* @param length the length of the new array
   */
  /*
   * The new array contains nulls, even if the old array did not. If we wanted to be accurate, we
   * would declare a return type of `@Nullable T[]`. However, we've decided not to think too hard
   * about arrays for now, as they're a mess. (We previously discussed this in the review of
   * ObjectArrays, which is the main caller of this method.)
   */

Complex logic ahead...
We have three sets of values, in order of precedence (last wins):
1. The builtin values.yaml defaults
2. The profile the user selects
3. Users input (-f or --set)

Unfortunately, Helm provides us (1) and (3) together (as .Values), making it hard to insert (2).

However, we can workaround this by placing all of (1) under a specific key (.Values.defaults).

//
// Below here we do the gradient algorithm. It works as follows:
//
// First we filter the tape to just the subset of operations we want to
// differentiate. In the process of doing so we count how many times each Tensor
// is used as an input to an op (so we know when we're done computing gradients
// for that Tensor). We also count, for each tape entry, how many of its output

We are going to include this `APIRouter` in the main `FastAPI` app, but first, let's check the dependencies and another `APIRouter`.

## Dependencies

We see that we are going to need some dependencies used in several places of the application.

So we put them in their own `dependencies` module (`app/dependencies.py`).

      if (retryRoute != null) {
        // Lock in the route because retryRoute() is racy and we don't want to call it twice.
        nextRouteToTry = retryRoute
        return true
      }
    }

    // If we have a routes left, use 'em.
    if (routeSelection?.hasNext() == true) return true

    // If we haven't initialized the route selector yet, assume it'll have at least one route.

    using a job called "copybara".

**5. Copy to Google Internal codebase and run internal CI**

-   Once the PR is in the Google codebase, we make sure it integrates well with
    its dependencies and the rest of the system.
-   Rarely, If the tests fail at this stage, we cannot merge the code.
-   If needed, we may come to you to make some changes. At times, it may not be

      // e.g. a certificate pinning error.
      return false
    }
    // An example of one we might want to retry with a different route is a problem connecting to a
    // proxy and would manifest as a standard IOException. Unless it is one we know we should not
    // retry, we return true and try a new route.
    return true
  }

  /**

	// Check if specified pools need to be removed from decommissioned pool.
	for k := range specifiedPools {
		pi, ok := rememberedPools[k]
		if !ok {
			// we do not have the pool anymore that we previously remembered, since all
			// the CLI checks out we can allow updates since we are mostly adding a pool here.
			update = true
		}
		if ok && pi.completed {

    return true;
  }

  static class NonTerminalImmutableMapEntry<K, V> extends ImmutableMapEntry<K, V> {
    /*
     * Yes, we sometimes set nextInKeyBucket to null, even for this "non-terminal" entry. We don't
     * do that with a plain NonTerminalImmutableMapEntry, but we do it with the BiMap-specific
     * subclass below. That's because the Entry might be non-terminal in the key bucket but terminal

    currently wrong for execution.

    <<NOTE:>> Do we want to fail when we cannot bind a mojo to the lifecycle
    because of offline/online status? That would probably indicate that the user
    was trying to do something they cannot succeed at for now...so we probably