Special case known good names for use-before-def

[jakebailey]

Fixes #391.

Sort of a hack, but these hardcoded names are pulled from the list of atoms in the Python grammar. I've still been unable to reproduce True/False being use-before-def (other than once without a debugger attached), but it shouldn't occur given that I now explicitly ignore them.

Unfortunately, this may mask a bug in ProjectState.BuiltinModule.GetMember, since I don't think there's a good reason why None should fail to be found...

[MikhailArkhipov]

This can happen when builtins module is not loaded/scraped, which would be a big problem in many other cases. Members are assigned in the AstBuiltinsPythonModule

            if (boolType != null) {
                _members["True"] = _members["False"] = new AstPythonConstant(boolType);
            }

This can be a threading issue such as code analyzed before builtins are loaded

[jakebailey]

Well, if you think there's a better way to approach this, I'd be curious to see what you think about the test case for None presented in the most recent comment in #391.

My previous fix to ignore ConstantExpression doesn't really work because the node being inspected is some sort of index expression. I think that fix was actually also masking this same problem, because None should have shown up in the builtins.

[MikhailArkhipov]

Might be falling to fallback in

            var moduleRef = await Modules.TryImportAsync(_builtinName, token).ConfigureAwait(false);
            if (moduleRef != null) {
                _builtinModule = (BuiltinModule)moduleRef.Module;
            } else {
                _builtinModule = new BuiltinModule(fallback, this);
                Modules.SetModule(_builtinName, BuiltinModule);
            }

[MikhailArkhipov]

On the None - try adding it to members as we do with True and False

                    if (typeId == BuiltinTypeId.NoneType) {
                        noneType = m as IPythonType;
                    }
...
_members["None"] = new AstPythonConstant(noneType);

[jakebailey]

Doing that does actually fix that Generator test (and is way less hacky), but I'm not exactly sure where how to approach finding how the analysis for builtins doesn't happen before the rest of the diagnostics happens.

[jakebailey]

Closing in favor of #447 for the None part. The True/False issue remains unsolved and will likely need to be approached in another way.