MSVC++ was resolved. (Github issue #2797)
- Coverage reporting could fail when modules were moved around after the build. Patch by Wenjun Si. (Github issue #2776)
- The code generated for deduplicated constants leaked some references. (Github issue #2750)
- The declaration of
sigismember()
inlibc.signal
was corrected. (Github issue #2756) - Crashes in compiler and test runner were fixed. (Github issue #2736, #2755)
- A C compiler warning about an invalid safety check was resolved. (Github issue #2731)
- Extensions compiled with MinGW-64 under Windows could misinterpret integer objects larger than 15 bit and return incorrect results. (Github issue #2670)
- Cython no longer requires the source to be writable when copying its data into a memory view slice. Patch by Andrey Paramonov. (Github issue #2644)
- Line tracing of
try
-statements generated invalid C code. (Github issue #2274) - When using the
warn.undeclared
directive, Cython's own code generated warnings that are now fixed. Patch by Nicolas Pauss. (Github issue #2685) - Cython's memoryviews no longer require strides for setting the shape field
but only the
PyBUF_ND
flag to be set. Patch by John Kirkham. (Github issue #2716) - Some C compiler warnings about unused memoryview code were fixed. Patch by Ho Cheuk Ting. (Github issue #2588)
- A C compiler warning about implicit signed/unsigned conversion was fixed. (Github issue #2729)
- Assignments to C++ references returned by
operator[]
could fail to compile. (Github issue #2671) - The power operator and the support for NumPy math functions were fixed in Pythran expressions. Patch by Serge Guelton. (Github issues #2702, #2709)
- Signatures with memory view arguments now show the expected type when embedded in docstrings. Patch by Matthew Chan and Benjamin Weigel. (Github issue #2634)
- Some
from ... cimport ...
constructs were not correctly considered when searching modified dependencies incythonize()
to decide whether to recompile a module. Patch by Kryštof Pilnáček. (Github issue #2638) - A struct field type in the
cpython.array
declarations was corrected. Patch by John Kirkham. (Github issue #2712)
- PEP-489 multi-phase module initialisation has been enabled again. Module reloads in other subinterpreters raise an exception to prevent corruption of the static module state.
- A set of
mypy
compatible PEP-484 declarations were added for Cython's C data types to integrate with static analysers in typed Python code. They are available in theCython/Shadow.pyi
module and describe the types in the specialcython
module that can be used for typing in Python code. Original patch by Julian Gethmann. (Github issue #1965) - Memoryviews are supported in PEP-484/526 style type declarations. (Github issue #2529)
@cython.nogil
is supported as a C-function decorator in Python code. (Github issue #2557)- Raising exceptions from nogil code will automatically acquire the GIL, instead
of requiring an explicit
with gil
block. - C++ functions can now be declared as potentially raising both C++ and Python exceptions, so that Cython can handle both correctly. (Github issue #2615)
cython.inline()
supports a directlanguage_level
keyword argument that was previously only available via a directive.- A new language level name
3str
was added that mostly corresponds to language level 3, but keeps unprefixed string literals as type 'str' in both Py2 and Py3, and the builtin 'str' type unchanged. This will become the default in the next Cython release and is meant to help user code a) transition more easily to this new default and b) migrate to Python 3 source code semantics without making support for Python 2.x difficult. - In CPython 3.6 and later, looking up globals in the module dict is almost as fast as looking up C globals. (Github issue #2313)
- For a Python subclass of an extension type, repeated method calls to non-overridden cpdef methods can avoid the attribute lookup in Py3.6+, which makes them 4x faster. (Github issue #2313)
- (In-)equality comparisons of objects to integer literals are faster. (Github issue #2188)
- Some internal and 1-argument method calls are faster.
- Modules that cimport many external extension types from other Cython modules execute less import requests during module initialisation.
- Constant tuples and slices are deduplicated and only created once per module. (Github issue #2292)
- The coverage plugin considers more C file extensions such as
.cc
and.cxx
. (Github issue #2266) - The
cythonize
command accepts compile time variable values (as set byDEF
) through the new-E
option. Patch by Jerome Kieffer. (Github issue #2315) pyximport
can import from namespace packages. Patch by Prakhar Goel. (Github issue #2294)- Some missing numpy and CPython C-API declarations were added. Patch by John Kirkham. (Github issues #2523, #2520, #2537)
- Declarations for the
pylifecycle
C-API functions were added in a new .pxd filecpython.pylifecycle
. - The Pythran support was updated to work with the latest Pythran 0.8.7. Original patch by Adrien Guinet. (Github issue #2600)
%a
is included in the string formatting types that are optimised into f-strings. In this case, it is also automatically mapped to%r
in Python 2.x.- New C macro
CYTHON_HEX_VERSION
to access Cython's version in the same style asPY_VERSION_HEX
. - Constants in
libc.math
are now declared asconst
to simplify their handling. - An additional
check_size
clause was added to thectypedef class
name specification to allow suppressing warnings when importing modules with backwards-compatiblePyTypeObject
size changes. Patch by Matti Picus. (Github issue #2627)
- The exception handling in generators and coroutines under CPython 3.7 was adapted to the newly introduced exception stack. Users of Cython 0.28 who want to support Python 3.7 are encouraged to upgrade to 0.29 to avoid potentially incorrect error reporting and tracebacks. (Github issue #1958)
- Crash when importing a module under Stackless Python that was built for CPython. Patch by Anselm Kruis. (Github issue #2534)
- 2-value slicing of typed sequences failed if the start or stop index was None. Patch by Christian Gibson. (Github issue #2508)
- Multiplied string literals lost their factor when they are part of another constant expression (e.g. 'x' * 10 + 'y' => 'xy').
- String formatting with the '%' operator didn't call the special
__rmod__()
method if the right side is a string subclass that implements it. (Python issue 28598) - The directive
language_level=3
did not apply to the first token in the source file. (Github issue #2230) - Overriding cpdef methods did not work in Python subclasses with slots. Note that this can have a performance impact on calls from Cython code. (Github issue #1771)
- Fix declarations of builtin or C types using strings in pure python mode. (Github issue #2046)
- Generator expressions and lambdas failed to compile in
@cfunc
functions. (Github issue #459) - Global names with
const
types were not excluded from star-import assignments which could lead to invalid C code. (Github issue #2621) - Several internal function signatures were fixed that lead to warnings in gcc-8. (Github issue #2363)
- The numpy helper functions
set_array_base()
andget_array_base()
were adapted to the current numpy C-API recommendations. Patch by Matti Picus. (Github issue #2528) - Some NumPy related code was updated to avoid deprecated API usage. Original patch by jbrockmendel. (Github issue #2559)
- Several C++ STL declarations were extended and corrected. Patch by Valentin Valls. (Github issue #2207)
- C lines of the module init function were unconditionally not reported in exception stack traces. Patch by Jeroen Demeyer. (Github issue #2492)
- When PEP-489 support is enabled, reloading the module overwrote any static module state. It now raises an exception instead, given that reloading is not actually supported.
- Object-returning, C++ exception throwing functions were not checking that the return value was non-null. Original patch by Matt Wozniski (Github Issue #2603)
- The source file encoding detection could get confused if the
c_string_encoding
directive appeared within the first two lines. (Github issue #2632) - Cython generated modules no longer emit a warning during import when the size of the NumPy array type is larger than what was found at compile time. Instead, this is assumed to be a backwards compatible change on NumPy side.
- Cython now emits a warning when no
language_level
(2, 3 or '3str') is set explicitly, neither as acythonize()
option nor as a compiler directive. This is meant to prepare the transition of the default language level from currently Py2 to Py3, since that is what most new users will expect these days. The future default will, however, not enforce unicode literals, because this has proven a major obstacle in the support for both Python 2.x and 3.x. The next major release is intended to make this change, so that it will parse all code that does not request a specific language level as Python 3 code, but withstr
literals. The language level 2 will continue to be supported for an indefinite time. - The documentation was restructured, cleaned up and examples are now tested. The NumPy tutorial was also rewritten to simplify the running example. Contributed by Gabriel de Marmiesse. (Github issue #2245)
- Cython compiles less of its own modules at build time to reduce the installed package size to about half of its previous size. This makes the compiler slightly slower, by about 5-7%.
- Extensions compiled with MinGW-64 under Windows could misinterpret integer objects larger than 15 bit and return incorrect results. (Github issue #2670)
- Multiplied string literals lost their factor when they are part of another constant expression (e.g. 'x' * 10 + 'y' => 'xy').
- The discouraged usage of GCC's attribute
optimize("Os")
was replaced by the similar attributecold
to reduce the code impact of the module init functions. (Github issue #2494) - A reference leak in Py2.x was fixed when comparing str to unicode for equality.
- Reallowing
tp_clear()
in a subtype of an@no_gc_clear
extension type generated an invalid C function call to the (non-existent) base type implementation. (Github issue #2309) - Exception catching based on a non-literal (runtime) tuple could fail to match the exception. (Github issue #2425)
- Compile fix for CPython 3.7.0a2. (Github issue #2477)
- Set iteration was broken in non-CPython since 0.28.
