JEP draft: Prepare to make final mean final

I think there's only one case where i ended up relaxing integrity, and i'm hoping that's temporary - it will take more time to fix than i was willing to spend.

I think Java's handling of this transition compares very favourably to how other languages have handled similar transitions from some old model to a new one (or evolution in general) in terms of balancing the needs of both old and new projects.

https://www.youtube.com/watch?v=2y5Pv4yN0b0&t=930s

Link is to the start of a sequence of three slides, the third of which is the slide in question.

For a more recent update on serialization, watch this talk “Serialization: A New Hope”: https://www.youtube.com/watch?v=mIbA2ymCWDs

The theory is, go through the constructor. However, some objects are designed to go through several steps before reaching the desired state.

If GSON must deserialize {…, state:”CONFIRMED”}, it needs to call new Transaction(account1, account2, amount), then .setState(STARTED) then .setState(PENDING) then .setState(PAID) then .setState(CONFIRMED) ? That’s the theory of the constructor and mutation methods guarding the state, so that it is physically impossible to reach a wrong state.

There is a convention that deserialization is an exception to this theory: It should be able to restore the object as-is, after for example a transfer over the wire. So it was conventionally enabled to set final variables of the object, but only at initialization and only for its own good. It was assumed that, even though GSON could reach a state that was unachievable through normal means, it was, after all, the role of the programmer to add the right annotations to avoid this.

So how do we do it now?

The moment you need to run some code from another language on your code to generate some other code or whatever, they break down, while Gradle can be used for pretty much anything.

In other words, cargo/go only solve the cache/parallelize/resolve task dependencies problem for "hard coded" special cases, the moment you strive away from that you are in a world of pain.

If I understand correctly, here's a C++ example that has undefined behavior:

    int foo() {
        const int x = 42;
        int *p = (int *)&x;
        *p += 1;
        return x;
    }

    int bar() {
        int x = 42;
        const int *const_p = &x;
        int *p = (int *)&const_p;
        *p += 1;
        return x;
    }

In C, at least, you can usually distinguish a true pointer to immutable data by using `restrict`.

The converse: people really write `field.setAccessible(true)` and are OK being aware that they are breaking another programmer's assumptions. That's the point of `field.setAccessible(true)` existing at all. It's monkey-patching.

Otherwise you might as well just delete the method and throw a `NoSuchMethodError` when someone calls it - nobody is calling `setAccessible(false)`

I have some experience hacking on Java Minecraft. Minecraft's code is what it is - take it or leave it. The more stuff you can do without modifying that code, the simpler the overall system is. If you can set a field with reflection then you don't need to modify that class to make the field public or non-final. In other words, monkey-patching. (Newer frameworks support more modification to Minecraft's own code with less effort, though.)

> This is not a good thing because interpreted mode is terribly slow. The JIT exists to make Java fast, and developers should not made this harder than necessary.

And it does that by being smart. It makes assumptions, compiles code given those assumptions, and recompiles it if they actually do not hold. For example, I expect that a method with a List parameter is most likely only ever called with one implementation of List, and the JVM knows this and will compile the method as if the parameter is ArrayList, and will compile the method for the specific implementation it sees in the first few calls, with a bail-out check forcing the method to be recompiled if it's ever not that implementation.

And if the JVM never sees a subclass of a certain class, it can statically dispatch all calls to references of that type, as if the class were final, even if it isn't marked final.

And, likewise, if the JVM never sees a write instruction to a certain field, except once in the constructor, it should be able to treat it as if the field were final, and avoid reloading it after method calls. If it loads a class with an instruction that writes that field, it has to recompile all those methods.

None of this should be news to the JVM engineers.

Even the JS standard library struggles with this. You just have to remember that .sort() modifies the array in place (and returns it), but .slice() does a shallow clone of the array. (Not a deep clone - that would be different again!)

While bad code can exist in any language, I get worried about too much const in code, because it means they failed at both 1) and 2) and instead there are usually seriously tricky protocols that must be observed to make the thing work. I often ran into code where people were sprinkling const all over the code to lock things down but they fundamentally did not understand the design and made it nearly impossible to evolve, unless you used casts to get rid of const, which defeats the whole purpose.

I'm not saying const doesn't have value, but it's weapon #3, not weapon #1.