// one. (StringCoding$StringDecoder vs. StreamDecoder).  StringCoding has a ton of special cases
    // theoretically we could duplicate all that logic here to try to beat 'new String' or at least
    // come close.
    USING_DECODER_WITH_SIZE_HINT {
      @Override
      String read(ByteSource byteSource, Charset cs) throws IOException {
        Optional<Long> size = byteSource.sizeIfKnown();

    // one. (StringCoding$StringDecoder vs. StreamDecoder).  StringCoding has a ton of special cases
    // theoretically we could duplicate all that logic here to try to beat 'new String' or at least
    // come close.
    USING_DECODER_WITH_SIZE_HINT {
      @Override
      String read(ByteSource byteSource, Charset cs) throws IOException {
        Optional<Long> size = byteSource.sizeIfKnown();

   *
   * Fundamentally, there's not really anything we can do about that. In the unlikely event that it
   * comes up in practice (maybe through some kind of sanitizer-like testing that intentionally
   * inflicts spurious interrupts on us?), we might have to accept some flakiness or disable some
   * tests, at least under whichever environment (JRE or Android) we see such problems.
   */

import java.math.BigInteger;
import org.jspecify.annotations.Nullable;

/**
 * A wrapper class for unsigned {@code long} values, supporting arithmetic operations.
 *
 * <p>In some cases, when speed is more important than code readability, it may be faster simply to
 * treat primitive {@code long} values as unsigned, using the methods from {@link UnsignedLongs}.
 *
 * <p>See the Guava User Guide article on <a

#   Googlers should replace "nightly_upload" with "public_cache,disk_cache" or
#   "rbe", if you have set up your system to use RBE (see further below).
#
# Here is how to choose your TFCI value:
# 1. A Python version must come first, because other scripts reference it.
#      Ex. py39  -- Python 3.9
#      Ex. py310 -- Python 3.10
#      Ex. py311 -- Python 3.11
#      Ex. py312 -- Python 3.12

        this.defaultLifeCycles = requireNonNull(defaultLifeCycles);
    }

    // These methods deal with construction intact Plugin object that look like they come from a standard
    // <plugin/> block in a Maven POM. We have to do some wiggling to pull the sources of information
    // together and this really shows the problem of constructing a sensible default configuration, but

   problems later. Be thoughtful and considerate when seeking out the appropriate forum for your
   questions. Those who are asked should be responsive and helpful.

 * **Step down considerately**: Members of every project come and go. When somebody leaves or
   disengages from the project, they should make it known and take the proper steps to ensure that
   others can pick up where they left off.

Compiling models via the recommended entry points described in
[XLA](https://www.tensorflow.org/xla) and
[JAX](https://jax.readthedocs.io/en/latest/jax-101/02-jitting.html)
documentation should be safe, while some of the testing and debugging tools that
come with the compiler are not designed to be used with untrusted data and
should be used with caution when working with untrusted models.

### Saved graphs and checkpoints

That would refer to some package above `app/`, with its own file `__init__.py`, etc. But we don't have that. So, that would throw an error in our example. 🚨

But now you know how it works, so you can use relative imports in your own apps no matter how complex they are. 🤓

### Add some custom `tags`, `responses`, and `dependencies` { #add-some-custom-tags-responses-and-dependencies }

         * simply rethrow an exception (e.getCause()) that may in rare cases have come from another
         * thread. To accomplish both goals, we wrap that IllegalArgumentException in a new
         * instance.
         */
        throw new IllegalArgumentException(e.getCause().getMessage(), e.getCause());
      }
      /*
       * If some other exception happened, we just propagate the wrapper