UnicodeEncodeError
in Py2 when%s
formatting is optimised for unicode strings. (Github issue #2276)- Work around a crash bug in g++ 4.4.x by disabling the size reduction setting of the module init function in this version. (Github issue #2235)
- Crash when exceptions occur early during module initialisation. (Github issue #2199)
abs()
is faster for Python long objects.- The C++11 methods
front()
andend()
were added to the declaration oflibcpp.string
. Patch by Alex Huszagh. (Github issue #2123) - The C++11 methods
reserve()
andbucket_count()
are declared forlibcpp.unordered_map
. Patch by Valentin Valls. (Github issue #2168)
- The copy of a read-only memoryview was considered read-only as well, whereas a common reason to copy a read-only view is to make it writable. The result of the copying is now a writable buffer by default. (Github issue #2134)
- The
switch
statement generation failed to apply recursively to the body of converted if-statements. NULL
was sometimes rejected as exception return value when the returned type is a fused pointer type. Patch by Callie LeFave. (Github issue #2177)- Fixed compatibility with PyPy 5.11. Patch by Matti Picus. (Github issue #2165)
- The NumPy tutorial was rewritten to use memoryviews instead of the older buffer declaration syntax. Contributed by Gabriel de Marmiesse. (Github issue #2162)
PyFrozenSet_New()
was accidentally used in PyPy where it is missing from the C-API.- Assignment between some C++ templated types were incorrectly rejected
when the templates mix
const
withctypedef
. (Github issue #2148) - Undeclared C++ no-args constructors in subclasses could make the compilation
fail if the base class constructor was declared without
nogil
. (Github issue #2157) - Bytes %-formatting inferred
basestring
(bytes or unicode) as result type in some cases wherebytes
would have been safe to infer. (Github issue #2153) None
was accidentally disallowed as typed return value ofdict.pop()
. (Github issue #2152)
- Cdef classes can now multiply inherit from ordinary Python classes.
(The primary base must still be a c class, possibly
object
, and the other bases must not be cdef classes.) - Type inference is now supported for Pythran compiled NumPy expressions. Patch by Nils Braun. (Github issue #1954)
- The
const
modifier can be applied to memoryview declarations to allow read-only buffers as input. (Github issues #1605, #1869) - C code in the docstring of a
cdef extern
block is copied verbatimly into the generated file. Patch by Jeroen Demeyer. (Github issue #1915) - When compiling with gcc, the module init function is now tuned for small code size instead of whatever compile flags were provided externally. Cython now also disables some code intensive optimisations in that function to further reduce the code size. (Github issue #2102)
- Decorating an async coroutine with
@cython.iterable_coroutine
changes its type at compile time to make it iterable. While this is not strictly in line with PEP-492, it improves the interoperability with old-style coroutines that useyield from
instead ofawait
. - The IPython magic has preliminary support for JupyterLab. (Github issue #1775)
- The new TSS C-API in CPython 3.7 is supported and has been backported. Patch by Naotoshi Seo. (Github issue #1932)
- Cython knows the new
Py_tss_t
type defined in PEP-539 and automatically initialises variables declared with that type toPy_tss_NEEDS_INIT
, a value which cannot be used outside of static assignments. - The set methods
.remove()
and.discard()
are optimised. Patch by Antoine Pitrou. (Github issue #2042) dict.pop()
is optimised. Original patch by Antoine Pitrou. (Github issue #2047)- Iteration over sets and frozensets is optimised. (Github issue #2048)
- Safe integer loops (< range(2^30)) are automatically optimised into C loops.
alist.extend([a,b,c])
is optimised into sequentiallist.append()
calls for short literal sequences.- Calls to builtin methods that are not specifically optimised into C-API calls now use a cache that avoids repeated lookups of the underlying C function. (Github issue #2054)
- Single argument function calls can avoid the argument tuple creation in some cases.
- Some redundant extension type checks are avoided.
- Formatting C enum values in f-strings is faster, as well as some other special cases.
- String formatting with the '%' operator is optimised into f-strings in simple cases.
- Subscripting (item access) is faster in some cases.
- Some
bytearray
operations have been optimised similar tobytes
. - Some PEP-484/526 container type declarations are now considered for loop optimisations.
- Indexing into memoryview slices with
view[i][j]
is now optimised intoview[i, j]
. - Python compatible
cython.*
types can now be mixed with type declarations in Cython syntax. - Name lookups in the module and in classes are faster.
- Python attribute lookups on extension types without instance dict are faster.
- Some missing signals were added to
libc/signal.pxd
. Patch by Jeroen Demeyer. (Github issue #1914) - The warning about repeated extern declarations is now visible by default. (Github issue #1874)
- The exception handling of the function types used by CPython's type slot functions was corrected to match the de-facto standard behaviour, so that code that uses them directly benefits from automatic and correct exception propagation. Patch by Jeroen Demeyer. (Github issue #1980)
- Defining the macro
CYTHON_NO_PYINIT_EXPORT
will prevent the module init function from being exported as symbol, e.g. when linking modules statically in an embedding setup. Patch by AraHaan. (Github issue #1944)
- If a module name is explicitly provided for an
Extension()
that is compiled viacythonize()
, it was previously ignored and replaced by the source file name. It can now be used to override the target module name, e.g. for compiling prefixed accelerator modules from Python files. (Github issue #2038) - The arguments of the
num_threads
parameter of parallel sections were not sufficiently validated and could lead to invalid C code. (Github issue #1957) - Catching exceptions with a non-trivial exception pattern could call into CPython with a live exception set. This triggered incorrect behaviour and crashes, especially in CPython 3.7.
- The signature of the special
__richcmp__()
method was corrected to recognise the type of the first argument asself
. It was previously treated as plain object, but CPython actually guarantees that it always has the correct type. Note: this can change the semantics of user code that previously relied onself
being untyped. - Some Python 3 exceptions were not recognised as builtins when running Cython under Python 2.
- Some async helper functions were not defined in the generated C code when compiling simple async code. (Github issue #2075)
- Line tracing did not include generators and coroutines. (Github issue #1949)
- C++ declarations for
unordered_map
were corrected. Patch by Michael Schatzow. (Github issue #1484) - Iterator declarations in C++
deque
andvector
were corrected. Patch by Alex Huszagh. (Github issue #1870) - The const modifiers in the C++
string
declarations were corrected, together with the coercion behaviour of string literals into C++ strings. (Github issue #2132) - Some declaration types in
libc.limits
were corrected. Patch by Jeroen Demeyer. (Github issue #2016) @cython.final
was not accepted on Python classes with an@cython.cclass
decorator. (Github issue #2040)- Cython no longer creates useless and incorrect
PyInstanceMethod
wrappers for methods in Python 3. Patch by Jeroen Demeyer. (Github issue #2105) - The builtin
bytearray
type could not be used as base type of cdef classes. (Github issue #2106)
- String forward references to extension types like
@cython.locals(x="ExtType")
failed to find the named type. (Github issue #1962) - NumPy slicing generated incorrect results when compiled with Pythran. Original patch by Serge Guelton (Github issue #1946).
- Fix "undefined reference" linker error for generators on Windows in Py3.3-3.5. (Github issue #1968)
- Adapt to recent C-API change of
PyThreadState
in CPython 3.7. - Fix signature of
PyWeakref_GetObject()
API declaration. Patch by Jeroen Demeyer (Github issue #1975).
- Comprehensions could incorrectly be optimised away when they appeared in boolean test contexts. (Github issue #1920)
- The special methods
__eq__
,__lt__
etc. in extension types did not type their first argument as the type of the class butobject
. (Github issue #1935) - Crash on first lookup of "cline_in_traceback" option during exception handling. (Github issue #1907)
- Some nested module level comprehensions failed to compile. (Github issue #1906)
- Compiler crash on some complex type declarations in pure mode. (Github issue #1908)
std::unordered_map.erase()
was declared with an incorrectvoid
return type inlibcpp.unordered_map
. (Github issue #1484)- Invalid use of C++
fallthrough
attribute before C++11 and similar issue in clang. (Github issue #1930) - Compiler crash on misnamed properties. (Github issue #1905)
- The Jupyter magic has a new debug option
--verbose
that shows details about the distutils invocation. Patch by Boris Filippov (Github issue #1881).
- Py3 list comprehensions in class bodies resulted in invalid C code. (Github issue #1889)
- Modules built for later CPython 3.5.x versions failed to import in 3.5.0/3.5.1. (Github issue #1880)
- Deallocating fused types functions and methods kept their GC tracking enabled, which could potentially lead to recursive deallocation attempts.
- Crash when compiling in C++ mode with old setuptools versions. (Github issue #1879)
- C++ object arguments for the constructor of Cython implemented C++ are now
passed by reference and not by value to allow for non-copyable arguments, such
as
unique_ptr
. - API-exported C++ classes with Python object members failed to compile. (Github issue #1866)
- Some issues with the new relaxed exception value handling were resolved.
- Python classes as annotation types could prevent compilation. (Github issue #1887)
- Cython annotation types in Python files could lead to import failures with a "cython undefined" error. Recognised types are now turned into strings.
- Coverage analysis could fail to report on extension modules on some platforms.
- Annotations could be parsed (and rejected) as types even with
annotation_typing=False
.
- PEP 489 support has been disabled by default to counter incompatibilities with import setups that try to reload or reinitialise modules.
