* seenExceptionsField to its exception
     *
     * Thread2: calls setException(), which returns false, CASes seenExceptionsField to its
     * exception, and wrongly believes that its exception is new (leading it to logging it when it
     * shouldn't)
     *
     * Our solution is for threads to CAS seenExceptionsField from null to a Set populated with _the

    period. The last release to remove non-`@Beta` APIs was Guava 21.0.) Even
    `@Deprecated` APIs will remain (again, unless they are `@Beta`). We have no
    plans to start removing things again, but officially, we're leaving our
    options open in case of surprises (like, say, a serious security problem).

3.  Guava has one dependency that is needed for linkage at runtime:
    `com.google.guava:failureaccess:1.0.3`. It also has

    return asCharSource(url, charset).readLines(callback);
  }

  /**
   * Reads all of the lines from a URL. The lines do not include line-termination characters, but do
   * include other leading and trailing whitespace.
   *
   * <p>This method returns a mutable {@code List}. For an {@code ImmutableList}, use {@code
   * Resources.asCharSource(url, charset).readLines()}.
   *
   * @param url the URL to read from

 * it delegates to other components that potentially encounter problems. Then, the problem reporter can focus on
 * providing a simple error message, leaving the donkey work of creating a nice model problem to this component.
 *
 */
public interface ModelProblemCollector {

    ProblemCollector<ModelProblem> getProblemCollector();

    default boolean hasErrors() {

Java serialization is inherently Java-centric and does not support cross-language scenarios well.

- **Type Safety:**
Java serialization does not enforce type safety as strictly as some alternatives, potentially leading to runtime errors.

## Decision

We do not use Java serialization.
Instead, we use custom serialization where we explicitly describe how data objects should be serialized and deserialized.

                String errMsg = "Missing Javadoc file for " + link + " in " + sourceFile.getName();
                if (link.startsWith("javadoc")) {
                    errMsg += " (You may need to remove the leading `javadoc` path component)";
                }
                errorsForFile.add(new Error(lineNumber, line, errMsg));
            }

		{"!\"#$%&'()*+,-.／:;<=>?@[\\]^_`{|}~", true},
		{"␀␁␂␃␄␅␆␇␈␉␊␋␌␍␎␏␐␑␒␓␔␕␖␗␘␙␚␛␜␝␞␟␡", true},
		{"trailing VT␋/trailing VT␋", true},
		{"␋leading VT/␋leading VT", true},
		{"~leading tilde", true},
		{"\rleading CR", true},
		{"\nleading LF", true},
		{"\tleading HT", true},
		{"trailing CR\r", true},
		{"trailing LF\n", true},
		{"trailing HT\t", true},
		// cases for which test should fail.

* **Machine Learning**: it normally requires lots of "matrix" and "vector" multiplications. Think of a huge spreadsheet with numbers and multiplying all of them together at the same time.

# TODO(b/173549186): Move Google-internal TF code out of learning/brain
package_group(
    name = "internal",
    packages = [
        "//devtools/python/indexer/...",
        "//learning/brain/keras/...",
        "//learning/brain/mlir/...",
        "//learning/brain/tfrt/...",
        "//learning/lib/ami/simple_ml/...",
        "//learning/pathways/...",
        "//learning/serving/contrib/tfrt/mlir/canonical_ops/...",

Here's a little bit of that history.

## Alternatives { #alternatives }

I have been creating APIs with complex requirements for several years (Machine Learning, distributed systems, asynchronous jobs, NoSQL databases, etc), leading several teams of developers.

As part of that, I needed to investigate, test and use many alternatives.

The history of **FastAPI** is in great part the history of its predecessors.