Changed some uses of Finalize as a static method name that confused the Mono compiler and people reading the code. Note that this may be a breaking change if anyone was calling
PythonEngine.Finalize(). If so, you should now use
Tweaked assembly lookup to ensure that assemblies can be found in the current working directory, even after changing directories using things like
🛠 Fixed some incorrect finalizers (thanks to Greg Chapman for the report) that may have caused some threading oddities.
👍 Tweaked support for out and ref parameters. If a method has a return type of void and a single ref or out parameter, that parameter will be returned as the result of the method. This matches the current behavior of IronPython and makes it more likely that code can be moved between Python for .NET and IP in the future.
🔨 Refactored part of the assembly manager to remove a potential case of thread-deadlock in multi-threaded applications.
__str__method to managed exceptions that returns the Message attribute of the exception and the StackTrace (if available).
💅 Implemented a workaround for the fact that exceptions cannot be new-style classes in the CPython interpreter. Managed exceptions can now be raised and caught naturally from Python (hooray!)
👍 Implemented support for invoking methods with out and ref parameters. Because there is no real equivalent to these in Python, methods that have out or ref parameters will return a tuple. The tuple will contain the result of the method as its first item, followed by out parameter values in the order of their declaration in the method signature.
🛠 Fixed a refcount problem that caused a crash when CLR was imported in an existing installed Python interpreter.
Added an automatic conversion from Python strings to
byte. This makes it easier to pass
bytedata to managed methods (or set properties, etc.) as a Python string without having to write explicit conversion code. Also works for sbyte arrays. Note that
sbytearrays returned from managed methods or obtained from properties or fields do not get converted to Python strings - they remain instances of
Added conversion of generic Python sequences to object arrays when appropriate (thanks to Mackenzie Straight for the patch).
📚 Added a bit of cautionary documentation for embedders, focused on correct handling of the Python global interpreter lock from managed code for code that calls into Python.
PyObject.FromManagedObjectnow correctly returns the Python None object if the input is a null reference. Also added a new
PyObject, making it easier to convert a Python-wrapped managed object to the real managed object.
🏁 Created a simple installer for windows platforms.
🔨 Refactored and fixed threading and global interpreter lock handling, which was badly broken before. Also added a number of threading and GIL-handling tests.
🛠 Related to the GIL fixes, added a note to embedders in the README about using the AcquireLock and ReleaseLock methods of the PythonEngine class to manage the GIL.
🛠 Fixed a problem in
Single <--> floatconversion for cultures that use different decimal symbols than Python.
Added a new
ReloadModulemethod to the
PythonEngineclass that hooks Python module reloading (
Added a new
StringAsModulemethod to the PythonEngine class that can create a module from a managed string of code.
Added a default
__str__implementation for Python wrappers of managed objects that calls the
ToStringmethod of the managed object.
🛠 Fixed a problem that made it impossible to override "special" methods like
__getitem__in subclasses of managed classes. Now the tests all pass, and there is much rejoicing.
Managed classes reflected to Python now have an
__doc__attribute that contains a listing of the class constructor signatures.
🛠 Fixed a problem that prevented passing null (None) for array arguments.
✅ Added a number of new argument conversion tests. Thanks to Laurent Caumont for giving Python for .NET a good workout with managed DirectX.
⚡️ Updated the bundled C Python runtime and libraries to Python 2.4. The current release is known to also run with Python 2.3. It is known not to work with older versions due to changes in CPython type object structure.
🛠 Mostly fixed the differences in the way that import works depending on whether you are using the bundled interpreter or an existing Python interpreter. The hack I used makes import work uniformly for imports done in Python modules. Unfortunately, there is still a limitation when using the interpreter interactively: you need to do
import CLRfirst before importing any sub-names when running with an existing Python interpreter.
The reason is that the first import of
CLRinstalls the CLR import hook, but for an existing interpreter the standard importer is still in control for the duration of that first import, so sub-names won't be found until the next import, which will use the now-installed hook.
👍 Added support to directly iterate over objects that support IEnumerator (as well as IEnumerable). Thanks to Greg Chapman for prodding me ;)
Added a section to the README dealing with rebuilding Python for .NET against other CPython versions.
🛠 Fixed a problem with accessing properties when only the interface for an object is known. For example,
ICollection(ob).Countfailed because Python for .NET mistakenly decided that Count was abstract.
🛠 Fixed some problems with how COM-based objects are exposed and how members of inherited interfaces are exposed. Thanks to Bruce Dodson for patches on this.
Changed the Runtime class to use a const string to target the appropriate CPython dll in DllImport attributes. Now you only have to change one line to target a new Python version.
- A dumb bug that could cause a crash on startup on some platforms was fixed. Decided to update the beta for this, as a number of people were running into the problem.