- Extension module initialisation follows
PEP 489 in CPython 3.5+, which
resolves several differences with regard to normal Python modules. This makes
the global names
__file__
and__path__
correctly available to module level code and improves the support for module-level relative imports. (Github issues #1715, #1753, #1035) - Asynchronous generators (PEP 525) and asynchronous comprehensions (PEP 530) have been implemented. Note that async generators require finalisation support in order to allow for asynchronous operations during cleanup, which is only available in CPython 3.6+. All other functionality has been backported as usual.
- Variable annotations are now parsed according to
PEP 526. Cython types (e.g.
cython.int
) are evaluated as C type declarations and everything else as Python types. This can be disabled with the directiveannotation_typing=False
. Note that most complex PEP-484 style annotations are currently ignored. This will change in future releases. (Github issue #1850) - Extension types (also in pure Python mode) can implement the normal special methods
__eq__
,__lt__
etc. for comparisons instead of the low-level__richcmp__
method. (Github issue #690) - New decorator
@cython.exceptval(x=None, check=False)
that makes the signature declarationsexcept x
,except? x
andexcept *
available to pure Python code. Original patch by Antonio Cuni. (Github issue #1653) - Signature annotations are now included in the signature docstring generated by
the
embedsignature
directive. Patch by Lisandro Dalcin (Github issue #1781). - The gdb support for Python code (
libpython.py
) was updated to the latest version in CPython 3.7 (git rev 5fe59f8). - The compiler tries to find a usable exception return value for cdef functions
with
except *
if the returned type allows it. Note that this feature is subject to safety limitations, so it is still better to provide an explicit declaration. - C functions can be assigned to function pointers with a compatible exception declaration, not only with exact matches. A side-effect is that certain compatible signature overrides are now allowed and some more mismatches of exception signatures are now detected and rejected as errors that were not detected before.
- The IPython/Jupyter magic integration has a new option
%%cython --pgo
for profile guided optimisation. It compiles the cell with PGO settings for the C compiler, executes it to generate a runtime profile, and then compiles it again using that profile for C compiler optimisation. Currently only tested with gcc. len(memoryview)
can be used in nogil sections to get the size of the first dimension of a memory view (shape[0]
). (Github issue #1733)- C++ classes can now contain (properly refcounted) Python objects.
- NumPy dtype subarrays are now accessible through the C-API. Patch by Gerald Dalley (Github issue #245).
- Resolves several issues with PyPy and uses faster async slots in PyPy3. Patch by Ronan Lamy (Github issues #1871, #1878).
- Extension types that were cimported from other Cython modules could disagree about the order of fused cdef methods in their call table. This could lead to wrong methods being called and potentially also crashes. The fix required changes to the ordering of fused methods in the call table, which may break existing compiled modules that call fused cdef methods across module boundaries, if these methods were implemented in a different order than they were declared in the corresponding .pxd file. (Github issue #1873)
- The exception state handling in generators and coroutines could lead to exceptions in the caller being lost if an exception was raised and handled inside of the coroutine when yielding. (Github issue #1731)
- Loops over
range(enum)
were not converted into C for-loops. Note that it is still recommended to use an explicit cast to a C integer type in this case. - Error positions of names (e.g. variables) were incorrectly reported after the name and not at the beginning of the name.
- Compile time
DEF
assignments were evaluated even when they occur inside of falsyIF
blocks. (Github issue #1796) - Disabling the line tracing from a trace function could fail. Original patch by Dmitry Trofimov. (Github issue #1769)
- Several issues with the Pythran integration were resolved.
- abs(signed int) now returns a signed rather than unsigned int. (Github issue #1837)
- Reading
frame.f_locals
of a Cython function (e.g. from a debugger or profiler could modify the module globals. (Github issue #1836) - Buffer type mismatches in the NumPy buffer support could leak a reference to the buffer owner.
- Using the "is_f_contig" and "is_c_contig" memoryview methods together could leave one of them undeclared. (Github issue #1872)
- Compilation failed if the for-in-range loop target was not a variable but a more complex expression, e.g. an item assignment. (Github issue #1831)
- Compile time evaluations of (partially) constant f-strings could show incorrect results.
- Escape sequences in raw f-strings (
fr'...'
) were resolved instead of passing them through as expected. - Some ref-counting issues in buffer error handling have been resolved.
- Type declarations in signature annotations are now parsed according to
PEP 484
typing. Only Cython types (e.g.
cython.int
) and Python builtin types are currently considered as type declarations. Everything else is ignored, but this will change in a future Cython release. (Github issue #1672) - The directive
annotation_typing
is nowTrue
by default, which enables parsing type declarations from annotations. - This release no longer supports Python 3.2.
cython.view.array
was missing.__len__()
.- Extension types with a
.pxd
override for their__releasebuffer__
slot (e.g. as provided by Cython for the Pythonarray.array
type) could leak a reference to the buffer owner on release, thus not freeing the memory. (Github issue #1638) - Auto-decoding failed in 0.26 for strings inside of C++ containers. (Github issue #1790)
- Compile error when inheriting from C++ container types. (Github issue #1788)
- Invalid C code in generators (declaration after code). (Github issue #1801)
- Arithmetic operations on
const
integer variables could generate invalid code. (Github issue #1798) - Local variables with names of special Python methods failed to compile inside of closures. (Github issue #1797)
- Problem with indirect Emacs buffers in cython-mode. Patch by Martin Albrecht (Github issue #1743).
- Extension types named
result
orPickleError
generated invalid unpickling code. Patch by Jason Madden (Github issue #1786). - Bazel integration failed to compile
.py
files. Patch by Guro Bokum (Github issue #1784). - Some include directories and dependencies were referenced with their absolute paths in the generated files despite lying within the project directory.
- Failure to compile in Py3.7 due to a modified signature of
_PyCFunctionFast()
- Pythran can be used as a backend for evaluating NumPy array expressions. Patch by Adrien Guinet (Github issue #1607).
- cdef classes now support pickling by default when possible.
This can be disabled with the
auto_pickle
directive. - Speed up comparisons of strings if their hash value is available. Patch by Claudio Freire (Github issue #1571).
- Support pyximport from zip files. Patch by Sergei Lebedev (Github issue #1485).
- IPython magic now respects the
__all__
variable and ignores names with leading-underscore (likeimport *
does). Patch by Syrtis Major (Github issue #1625). abs()
is optimised for C complex numbers. Patch by da-woods (Github issue #1648).- The display of C lines in Cython tracebacks can now be enabled at runtime
via
import cython_runtime; cython_runtime.cline_in_traceback=True
. The default has been changed to False. - The overhead of calling fused types generic functions was reduced.
- "cdef extern" include files are now also searched relative to the current file. Patch by Jeroen Demeyer (Github issue #1654).
- Optional optimization for re-acquiring the GIL, controlled by the fast_gil directive.
- Item lookup/assignment with a unicode character as index that is typed
(explicitly or implicitly) as
Py_UCS4
orPy_UNICODE
used the integer value instead of the Unicode string value. Code that relied on the previous behaviour now triggers a warning that can be disabled by applying an explicit cast. (Github issue #1602) - f-string processing was adapted to changes in PEP 498 and CPython 3.6.
- Invalid C code when decoding from UTF-16(LE/BE) byte strings. (Github issue #1696)
- Unicode escapes in 'ur' raw-unicode strings were not resolved in Py2 code. Original patch by Aaron Gallagher (Github issue #1594).
- File paths of code objects are now relative. Original patch by Jelmer Vernooij (Github issue #1565).
- Decorators of cdef class methods could be executed twice. Patch by Jeroen Demeyer (Github issue #1724).
- Dict iteration using the Py2
iter*
methods failed in PyPy3. Patch by Armin Rigo (Github issue #1631). - Several warnings in the generated code are now suppressed.
- The
unraisable_tracebacks
option now defaults toTrue
. - Coercion of C++ containers to Python is no longer automatic on attribute access (Github issue #1521).
- Access to Python attributes of cimported modules without the corresponding import is now a compile-time (rather than runtime) error.
- Do not use special dll linkage for "cdef public" functions. Patch by Jeroen Demeyer (Github issue #1687).
- cdef/cpdef methods must match their declarations. See Github Issue #1732. This is now a warning and will be an error in future releases.
- Fixes several issues with C++ template deduction.
- Fixes a issue with bound method type inference (Github issue #551).
- Fixes a bug with cascaded tuple assignment (Github issue #1523).
- Fixed or silenced many Clang warnings.
- Fixes bug with powers of pure real complex numbers (Github issue #1538).
- Fixes a bug with
isinstance(o, Exception)
(Github issue #1496). - Fixes bug with
cython.view.array
missing utility code in some cases (Github issue #1502).
- The distutils extension
Cython.Distutils.build_ext
has been reverted, temporarily, to beold_build_ext
to give projects time to migrate. The new build_ext is available asnew_build_ext
.
- def/cpdef methods of cdef classes benefit from Cython's internal function implementation, which enables introspection and line profiling for them. Implementation sponsored by Turbostream (www.turbostream-cfd.com).
- Calls to Python functions are faster, following the recent "FastCall" optimisations that Victor Stinner implemented for CPython 3.6. See https://bugs.python.org/issue27128 and related issues.
- The new METH_FASTCALL calling convention for PyCFunctions is supported in CPython 3.6. See https://bugs.python.org/issue27810
- Initial support for using Cython modules in Pyston. Patch by Boxiang Sun.
- Dynamic Python attributes are allowed on cdef classes if an attribute
cdef dict __dict__
is declared in the class. Patch by empyrical. - Cython implemented C++ classes can make direct calls to base class methods. Patch by empyrical.
