Search code, repositories, users, issues, pull requests...
perfIO provides buffered streaming I/O abstractions for both binary and text data. The design is optimized for performance. The main perfIO classes correspond roughly to the following standard library abstractions:
Reading and writing binary data using a sufficiently large buffer is faster than ByteBuffer, but with the full generality of streaming I/O that does not require a fixed buffer size. Data from the included benchmarks (measured on Temurin OpenJDK 22.0.2 on Linux x68-64) shows typical speed-ups between 3x and 40x in different scenarios compared to standard library abstractions for streaming I/O.
InputStream and OutputStream use the decorator pattern. You typically use multiple nested streams that have to make all of their reads/writes through a standard set of methods that operate on individual bytes or byte arrays. You can easily end up with polymorphic dispatch (because you are always calling methods of the same interface with different implementations) and double buffering.
By contrast, BufferedInput and BufferedOutput are built around a single buffer (byte array, ByteBuffer or MemorySegment) with a position and limit. All reads and writes are made directly to the buffer. There is no method dispatch overhead no matter how deeply nested the objects are. Almost all reads and writes (except the ones that require re-buffering) are done with a single range check plus increment. Any additional checks are pushed out of the inner loop into the buffering.
