QFuture::cancel() is impractical on continuation chains

QFuture::cancel() behaves unintuitively and is impractical when used in conjunction with continuation chains.

Specifically, there's no obvious way to cancel such a chain. Suppose we have:

auto f = foo()
    .then(...)  // a
    .then(...)  // b
    .then(...)  // c
    .then(...); // d

Calling f.cancel() will only cancel the very last continuation (`d`), nothing else. This is contrary to basic expectations - I'd expect every not-yet-finished continuation to be canceled as well. It's also not documented.

Having read the impl of QFutureInterfaceBase::cancel(), I thought that I could maybe cancel() the very first future:

auto f = foo();
auto firstF = f;
f = f
    .then(...)  // a
    .then(...)  // b
    .then(...)  // c
    .then(...); // d
firstF.cancel();

But that only works when `firstF` has not yet finished! If it is finished already by the time I cancel it, nothing happens and the chain runs to the end.

In the end, I have to keep a list of all intermediate futures and cancel each one in order to reliably cancel the chain:

QList<QFuture<void>> futures;
auto f = foo();
futures << f;
f = f.then(...)  // a
futures << f;
f = f.then(...)  // b
futures << f;
f = f.then(...)  // c
futures << f;
f = f.then(...); // d
futures << f;

for (auto &f : futures) {
    f.cancel();
}

This is super inconvenient (and unexpected).

It would be cool if there was an overload to forcibly cancel the whole chain that a future belongs to.

Reproduction

    QPromise<void> p1, p2;
    p1.start();
    p2.start();

    auto f = QtFuture::makeReadyVoidFuture()
    .then([f = p1.future()]() {
        qDebug() << "Init";
        return f;
    }).unwrap().then([f = p2.future()]() {
        qDebug() << "Continue1";
        return f;
    }).unwrap().then([&]{
        qDebug() << "Continue2";
    }).onCanceled([] {
        qDebug() << "Cancelled";
    });

    p1.finish();
    f.cancel();
    p2.finish();

Expected output:

Init
Continue1
Cancelled

Actual output:

Init
Continue1
Continue2

Note: Is the problem actually the nested and unwrapped futures?