- C++ classes can now have typedef members. STL containers updated with value_type.
- New directive
cython.no_gc
to fully disable GC for a cdef class. Patch by Claudio Freire. - Buffer variables are no longer excluded from
locals()
. Patch by da-woods. - Building f-strings is faster, especially when formatting C integers.
- for-loop iteration over "std::string".
libc/math.pxd
providese
andpi
as alias constants to simplify usage as a drop-in replacement for Python's math module.- Speed up cython.inline().
- Binary lshift operations with small constant Python integers are faster.
- Some integer operations on Python long objects are faster in Python 2.7.
- Support for the C++
typeid
operator. - Support for bazel using a the pyx_library rule in //Tools:rules.bzl.
- Division of complex numbers avoids overflow by using Smith's method.
- Some function signatures in
libc.math
andnumpy.pxd
were incorrect. Patch by Michael Seifert.
- The "%%cython" IPython/jupyter magic now defaults to the language level of the current jupyter kernel. The language level can be set explicitly with "%%cython -2" or "%%cython -3".
- The distutils extension
Cython.Distutils.build_ext
has now been updated to use cythonize which properly handles dependencies. The old extension can still be found inCython.Distutils.old_build_ext
and is now deprecated. directive_defaults
is no longer available inCython.Compiler.Options
, useget_directive_defaults()
instead.
- IPython cell magic was lacking a good way to enable Python 3 code semantics. It can now be used as "%%cython -3".
- Follow a recent change in PEP 492
and CPython 3.5.2 that now requires the
__aiter__()
method of asynchronous iterators to be a simpledef
method instead of anasync def
method. - Coroutines and generators were lacking the
__module__
special attribute. - C++
std::complex
values failed to auto-convert from and to Python complex objects. - Namespaced C++ types could not be used as memory view types due to lack of name mangling. Patch by Ivan Smirnov.
- Assignments between identical C++ types that were declared with differently typedefed template types could fail.
- Rebuilds could fail to evaluate dependency timestamps in C++ mode. Patch by Ian Henriksen.
- Macros defined in the
distutils
compiler option do not require values anymore. Patch by Ian Henriksen. - Minor fixes for MSVC, Cygwin and PyPy.
- PEP 498: Literal String Formatting (f-strings). Original patch by Jelle Zijlstra.
- PEP 515: Underscores as visual separators in number literals.
- Parser was adapted to some minor syntax changes in Py3.6, e.g. https://bugs.python.org/issue9232
- The embedded C code comments that show the original source code
can be discarded with the new directive
emit_code_comments=False
. - Cpdef enums are now first-class iterable, callable types in Python.
- Ctuples can now be declared in pure Python code.
- Posix declarations for DLL loading and stdio extensions were added. Patch by Lars Buitinck.
- The Py2-only builtins
unicode()
,xrange()
,reduce()
andlong
are now also available in compile timeDEF
expressions when compiling with Py3. - Exception type tests have slightly lower overhead. This fixes ticket 868.
- @property syntax fully supported in cdef classes, old syntax deprecated.
- C++ classes can now be declared with default template parameters.
- C++ exceptions raised by overloaded C++ operators were not always handled. Patch by Ian Henriksen.
- C string literals were previously always stored as non-const global variables in the module. They are now stored as global constants when possible, and otherwise as non-const C string literals in the generated code that uses them. This improves compatibility with strict C compiler options and prevents non-const strings literals with the same content from being incorrectly merged.
- Compile time evaluated
str
expressions (DEF
) now behave in a more useful way by turning into Unicode strings when compiling under Python 3. This allows using them as intermediate values in expressions. Previously, they always evaluated to bytes objects. isinf()
declarations inlibc/math.pxd
andnumpy/math.pxd
now reflect the actual tristateint
return value instead of usingbint
.- Literal assignments to ctuples avoid Python tuple round-trips in some more corner cases.
- Iteration over
dict(...).items()
failed to get optimised when dict arguments included keyword arguments. - cProfile now correctly profiles cpdef functions and methods.
- Compile errors and warnings in integer type conversion code. This fixes ticket 877. Patches by Christian Neukirchen, Nikolaus Rath, Ian Henriksen.
- Reference leak when
*args
argument was reassigned in closures. - Truth-testing Unicode strings could waste time and memory in Py3.3+.
- Return values of async functions could be ignored and replaced by
None
. - Compiler crash in CPython 3.6.
- Fix prange() to behave identically to range(). The end condition was miscalculated when the range was not exactly divisible by the step.
- Optimised
all(genexpr)
/any(genexpr)
calls could warn about unused code. This fixes ticket 876.
- Memory leak when calling Python functions in PyPy.
- Compilation problem with MSVC in C99-ish mode.
- Warning about unused values in a helper macro.
- Invalid C code for some builtin methods. This fixes ticket 856 again.
- Incorrect C code in helper functions for PyLong conversion and string decoding. This fixes ticket 863, ticket 864 and ticket 865. Original patch by Nikolaus Rath.
- Large folded or inserted integer constants could use too small C integer types and thus trigger a value wrap-around.
- The coroutine and generator types of Cython now also register directly
with the
Coroutine
andGenerator
ABCs in thebackports_abc
module if it can be imported. This fixes ticket 870.
- Compiler crash when analysing some optimised expressions.
- Coverage plugin was adapted to coverage.py 4.0 beta 2.
- C++ destructor calls could fail when '&' operator is overwritten.
- Incorrect C literal generation for large integers in compile-time evaluated DEF expressions and constant folded expressions.
- Byte string constants could end up as Unicode strings when originating from compile-time evaluated DEF expressions.
- Invalid C code when caching known builtin methods. This fixes ticket 860.
ino_t
inposix.types
was not declared asunsigned
.- Declarations in
libcpp/memory.pxd
were missingoperator!()
. Patch by Leo Razoumov. - Static cdef methods can now be declared in .pxd files.
- Invalid C code for generators. This fixes ticket 858.
- Invalid C code for some builtin methods. This fixes ticket 856.
- Invalid C code for unused local buffer variables. This fixes ticket 154.
- Test failures on 32bit systems. This fixes ticket 857.
- Code that uses
from xyz import *
and global C struct/union/array variables could fail to compile due to missing helper functions. This fixes ticket 851. - Misnamed PEP 492 coroutine property
cr_yieldfrom
renamed tocr_await
to match CPython. - Missing deallocation code for C++ object attributes in certain extension class hierarchies.
- Crash when async coroutine was not awaited.
- Compiler crash on
yield
in signature annotations and default argument values. Both are forbidden now. - Compiler crash on certain constructs in
finally
clauses. - Cython failed to build when CPython's pgen is installed.
- PEP 492 (async/await) was implemented.
- PEP 448 (Additional Unpacking Generalizations) was implemented.
- Support for coverage.py 4.0+ can be enabled by adding the plugin "Cython.Coverage" to the ".coveragerc" config file.
- Annotated HTML source pages can integrate (XML) coverage reports.
- Tracing is supported in
nogil
functions/sections and module init code. - When generators are used in a Cython module and the module imports the modules "inspect" and/or "asyncio", Cython enables interoperability by patching these modules during the import to recognise Cython's internal generator and coroutine types. This can be disabled by C compiling the module with "-D CYTHON_PATCH_ASYNCIO=0" or "-D CYTHON_PATCH_INSPECT=0"
- When generators or coroutines are used in a Cython module, their types
are registered with the
Generator
andCoroutine
ABCs in thecollections
orcollections.abc
stdlib module at import time to enable interoperability with code that needs to detect and process Python generators/coroutines. These ABCs were added in CPython 3.5 and are available for older Python versions through thebackports_abc
module on PyPI. See https://bugs.python.org/issue24018 - Adding/subtracting/dividing/modulus and equality comparisons with constant Python floats and small integers are faster.
- Binary and/or/xor/rshift operations with small constant Python integers are faster.
- When called on generator expressions, the builtins
all()
,any()
,dict()
,list()
,set()
,sorted()
andunicode.join()
avoid the generator iteration overhead by inlining a part of their functionality into the for-loop. - Keyword argument dicts are no longer copied on function entry when they are not being used or only passed through to other function calls (e.g. in wrapper functions).
- The
PyTypeObject
declaration incpython.object
was extended. - The builtin
type
type is now declared as PyTypeObject in source, allowing for extern functions taking type parameters to have the correct C signatures. Note that this might break code that usestype
just for passing around Python types in typed variables. Removing the type declaration provides a backwards compatible fix. wraparound()
andboundscheck()
are available as no-ops in pure Python mode.- Const iterators were added to the provided C++ STL declarations.
- Smart pointers were added to the provided C++ STL declarations. Patch by Daniel Filonik.
NULL
is allowed as default argument when embedding signatures. This fixes ticket 843.- When compiling with
--embed
, the internal module name is changed to__main__
to allow arbitrary program names, including those that would be invalid for modules. Note that this prevents reuse of the generated C code as an importable module. - External C++ classes that overload the assignment operator can be used. Patch by Ian Henriksen.
- Support operator bool() for C++ classes so they can be used in if statements.
- Calling "yield from" from Python on a Cython generator that returned a value triggered a crash in CPython. This is now being worked around. See https://bugs.python.org/issue23996
- Language level 3 did not enable true division (a.k.a. float division) for integer operands.
