Classes with virtual ref(), unref() and get_ref_count() methods, like RecorderBase, could not be returned by PT() from methods because they didn't inherit from ReferenceCount. However, classes do not need to inherit ReferenceCount to be able to be tracked by a PointerTo, and defining an abstract base class with pure virtual ref()/unref()/get_ref_count() is a way to avoid dual inheritance of ReferenceCount.
We need third-party extensions that link with the Panda3D libraries to continue to work when shipping them with the optimized libraries that deploy-ng uses. To do this, we need the optimized build not to omit symbols that these extensions might depend on.
On MSVC (not with GCC/clang), adding `static` changes the mangled symbol name, so we shouldn't add that when building with NDEBUG. On GCC/clang, it doesn't, but adding `const` does, and C++11 rules make `constexpr` methods implicitly `const`, so I've removed the `constexpr` variants from NotifyCategoryProxy for now. Hopefully the compiler is still smart enough to compile out any references when compiling with NDEBUG.
Previously it was only possible to set this in the default-loaded Config.prc file; now it is possible to set this at any time, though it can only be set once from its default value and not changed after it has already been set to something other than the default value.
This happens if load_dso fails due to being unable to find the file, but not setting the dlerror flag and therefore causing a misleading error message to be shown.
The main motivation behind this change is to get rid of a
signed integer overflow that sometimes happens in the prime
multiplication step, which is (per the C++ spec) undefined
behavior. However, it's probably for the best to use only
unsigned int when the function is clearly trying to avoid
negative values.
Not that I suspect it matters much, but I have also heavily
tested that the behavior of the function is unchanged (at
least on PC hardware - signed integer overflow doesn't
behave portably) although it may now be slightly faster due
to the fact that I have removed the floating-point math.