* {@snippet :
   * Stream.of("foo", "quux", "banana", "elephant")
   *     .collect(least(2, comparingInt(String::length)))
   * // returns {"foo", "quux"}
   * }
   *
   * <p>This {@code Collector} uses O(k) memory and takes expected time O(n) (worst-case O(n log
   * k)), as opposed to e.g. {@code Stream.sorted(comparator).limit(k)}, which currently takes O(n
   * log n) time and O(n) space.
   *

models or graphs is equivalent to running untrusted code.

If you need to run untrusted models, execute them inside a
[**sandbox**](https://developers.google.com/code-sandboxing). Memory corruptions
in TensorFlow ops can be recognized as security issues only if they are
reachable and exploitable through production-grade, benign models.

### Compilation

   - sanjoy
# filesystem path
filesystem_path:
   - tensorflow/c/experimental/filesystem
# security path
security_path:
   - tensorflow/security
# words checklist
segfault_memory:
   - segfault
   - memory leaks
# assignees
filesystem_security_assignee:
   - mihaimaruseac
   
# Cuda Comment
cuda_comment: >
   From the template it looks like you are installing **TensorFlow** (TF) prebuilt binaries:

- A maximum of 100M inside a single zip is allowed. However, a reasonable limit of 100,000 files inside a single ZIP archive is recommended for best performance and memory usage trade-off.

## Content-Type

 * lock, but also maximizes memory footprint. On the other extreme, one could use a single lock for
 * all tasks, which minimizes memory footprint but also minimizes concurrency. Instead of choosing
 * either of these extremes, {@code Striped} allows the user to trade between required concurrency
 * and memory footprint. For example, if a set of tasks are CPU-bound, one could easily create a

# ADR-0002 - Avoid using Java serialization

## Date

2012-12-01

## Context

In Gradle we often need to serialize in-memory objects for caching, or to transmit them across process barriers, etc.
Java serialization is one way to implement this, however, despite its simplicity of implementation, it has several drawbacks:

- **Performance:**
Java's built-in serialization mechanism is often slower compared to other serialization solutions.

                String.format(Locale.ROOT, "Caught an exception in %s.  Shutting down.", t),
                e);
      } catch (Throwable errorInLogging) { // sneaky checked exception
        // If logging fails, e.g. due to missing memory, at least try to log the
        // message and the cause for the failed logging.
        System.err.println(e.getMessage());
        System.err.println(errorInLogging.getMessage());
      } finally {
        runtime.exit(1);

		return err
	}

	_, vssLimit, err := sys.GetMaxMemoryLimit()
	if err != nil {
		return err
	}

	if vssLimit > 0 && vssLimit < humanize.GiByte {
		logger.Info("WARNING: maximum virtual memory limit (%s) is too small for 'go runtime', please consider setting `ulimit -v` to unlimited",
			humanize.IBytes(vssLimit))
	}

	if ctx.MemLimit > 0 {
		debug.SetMemoryLimit(int64(ctx.MemLimit))
	}

 * }
 * public void testFooLeak() {
 *   GcFinalization.awaitClear(fooWeakRef());
 * }
 * }
 *
 * <p>This class cannot currently be used to test soft references, since this class does not try to
 * create the memory pressure required to cause soft references to be cleared.
 *
 * <p>This class only provides testing utilities. It is not designed for direct use in production or
 * for benchmarking.
 *
 * @author mike nonemacher

limitations under the License.
==============================================================================*/
#ifndef TENSORFLOW_C_EAGER_ABSTRACT_CONTEXT_H_
#define TENSORFLOW_C_EAGER_ABSTRACT_CONTEXT_H_

#include <memory>

#include "tensorflow/c/eager/abstract_function.h"
#include "tensorflow/c/eager/abstract_operation.h"

namespace tensorflow {

// Abstract interface to a context.
//