- Functions with fused argument types that included a generic 'object' fallback could end up using that fallback also for other explicitly listed object types.
- Relative cimports could accidentally fall back to trying an absolute cimport on failure.
- The result of calling a C struct constructor no longer requires an intermediate assignment when coercing to a Python dict.
- C++ exception declarations with mapping functions could fail to compile when pre-declared in .pxd files.
cpdef void
methods are now permitted.abs(cint)
could fail to compile in MSVC and used sub-optimal code in C++. Patch by David Vierra, original patch by Michael Enßlin.- Buffer index calculations using index variables with small C integer types could overflow for large buffer sizes. Original patch by David Vierra.
- C unions use a saner way to coerce from and to Python dicts.
- When compiling a module
foo.pyx
, the directories insys.path
are no longer searched when looking forfoo.pxd
. Patch by Jeroen Demeyer. - Memory leaks in the embedding main function were fixed. Original patch by Michael Enßlin.
- Some complex Python expressions could fail to compile inside of finally clauses.
- Unprefixed 'str' literals were not supported as C varargs arguments.
- Fixed type errors in conversion enum types to/from Python. Note that this imposes stricter correctness requirements on enum declarations.
- Changed mangling scheme in header files generated by
cdef api
declarations. - Installation under CPython 3.3+ no longer requires a pass of the 2to3 tool. This also makes it possible to run Cython in Python 3.3+ from a source checkout without installing it first. Patch by Petr Viktorin.
jedi-typer.py
(inTools/
) was extended and renamed tojedityper.py
(to make it importable) and now works with and requires Jedi 0.9. Patch by Tzer-jen Wei.
- Crash when returning values on generator termination.
- In some cases, exceptions raised during internal isinstance() checks were not propagated.
- Runtime reported file paths of source files (e.g for profiling and tracing) are now relative to the build root directory instead of the main source file.
- Tracing exception handling code could enter the trace function with an active exception set.
- The internal generator function type was not shared across modules.
- Comparisons of (inferred) ctuples failed to compile.
- Closures inside of cdef functions returning
void
failed to compile. - Using
const
C++ references in intermediate parts of longer expressions could fail to compile. - C++ exception declarations with mapping functions could fail to compile when pre-declared in .pxd files.
- C++ compilation could fail with an ambiguity error in recent MacOS-X Xcode versions.
- C compilation could fail in pypy3.
- Fixed a memory leak in the compiler when compiling multiple modules.
- When compiling multiple modules, external library dependencies could leak into later compiler runs. Fix by Jeroen Demeyer. This fixes ticket 845.
- C functions can coerce to Python functions, which allows passing them around as callable objects.
- C arrays can be assigned by value and auto-coerce from Python iterables and to Python lists (and tuples).
- Extern C functions can now be declared as cpdef to export them to the module's Python namespace. Extern C functions in pxd files export their values to their own module, iff it exists.
- Anonymous C tuple types can be declared as (ctype1, ctype2, ...).
- PEP 479: turn accidental StopIteration exceptions that exit generators into a RuntimeError, activated with future import "generator_stop".
- Looping over
reversed(range())
is optimised in the same way asrange()
. Patch by Favian Contreras.
- Mismatching 'except' declarations on signatures in .pxd and .pyx files failed to produce a compile error.
- Failure to find any files for the path pattern(s) passed into
cythonize()
is now an error to more easily detect accidental typos. - The
logaddexp
family of functions innumpy.math
now has correct declarations. - In Py2.6/7 and Py3.2, simple Cython memory views could accidentally be interpreted as non-contiguous by CPython, which could trigger a CPython bug when copying data from them, thus leading to data corruption. See CPython issues 12834 and 23349.
- Preliminary support for defining the Cython language with a formal grammar. To try parsing your files against this grammar, use the --formal_grammar directive. Experimental.
_
is no longer considered a cacheable builtin as it could interfere with gettext.- Cythonize-computed metadata now cached in the generated C files.
- Several corrections and extensions in numpy, cpython, and libcpp pxd files.
- Crash when assigning a C value to both a Python and C target at the same time.
- Automatic coercion from C++ strings to
str
generated incomplete code that failed to compile. - Declaring a constructor in a C++ child class erroneously required a default constructor declaration in the super class.
resize_smart()
incpython.array
was broken.- Functions in
libcpp.cast
are now declared asnogil
. - Some missing C-API declarations were added.
- Py3 main code in embedding program code was lacking casts.
- Exception related to distutils "Distribution" class type in pyximport under latest CPython 2.7 and 3.4 releases when setuptools is being imported later.
- New
cythonize
option-a
to generate the annotated HTML source view. - Missing C-API declarations in
cpython.unicode
were added. - Passing
language='c++'
into cythonize() globally enables C++ mode for all modules that were not passed as Extension objects (i.e. only source files and file patterns). Py_hash_t
is a known type (used in CPython for hash values).PySlice_*()
C-API functions are available from thecpython.slice
module.- Allow arrays of C++ classes.
- Reference leak for non-simple Python expressions in boolean and/or expressions.
- To fix a name collision and to reflect availability on host platforms, standard C declarations [ clock(), time(), struct tm and tm* functions ] were moved from posix/time.pxd to a new libc/time.pxd. Patch by Charles Blake.
- Rerunning unmodified modules in IPython's cython support failed. Patch by Matthias Bussonier.
- Casting C++
std::string
to Python byte strings failed when auto-decoding was enabled. - Fatal exceptions in global module init code could lead to crashes if the already created module was used later on (e.g. through a stale reference in sys.modules or elsewhere).
cythonize.py
script was not installed on MS-Windows.
- Compilation no longer fails hard when unknown compilation options are passed. Instead, it raises a warning and ignores them (as it did silently before 0.21). This will be changed back to an error in a future release.
- C (cdef) functions allow inner Python functions.
- Enums can now be declared as cpdef to export their values to the module's Python namespace. Cpdef enums in pxd files export their values to their own module, iff it exists.
- Allow @staticmethod decorator to declare static cdef methods. This is especially useful for declaring "constructors" for cdef classes that can take non-Python arguments.
- Taking a
char*
from a temporary Python string object is safer in more cases and can be done inside of non-trivial expressions, including arguments of a function call. A compile time error is raised only when such a pointer is assigned to a variable and would thus exceed the lifetime of the string itself. - Generators have new properties
__name__
and__qualname__
that provide the plain/qualified name of the generator function (following CPython 3.5). See https://bugs.python.org/issue21205 - The
inline
function modifier is available as a decorator@cython.inline
in pure mode. - When cygdb is run in a virtualenv, it enables the same virtualenv inside of the debugger. Patch by Marc Abramowitz.
- PEP 465: dedicated infix operator for matrix multiplication (A @ B).
- HTML output of annotated code uses Pygments for code highlighting and generally received a major overhaul by Matthias Bussonier.
- IPython magic support is now available directly from Cython with the command "%load_ext cython". Cython code can directly be executed in a cell when marked with "%%cython". Code analysis is available with "%%cython -a". Patch by Martín Gaitán.
- Simple support for declaring Python object types in Python signature
annotations. Currently requires setting the compiler directive
annotation_typing=True
. - New directive
use_switch
(defaults to True) to optionally disable the optimization of chained if statement to C switch statements. - Defines dynamic_cast et al. in
libcpp.cast
and C++ heap data structure operations inlibcpp.algorithm
. - Shipped header declarations in
posix.*
were extended to cover more of the POSIX API. Patches by Lars Buitinck and Mark Peek.
- Simple calls to C implemented Python functions/methods are faster. This also speeds up many operations on builtins that Cython cannot otherwise optimise.
- The "and"/"or" operators try to avoid unnecessary coercions of their arguments. They now evaluate the truth value of each argument independently and only coerce the final result of the whole expression to the target type (e.g. the type on the left side of an assignment). This also avoids reference counting overhead for Python values during evaluation and generally improves the code flow in the generated C code.
- The Python expression "2 ** N" is optimised into bit shifting. See https://bugs.python.org/issue21420
- Cascaded assignments (a = b = ...) try to minimise the number of type coercions.
- Calls to
slice()
are translated to a straight C-API call.
- Crash when assigning memory views from ternary conditional expressions.
- Nested C++ templates could lead to unseparated ">>" characters being generated into the C++ declarations, which older C++ compilers could not parse.
- Sending SIGINT (Ctrl-C) during parallel cythonize() builds could hang the child processes.
- No longer ignore local setup.cfg files for distutils in pyximport. Patch by Martin Teichmann.
- Taking a
char*
from an indexed Python string generated unsafe reference counting code. - Set literals now create all of their items before trying to add them to the set, following the behaviour in CPython. This makes a difference in the rare case that the item creation has side effects and some items are not hashable (or if hashing them has side effects, too).
- Cython no longer generates the cross product of C functions for code
that uses memory views of fused types in function signatures (e.g.
cdef func(floating[:] a, floating[:] b)
). This is considered the expected behaviour by most users and was previously inconsistent with other structured types like C arrays. Code that really wants all type combinations can create the same fused memoryview type under different names and use those in the signature to make it clear which types are independent. - Names that were unknown at compile time were looked up as builtins at runtime but not as global module names. Trying both lookups helps with globals() manipulation.
