Description

This PR proposes adding checks to stubtest so that it raises an error if a variable is annotated as being writeable in the stub, but is a read-only property at runtime.

New errors reported by stubtest with this patch applied, when checking typeshed

0; however, this patch was used to prepare the following PR, fixing roughly 50 hits:

• stdlib: fix many attributes which are read-only at runtime but read-write in the stub typeshed#7395

Things I tried, which ultimately proved unfruitful

• Checking the opposite scenario as well: raising an error if an attribute was marked as read-only in a stub but was a read-write property at runtime. This resulted in many false positives and 0 true positives.
• Adding a similar check for other common descriptors, such as instances of types.GetSetDescriptorType. Unfortunately, it is much harder to accurately determine whether or not an instance of GetSetDescriptorType is a read-only data descriptor, or a read-write data descriptor. (All instances of GetSetDescriptorType have a __set__ method, but many of these __set__ methods raise AttributeError. Whereas with properties you can do a simple if prop.fset is None check, there's no equivalent test for GetSetDescriptors.)

Test Plan

Four Five Four new test cases added, one modified.

cc. @hauntsaninja