Last time we talked about .NET dependent handle. It is a handle that promotes secondary if primary is promoted - as if there is a imaginary reference between them. This time let’s take a look at another secret handle - ref-counted handle. A ref-counted handle is a special handle that will become either strong or weak depending on the ref count. It’s only used in COM interop today internally in the CLR.
You can find its definition in gcinterface.h
* REFCOUNTED HANDLES
* Refcounted handles are handles that behave as strong handles while the
* refcount on them is greater than 0 and behave as weak handles otherwise.
* N.B. These are currently NOT general purpose.
* The implementation is tied to COM Interop.
HNDTYPE_REFCOUNTED = 5,
.NET publicly has documented 4 kind of handles:
Weak (also called Short Weak) - Don’t keep target object alive and will return null when object is gone. The target will become null when the object enters for finalization.
WeakTrackResurrection (also called Long Weak) - Don’t keep target object alive and will return null when object is gone. It’ll return the object even when the object is being finalized or resurrected.
Normal (also called strong) - keeps target object alive. If you are not careful, you may leak the object.
Pinned - Keeps the target object alive and prevents GC from moving this object around. Useful when you are passing this object to native code, as native code won’t know if GC moved it. Note that using a lot of pinning handles may degrade GC performance. The most common offender is pinned strings/arrays in interop calls.
You can also find them described in GCHandle enumeration.
However, besides these 4 types, there are actually more secret internal handle types that are not exposed. In this post I’ll be talking about dependent handle, and why it is totally awesome.
Last time we’ve looked at using ctypes to call C API, and writing extension module using Python/C API. Now we can finally tie these two together - looking at how ctypes is actually implemented using mix of Python/C API and Python code.
- You can find CPython source code here.
- ctypes’ C implementation is here
- ctypes’ python implementation is here.
In my previous post we’ve briefly looked at using ctypes module to call C APIs. In this post let’s dive into a completely different approach - writing a C extension using Python/C API. From the C extension you can then call whatever C/C++ code you want. This is a pretty popular approach to expose a 3rd party C library as python module. The downside is this module would only be loaded by CPython, which is usually not a big issue - unless you use other Python implementation like PyPy.
Recently I’ve been evaluating Python interop technologies for a project at work and I think it’ll made an interesting blog series.