- Fixed stl container conversion for typedef element types.
obj.pop(x)
truncated large C integer values of x toPy_ssize_t
.__init__.pyc
is recognised as marking a package directory (in addition to .py, .pyx and .pxd).- Syntax highlighting in
cython-mode.el
for Emacs no longer incorrectly highlights keywords found as part of longer names. - Correctly handle
from cython.submodule cimport name
. - Fix infinite recursion when using super with cpdef methods.
- No-args
dir()
was not guaranteed to return a sorted list.
- The header line in the generated C files no longer contains the timestamp but only the Cython version that wrote it. This was changed to make builds more reproducible.
- Removed support for CPython 2.4, 2.5 and 3.1.
- The licensing implications on the generated code were clarified to avoid legal constraints for users.
- Some optimisations for set/frozenset instantiation.
- Support for C++ unordered_set and unordered_map.
- Access to attributes of optimised builtin methods (e.g.
[].append.__name__
) could fail to compile. - Memory leak when extension subtypes add a memory view as attribute to those of the parent type without having Python object attributes or a user provided dealloc method.
- Compiler crash on readonly properties in "binding" mode.
- Auto-encoding with
c_string_encoding=ascii
failed in Py3.3. - Crash when subtyping freelist enabled Cython extension types with
Python classes that use
__slots__
. - Freelist usage is restricted to CPython to avoid problems with other Python implementations.
- Memory leak in memory views when copying overlapping, contiguous slices.
- Format checking when requesting non-contiguous buffers from
cython.array
objects was accidentally omitted in Py3. - C++ destructor calls in extension types could fail to compile in clang.
- Buffer format validation failed for sequences of strings in structs.
- Docstrings on extension type attributes in .pxd files were rejected.
- Build error under recent MacOS-X versions where
isspace()
could not be resolved by clang. - List/Tuple literals multiplied by more than one factor were only multiplied by the last factor instead of all.
- Lookups of special methods (specifically for context managers) could fail in Python <= 2.6/3.1.
- Local variables were erroneously appended to the signature introspection of Cython implemented functions with keyword-only arguments under Python 3.
- In-place assignments to variables with inferred Python builtin/extension types could fail with type errors if the result value type was incompatible with the type of the previous value.
- The C code generation order of cdef classes, closures, helper code, etc. was not deterministic, thus leading to high code churn.
- Type inference could fail to deduce C enum types.
- Type inference could deduce unsafe or inefficient types from integer assignments within a mix of inferred Python variables and integer variables.
- Support for CPython 3.4.
- Support for calling C++ template functions.
yield
is supported infinally
clauses.- The C code generated for finally blocks is duplicated for each exit case to allow for better optimisations by the C compiler.
- Cython tries to undo the Python optimisationism of assigning a bound method to a local variable when it can generate better code for the direct call.
- Constant Python float values are cached.
- String equality comparisons can use faster type specific code in more cases than before.
- String/Unicode formatting using the '%' operator uses a faster C-API call.
bytearray
has become a known type and supports coercion from and to C strings. Indexing, slicing and decoding is optimised. Note that this may have an impact on existing code due to type inference.- Using
cdef basestring stringvar
and function arguments typed asbasestring
is now meaningful and allows assigning exactlystr
andunicode
objects, but no subtypes of these types. - Support for the
__debug__
builtin. - Assertions in Cython compiled modules are disabled if the running Python interpreter was started with the "-O" option.
- Some types that Cython provides internally, such as functions and generators, are now shared across modules if more than one Cython implemented module is imported.
- The type inference algorithm works more fine granular by taking the results of the control flow analysis into account.
- A new script in
bin/cythonize
provides a command line frontend to the cythonize() compilation function (including distutils build). - The new extension type decorator
@cython.no_gc_clear
prevents objects from being cleared during cyclic garbage collection, thus making sure that object attributes are kept alive until deallocation. - During cyclic garbage collection, attributes of extension types that cannot create reference cycles due to their type (e.g. strings) are no longer considered for traversal or clearing. This can reduce the processing overhead when searching for or cleaning up reference cycles.
- Package compilation (i.e.
__init__.py
files) now works, starting with Python 3.3. - The cython-mode.el script for Emacs was updated. Patch by Ivan Andrus.
- An option common_utility_include_dir was added to cythonize() to save oft-used utility code once in a separate directory rather than as part of each generated file.
unraisable_tracebacks
directive added to control printing of tracebacks of unraisable exceptions.
- Abstract Python classes that subtyped a Cython extension type failed to raise an exception on instantiation, and thus ended up being instantiated.
set.add(a_tuple)
andset.discard(a_tuple)
failed with a TypeError in Py2.4.- The PEP 3155
__qualname__
was incorrect for nested classes and inner classes/functions declared asglobal
. - Several corner cases in the try-finally statement were fixed.
- The metaclass of a Python class was not inherited from its parent
class(es). It is now extracted from the list of base classes if not
provided explicitly using the Py3
metaclass
keyword argument. In Py2 compilation mode, a__metaclass__
entry in the class dict will still take precedence if not using Py3 metaclass syntax, but only after creating the class dict (which may have been done by a metaclass of a base class, see PEP 3115). It is generally recommended to use the explicit Py3 syntax to define metaclasses for Python types at compile time. - The automatic C switch statement generation behaves more safely for heterogeneous value types (e.g. mixing enum and char), allowing for a slightly wider application and reducing corner cases. It now always generates a 'default' clause to avoid C compiler warnings about unmatched enum values.
- Fixed a bug where class hierarchies declared out-of-order could result in broken generated code.
- Fixed a bug which prevented overriding const methods of C++ classes.
- Fixed a crash when converting Python objects to C++ strings fails.
- In Py3 compilation mode, Python2-style metaclasses declared by a
__metaclass__
class dict entry are ignored. - In Py3.4+, the Cython generator type uses
tp_finalize()
for safer cleanup instead oftp_del()
.
- Some standard declarations were fixed or updated, including the previously
incorrect declaration of
PyBuffer_FillInfo()
and some missing bits inlibc.math
. - Heap allocated subtypes of
type
used the wrong base type struct at the C level. - Calling the unbound method dict.keys/value/items() in dict subtypes could call the bound object method instead of the unbound supertype method.
- "yield" wasn't supported in "return" value expressions.
- Using the "bint" type in memory views lead to unexpected results. It is now an error.
- Assignments to global/closure variables could catch them in an illegal state while deallocating the old value.
- Completely empty C-API structs for extension type slots (protocols like number/mapping/sequence) are no longer generated into the C code.
- Docstrings that directly follow a public/readonly attribute declaration in a cdef class will be used as docstring of the auto-generated property. This fixes ticket 206.
- The automatic signature documentation tries to preserve more semantics
of default arguments and argument types. Specifically,
bint
arguments now appear as typebool
. - A warning is emitted when negative literal indices are found inside of
a code section that disables
wraparound
handling. This helps with fixing invalid code that might fail in the face of future compiler optimisations. - Constant folding for boolean expressions (and/or) was improved.
- Added a build_dir option to cythonize() which allows one to place the generated .c files outside the source tree.
isinstance(X, type)
failed to get optimised into a call toPyType_Check()
, as done for other builtin types.- A spurious
from datetime cimport *
was removed from the "cpython" declaration package. This means that the "datetime" declarations (added in 0.19) are no longer available directly from the "cpython" namespace, but only from "cpython.datetime". This is the correct way of doing it because the declarations refer to a standard library module, not the core CPython C-API itself. - The C code for extension types is now generated in topological order instead of source code order to avoid C compiler errors about missing declarations for subtypes that are defined before their parent.
- The
memoryview
type name no longer shows up in the module dict of modules that use memory views. This fixes trac ticket 775. - Regression in 0.19 that rejected valid C expressions from being used in C array size declarations.
- In C++ mode, the C99-only keyword
restrict
could accidentally be seen by the GNU C++ compiler. It is now specially handled for both GCC and MSVC. - Testing large (> int) C integer values for their truth value could fail due to integer wrap-around.
- New directives
c_string_type
andc_string_encoding
to more easily and automatically convert between C strings and the different Python string types. - The extension type flag
Py_TPFLAGS_HAVE_VERSION_TAG
is enabled by default on extension types and can be disabled using thetype_version_tag
compiler directive. - EXPERIMENTAL support for simple Cython code level line tracing. Enabled by the "linetrace" compiler directive.
- Cython implemented functions make their argument and return type annotations
available through the
__annotations__
attribute (PEP 3107). - Access to non-cdef module globals and Python object attributes is faster.
Py_UNICODE*
coerces from and to Python unicode strings. This is helpful when talking to Windows APIs, which use compatible wchar_t arrays for strings. Note that thePy_UNICODE
type is otherwise deprecated as of CPython 3.3.isinstance(obj, basestring)
is optimised. In Python 3 it only tests for instances ofstr
(i.e. Py2unicode
).- The
basestring
builtin is mapped tostr
(i.e. Py2unicode
) when compiling the generated C code under Python 3. - Closures use freelists, which can speed up their creation quite substantially. This is also visible for short running generator expressions, for example.
- A new class decorator
@cython.freelist(N)
creates a static freelist of N instances for an extension type, thus avoiding the costly allocation step if possible. This can speed up object instantiation by 20-30% in suitable scenarios. Note that freelists are currently only supported for base types, not for types that inherit from others. - Fast extension type instantiation using the
Type.__new__(Type)
idiom has gained support for passing arguments. It is also a bit faster for types defined inside of the module. - The Python2-only dict methods
.iter*()
and.view*()
(requires Python 2.7) are automatically mapped to the equivalent keys/values/items methods in Python 3 for typed dictionaries. - Slicing unicode strings, lists and tuples is faster.
