Incorrect handling of nullable types defined in typedefs

[mtomasz-chromium]

The following code will compile, but it should not.

/**
 * @param {!FooBar} foo
 */
function say(foo) {
};

/**
 * @typedef {?{
 *   name: string
 * }}
 */
var FooBar;

This bug is very bad, as it makes us believe that FooBar is non-nullable, but it's not true. In Chromium we use this syntax a lot for defining new types in externs files, and we have possibly a lot of bugs which are not caught by the Closure Compiler.

[dimvar]

As a rule of thumb, you should only be using ! followed by a class or interface name. It is harder for the compiler to restrict the type if an arbitrary type can follow the !.

Of course, it would be good if the compiler warned when it can't handle ! properly, but it doesn't currently.

[dimvar]

Also, you may want to try the new (still experimental) type inference for your project, which does the right thing in this example. To enable it, pass the flag --new_type_inf to the compiler.