Exceptions raised as a result of getting or setting properties were not very helpful (target invokation exception). This has been changed to pass through the inner exception the way that methods do, which is much more likely to be the real exception that caused the problem.
🔨 Events were refactored as the implementation was based on some bad assumptions. As a result, subscription and unsubscription now works correctly. A change from beta 1 is that event objects are no longer directly callable - this was not appropriate, since the internal implementation of an event is private and cant work reliably. Instead, you should the appropriate
OnSomeEventmethod published by a class to fire an event.
The distribution did not include the key file, making it a pain for people to build from source. Added the key file to the distribution buildout for beta 2.
Assemblies can now be found and loaded if they are on the PYTHONPATH. Previously only the appbase and the GAC were checked. The system now checks PYTHONPATH first, then the appbase, then the GAC.
🛠 Fixed a bug in constructor invokation during object instantiation.
Added the baseline of the managed embedding API. Some of the details are still subject to change based on some real-world use and feedback.
The embedding API is based on the
PyObjectclass, along with a number of specific
PyList, (etc.) classes that expose the respective interfaces of the built-in Python types. The basic structure and usage is intended be familar to anyone who has used Python / C++ wrapper libraries like CXX or Boost.
✅ Started integrating NUnit2 to support unit tests for the embedding layer - still need to add the embedding tests (many already exist, but were written for an older version of NUnit).
👍 Added Python iteration protocol support for arrays and managed objects that implement IEnumerable. This means that you can now use the Python idiom
for item in object:on any array or IEnumerable object.
Added automatic conversion from Python sequence types to managed array types. This means, for example, that you can now call a managed method like AddRange that expects an array with any Python object that supports the Python sequence protocol, provided the items of the sequence are convertible to the item type of the managed array.
Added new demo scripts, mostly more substantial winforms examples.
✅ Finished the unit tests for event support, and fixed lots of problems with events and delegates as a result. This is one of the trickier parts of the integration layer, and there is good coverage of these in the unit tests now.
Did a fair amount of profiling with an eval version of ANTS (which is quite nice, BTW) and made a few changes as a result.
🔨 Type management was refactored, fixing the issue that caused segfaults when GC was enabled. Unit tests, stress tests and demo apps now all run nicely with Python GC enabled. There are one or two things left to fix, but the fixes should not have any user impact.
Changed to base PythonNet on Python 2.3.2. This is considered the most stable release, and a good 25 - 30% faster as well.
Added a new
CLR.dllthat acts as an extension module that allows an existing unmodified Python 2.3 installation to simply
import CLRto bootstrap the managed integration layer.
A bug was causing managed methods to only expose overloads declared in a particular class, hiding inherited overloads of the same name. Fixed the bug and added some unit tests.
Added a virtual
__doc__attribute to managed methods that contains the signature of the method. This also means that the Python
helpfunction now provides signature info when used on a managed class.
Calling managed methods and events
unbound(passing the instance as the first argument) now works. There is a caveat for methods - if a class declares both static and instance methods with the same name, it is not possible to call that instance method unbound (the static method will always be called).
👍 Overload selection for overloaded methods is now much better and uses a method resolution algorithm similar to that used by Jython.
Changed the managed python.exe wrapper to run as an STA thread, which seems to be more compatible with winforms apps. This needs a better solution long-term. One possibility would be a command line switch so that -sta or -mta could control the python.exe apartment state.
👍 Added support for the Python boolean type (True, False). Bool values now appear as True or False to Python.
✅ Added a Mono makefile. Thanks to Camilo Uribe for help in testing and working out problems on Mono. Note that it not currently possible to build PythonNet using mono, due to the use of some IL attributes that the mono assembler / disassembler doesn't support yet.
Preliminary tests show that PythonNet does actually run under mono, though the test suite bombs out before the end with an "out of memory" error from the mono runtime. It's just a guess at this point, but I suspect there may be a limited pool for allocating certain reflection structures, and Python uses the reflection infrastructure quite heavily.
🚚 Removed decoys like the non-working embedding APIs; lots of internal refactoring.
👍 Implemented indexer support. Managed instances that implement indexers can now be used naturally from Python (e.g.
👍 Implemented sequence protocol support for managed arrays.
Implemented basic thread state management; calls to managed methods no longer block Python. I won't go so far as to say the thread choreography is "finished", as I don't have a comprehensive set of tests to back that up yet (and it will take some work to write a sufficiently large and evil set of tests).
🛠 Fixed a bug that caused conversions of managed strings to PyUnicode to produce mangled values in certain situations.
🛠 Fixed a number of problems related to subclassing a managed class, including the fact that it didn't work :)
🚀 Fixed all of the bugs that were causing tests to fail. This release contains all new bugs and new failing tests. Progress! :)
- 🚀 Initial (mostly) working experimental release.
🚀 [unreleased]: ../../compare/v2.3.0...HEAD