- list.append() is faster on average.
raise Exception() from None
suppresses the exception context in Py3.3.- Py3 compatible
exec(tuple)
syntax is supported in Py2 code. - Keyword arguments are supported for cdef functions.
- External C++ classes can be declared nogil. Patch by John Stumpo. This fixes trac ticket 805.
- 2-value slicing of unknown objects passes the correct slice when the
getitem
protocol is used instead of thegetslice
protocol (especially in Python 3), i.e.None
values for missing bounds instead of[0,maxsize]
. It is also a bit faster in some cases, e.g. for constant bounds. This fixes trac ticket 636. - Cascaded assignments of None values to extension type variables failed with
a
TypeError
at runtime. - The
__defaults__
attribute was not writable for Cython implemented functions. - Default values of keyword-only arguments showed up in
__defaults__
instead of__kwdefaults__
(which was not implemented). Both are available for Cython implemented functions now, as specified in Python 3.x. yield
works inside ofwith gil
sections. It previously lead to a crash. This fixes trac ticket 803.- Static methods without explicitly named positional arguments (e.g. having only
*args
) crashed when being called. This fixes trac ticket 804. dir()
without arguments previously returned an unsorted list, which now gets sorted as expected.dict.items()
,dict.keys()
anddict.values()
no longer return lists in Python 3.- Exiting from an
except-as
clause now deletes the exception in Python 3 mode. - The declarations of
frexp()
andldexp()
inmath.pxd
were incorrect.
- Named Unicode escapes ("N{...}") are supported.
- Python functions/classes provide the special attribute "__qualname__" as defined by PEP 3155.
- Added a directive
overflowcheck
which raises an OverflowException when arithmetic with C ints overflow. This has a modest performance penalty, but is much faster than using Python ints. - Calls to nested Python functions are resolved at compile time.
- Type inference works across nested functions.
py_bytes_string.decode(...)
is optimised.- C
const
declarations are supported in the language.
- Automatic C++ exception mapping didn't work in nogil functions (only in "with nogil" blocks).
- Garbage collection triggered during deallocation of container classes could lead to a double-deallocation.
- During final interpreter cleanup (with types cleanup enabled at compile time), extension types that inherit from base types over more than one level that were cimported from other modules could lead to a crash.
- Weak-reference support in extension types (with a
cdef __weakref__
attribute) generated incorrect deallocation code. - In CPython 3.3, converting a Unicode character to the Py_UNICODE type could fail to raise an overflow for non-BMP characters that do not fit into a wchar_t on the current platform.
- Negative C integer constants lost their longness suffix in the generated C code.
cythonize()
gained a best effort compile mode that can be used to simply ignore .py files that fail to compile.
- Replacing an object reference with the value of one of its cdef attributes could generate incorrect C code that accessed the object after deleting its last reference.
- C-to-Python type coercions during cascaded comparisons could generate invalid C code, specifically when using the 'in' operator.
- "obj[1,]" passed a single integer into the item getter instead of a tuple.
- Cyclic imports at module init time did not work in Py3.
- The names of C++ destructors for template classes were built incorrectly.
- In pure mode, type casts in Cython syntax and the C ampersand operator are now rejected. Use the pure mode replacements instead.
- In pure mode, C type names and the sizeof() function are no longer recognised as such and can be used as normal Python names.
- The extended C level support for the CPython array type was declared too late to be used by user defined classes.
- C++ class nesting was broken.
- Better checking for required nullary constructors for stack-allocated C++ instances.
- Remove module docstring in no-docstring mode.
- Fix specialization for varargs function signatures.
- Fix several compiler crashes.
- An experimental distutils script for compiling the CPython standard library was added as Tools/cystdlib.py.
- A reference leak was fixed in the new dict iteration code when the loop target was not a plain variable but an unpacked tuple.
- Memory views did not handle the special case of a NULL buffer strides value, as allowed by PEP3118.
- Alpha quality support for compiling and running Cython generated extension modules in PyPy (through cpyext). Note that this requires at least PyPy 1.9 and in many cases also adaptations in user code, especially to avoid borrowed references when no owned reference is being held directly in C space (a reference in a Python list or dict is not enough, for example). See the documentation on porting Cython code to PyPy.
- "yield from" is supported (PEP 380) and a couple of minor problems with generators were fixed.
- C++ STL container classes automatically coerce from and to the equivalent Python container types on typed assignments and casts. Note that the data in the containers is copied during this conversion.
- C++ iterators can now be iterated over using "for x in cpp_container" whenever cpp_container has begin() and end() methods returning objects satisfying the iterator pattern (that is, it can be incremented, dereferenced, and compared (for non-equality)).
- cdef classes can now have C++ class members (provided a zero-argument constructor exists)
- A new cpython.array standard cimport file allows to efficiently talk to the stdlib array.array data type in Python 2. Since CPython does not export an official C-API for this module, it receives special casing by the compiler in order to avoid setup overhead on user side. In Python 3, both buffers and memory views on the array type already worked out of the box with earlier versions of Cython due to the native support for the buffer interface in the Py3 array module.
- Fast dict iteration is now enabled optimistically also for untyped variables when the common iteration methods are used.
- The unicode string processing code was adapted for the upcoming CPython 3.3 (PEP 393, new Unicode buffer layout).
- Buffer arguments and memory view arguments in Python functions can be declared "not None" to raise a TypeError on None input.
- c(p)def functions in pure mode can specify their return type with "@cython.returns()".
- Automatic dispatch for fused functions with memoryview arguments
- Support newaxis indexing for memoryviews
- Support decorators for fused functions
- Old-style Py2 imports did not work reliably in Python 3.x and were broken in Python 3.3. Regardless of this fix, it's generally best to be explicit about relative and global imports in Cython code because old-style imports have a higher overhead. To this end, "from __future__ import absolute_import" is supported in Python/Cython 2.x code now (previous versions of Cython already used it when compiling Python 3 code).
- Stricter constraints on the "inline" and "final" modifiers. If your code does not compile due to this change, chances are these modifiers were previously being ignored by the compiler and can be removed without any performance regression.
- Exceptions are always instantiated while raising them (as in Python), instead of risking to instantiate them in potentially unsafe situations when they need to be handled or otherwise processed.
- locals() properly ignores names that do not have Python compatible types (including automatically inferred types).
- Some garbage collection issues of memory views were fixed.
- numpy.pxd compiles in Python 3 mode.
- Several C compiler warnings were fixed.
- Several bugs related to memoryviews and fused types were fixed.
- Several bug-fixes and improvements related to cythonize(), including ccache-style caching.
- libc.string provides a convenience declaration for const uchar in addition to const char.
- User declared char* types are now recognised as such and auto-coerce to and from Python bytes strings.
- callable() and next() compile to more efficient C code.
- list.append() is faster on average.
- Modules generated by @cython.inline() are written into the directory pointed to by the environment variable CYTHON_CACHE_DIR if set.
- Enhancements to Cython's function type (support for weak references, default arguments, code objects, dynamic attributes, classmethods, staticmethods, and more)
- Fused Types - Template-like support for functions and methods CEP 522 (docs)
- Typed views on memory - Support for efficient direct and indirect buffers (indexing, slicing, transposing, ...) CEP 517 (docs)
- super() without arguments
- Final cdef methods (which translate into direct calls on known instances)
- fix alignment handling for record types in buffer support
- support default arguments for closures
- search sys.path for pxd files
- support C++ template casting
- faster traceback building and faster generator termination
- support inplace operators on indexed buffers
- allow nested prange sections
- Generators (yield) - Cython has full support for generators, generator expressions and PEP 342 coroutines.
- The nonlocal keyword is supported.
- Re-acquiring the gil: with gil - works as expected within a nogil context.
- OpenMP support: prange.
- Control flow analysis prunes dead code and emits warnings and errors about uninitialised variables.
- Debugger command cy set to assign values of expressions to Cython variables and cy exec counterpart $cy_eval().
- Exception chaining PEP 3134.
- Relative imports PEP 328.
- Improved pure syntax including cython.cclass, cython.cfunc, and cython.ccall.
- The with statement has its own dedicated and faster C implementation.
- Support for del.
- Boundschecking directives implemented for builtin Python sequence types.
- Several updates and additions to the shipped standard library .pxd files.
- Forward declaration of types is no longer required for circular references.
- Uninitialized variables are no longer initialized to None and accessing them has the same semantics as standard Python.
- globals() now returns a read-only dict of the Cython module's globals, rather than the globals of the first non-Cython module in the stack
- Many C++ exceptions are now special cased to give closer Python counterparts. This means that except+ functions that formerly raised generic RuntimeErrors may raise something else such as ArithmeticError.
- The inlined generator expressions (introduced in Cython 0.13) were disabled in favour of full generator expression support. This breaks code that previously used them inside of cdef functions (usage in def functions continues to work) and induces a performance regression for cases that continue to work but that were previously inlined. We hope to reinstate this feature in the near future.
- The gdb debugging support was extended to include all major Cython features, including closures.
- raise MemoryError() is now safe to use as Cython replaces it with the correct C-API call.
- Decorators on special methods of cdef classes now raise a compile time error rather than being ignored.
- In Python 3 language level mode (-3 option), the 'str' type is now mapped to 'unicode', so that cdef str s declares a Unicode string even when running in Python 2.
- Python classes can now be nested and receive a proper closure at definition time.
- Redefinition is supported for Python functions, even within the same scope.
- Lambda expressions are supported in class bodies and at the module level.
- Metaclasses are supported for Python classes, both in Python 2 and Python 3 syntax. The Python 3 syntax (using a keyword argument in the type declaration) is preferred and optimised at compile time.
- "final" extension classes prevent inheritance in Python space. This feature is available through the new "cython.final" decorator. In the future, these classes may receive further optimisations.
- "internal" extension classes do not show up in the module dictionary. This feature is available through the new "cython.internal" decorator.
- Extension type inheritance from builtin types, such as "cdef class MyUnicode(unicode)", now works without further external type redeclarations (which are also strongly discouraged now and continue to issue a warning).
- GDB support. http://docs.cython.org/src/userguide/debugging.html
- A new build system with support for inline distutils directives, correct dependency tracking, and parallel compilation. https://github.com/cython/cython/wiki/enhancements-distutils_preprocessing
- Support for dynamic compilation at runtime via the new cython.inline function and cython.compile decorator. https://github.com/cython/cython/wiki/enhancements-inline
- "nogil" blocks are supported when compiling pure Python code by writing "with cython.nogil".
- Iterating over arbitrary pointer types is now supported, as is an optimized version of the in operator, e.g. x in ptr[a:b].
- In parallel assignments, the right side was evaluated in reverse order in 0.13. This could result in errors if it had side effects (e.g. function calls).
- In some cases, methods of builtin types would raise a SystemError instead of an AttributeError when called on None.
- Constant tuples are now cached over the lifetime of an extension module, just like CPython does. Constant argument tuples of Python function calls are also cached.
- Closures have tightened to include exactly the names used in the inner functions and classes. Previously, they held the complete locals of the defining function.
- The builtin "next()" function in Python 2.6 and later is now implemented internally and therefore available in all Python versions. This makes it the preferred and portable way of manually advancing an iterator.
- In addition to the previously supported inlined generator expressions in 0.13, "sorted(genexpr)" can now be used as well. Typing issues were fixed in "sum(genexpr)" that could lead to invalid C code being generated. Other known issues with inlined generator expressions were also fixed that make upgrading to 0.14 a strong recommendation for code that uses them. Note that general generators and generator expressions continue to be not supported.
- Inplace arithmetic operators now respect the cdivision directive and are supported for complex types.
- Typing a variable as type "complex" previously gave it the Python object type. It now uses the appropriate C/C++ double complex type. A side-effect is that assignments and typed function parameters now accept anything that Python can coerce to a complex, including integers and floats, and not only complex instances.
- Large integer literals pass through the compiler in a safer way. To prevent truncation in C code, non 32-bit literals are turned into Python objects if not used in a C context. This context can either be given by a clear C literal suffix such as "UL" or "LL" (or "L" in Python 3 code), or it can be an assignment to a typed variable or a typed function argument, in which case it is up to the user to take care of a sufficiently large value space of the target.
- Python functions are declared in the order they appear in the file, rather than all being created at module creation time. This is consistent with Python and needed to support, for example, conditional or repeated declarations of functions. In the face of circular imports this may cause code to break, so a new --disable-function-redefinition flag was added to revert to the old behavior. This flag will be removed in a future release, so should only be used as a stopgap until old code can be fixed.
- Closures are fully supported for Python functions. Cython supports inner functions and lambda expressions. Generators and generator expressions are not supported in this release.
- Proper C++ support. Cython knows about C++ classes, templates and overloaded function signatures, so that Cython code can interact with them in a straight forward way.
- Type inference is enabled by default for safe C types (e.g. double, bint, C++ classes) and known extension types. This reduces the need for explicit type declarations and can improve the performance of untyped code in some cases. There is also a verbose compile mode for testing the impact on user code.
- Cython's for-in-loop can iterate over C arrays and sliced pointers. The type of the loop variable will be inferred automatically in this case.
- The Py_UNICODE integer type for Unicode code points is fully supported, including for-loops and 'in' tests on unicode strings. It coerces from and to single character unicode strings. Note that untyped for-loop variables will automatically be inferred as Py_UNICODE when iterating over a unicode string. In most cases, this will be much more efficient than yielding sliced string objects, but can also have a negative performance impact when the variable is used in a Python context multiple times, so that it needs to coerce to a unicode string object more than once. If this happens, typing the loop variable as unicode or object will help.
- The built-in functions any(), all(), sum(), list(), set() and dict() are inlined as plain for loops when called on generator expressions. Note that generator expressions are not generally supported apart from this feature. Also, tuple(genexpr) is not currently supported - use tuple([listcomp]) instead.
- More shipped standard library declarations. The python_* and stdlib/stdio .pxd files have been deprecated in favor of clib.* and cpython[.*] and may get removed in a future release.
- Pure Python mode no longer disallows non-Python keywords like 'cdef', 'include' or 'cimport'. It also no longer recognises syntax extensions like the for-from loop.
- Parsing has improved for Python 3 syntax in Python code, although not all features are correctly supported. The missing Python 3 features are being worked on for the next release.
- from __future__ import print_function is supported in Python 2.6 and later. Note that there is currently no emulation for earlier Python versions, so code that uses print() with this future import will require at least Python 2.6.
- New compiler directive language_level (valid values: 2 or 3) with corresponding command line options -2 and -3 requests source code compatibility with Python 2.x or Python 3.x respectively. Language level 3 currently enforces unicode literals for unprefixed string literals, enables the print function (requires Python 2.6 or later) and keeps loop variables in list comprehensions from leaking.
- Loop variables in set/dict comprehensions no longer leak into the surrounding scope (following Python 2.7). List comprehensions are unchanged in language level 2.
- print >> stream
The availability of type inference by default means that Cython will also infer the type of pointers on assignments. Previously, code like this:
cdef char* s = ... untyped_variable = s
would convert the char* to a Python bytes string and assign that. This is no longer the case and no coercion will happen in the example above. The correct way of doing this is through an explicit cast or by typing the target variable, i.e.
cdef char* s = ... untyped_variable1 = <bytes>s untyped_variable2 = <object>s cdef object py_object = s cdef bytes bytes_string = s
bool is no longer a valid type name by default. The problem is that it's not clear whether bool should refer to the Python type or the C++ type, and expecting one and finding the other has already led to several hard-to-find bugs. Both types are available for importing: you can use from cpython cimport bool for the Python bool type, and from libcpp cimport bool for the C++ type. bool is still a valid object by default, so one can still write bool(x).
__getsegcount__
is now correctly typed to take aPy_size_t*
rather than anint*
.
- Type inference improvements.
- There have been several bug fixes and improvements to the type inferencer.
- Notably, there is now a "safe" mode enabled by setting the infer_types directive to None. (The None here refers to the "default" mode, which will be the default in 0.13.) This safe mode limits inference to Python object types and C doubles, which should speed up execution without affecting any semantics such as integer overflow behavior like infer_types=True might. There is also an infer_types.verbose option which allows one to see what types are inferred.
- The boundscheck directive works for lists and tuples as well as buffers.
- len(s) and s.decode("encoding") are efficiently supported for char* s.
- Cython's INLINE macro has been renamed to CYTHON_INLINE to reduce conflict and has better support for the MSVC compiler on Windows. It is no longer clobbered if externally defined.
- Revision history is now omitted from the source package, resulting in a 85% size reduction. Running make repo will download the history and turn the directory into a complete Mercurial working repository.
- Cython modules don't need to be recompiled when the size of an external type grows. (A warning, rather than an error, is produced.) This should be helpful for binary distributions relying on NumPy.
- Several other bugs and minor improvements have been made. This release should be fully backwards compatible with 0.12.
- Type inference with the infer_types directive
- Seamless C++ complex support
- Fast extension type instantiation using the normal Python meme obj = MyType.__new__(MyType)
- Improved support for Py3.1
- Cython now runs under Python 3.x using the 2to3 tool
- unittest support for doctests in Cython modules
- Optimised handling of C strings (char*): for c in cstring[2:50] and cstring.decode()
- Looping over c pointers: for i in intptr[:50].
- pyximport improvements
- cython_freeze improvements
- Many bug fixes
- Many other optimisation, e.g. enumerate() loops, parallel swap assignments (a,b = b,a), and unicode.encode()
- More complete numpy.pxd
There's now native complex floating point support! C99 complex will be used if complex.h is included, otherwise explicit complex arithmetic working on all C compilers is used. [Robert Bradshaw]
cdef double complex a = 1 + 0.3j cdef np.ndarray[np.complex128_t, ndim=2] arr = \ np.zeros(10, np.complex128)
Cython can now generate a main()-method for embedding of the Python interpreter into an executable (see #289) [Robert Bradshaw]
@wraparound directive (another way to disable arr[idx] for negative idx) [Dag Sverre Seljebotn]
Correct support for NumPy record dtypes with different alignments, and "cdef packed struct" support [Dag Sverre Seljebotn]
@callspec directive, allowing custom calling convention macros [Lisandro Dalcin]
- Bug fixes and smaller improvements. For the full list, see [1].