Performance degradation in ImageReader#getNumImages for JPEGImageReader (3.6.2 -> 3.6.3)

Describe the bug Calling ImageReader#getNumImages for JPEGImageReader execution time takes much more time starting with 3.6.3.

Version information

1. TwelveMonkeys ImageIO 3.6.2 is fine, 3.6.3 (also checked 3.8.2) are having the issue

2. Java openjdk version “1.8.0_302” OpenJDK Runtime Environment (Temurin)(build 1.8.0_302-b08) OpenJDK 64-Bit Server VM (Temurin)(build 25.302-b08, mixed mode)

To Reproduce Steps to reproduce the behavior: Use below test code

1. Compile the below sample code

Expected behavior Ideally it should be as fast as before.

Example code

ImageIO.scanForPlugins();
InputStream inputStream = RenditionServiceTest.class.getResourceAsStream("/images/image.jpg");
ImageInputStream iis = ImageIO.createImageInputStream(inputStream);
Iterator<ImageReader> iterator = ImageIO.getImageReaders(iis);
ImageReader reader = iterator.next();
reader.setInput(iis);
reader.getNumImages(true);

The above code executes in ~500ms for 3.6.2, ~9000ms for 3.6.3.

Additional context This commit seems to have removed the buffered image input stream (which made it fast): https://github.com/haraldk/TwelveMonkeys/commit/d5e664cdcc5bc9be0ba3f935e1b99dd3767205d8#diff-6e4d5b2b35650d42a59dfda91a85dbf159787a09272c0c3ffcc00a964c3dfe5aL750

[sghpjuikit]

Using setCache(false) for 8MB losless webp:

• ImageIO.createImageInputStream(FileInputStream(file).buffered(bufferSize)) // 1.5s (bufferSize 8k vs 124k vs 1M has 0 impact)
• ImageIO.createImageInputStream(FileInputStream(file)) // 22s
• ImageIO.createImageInputStream(file) // 26s
• ImageIO.createImageInputStream(MemoryCacheImageInputStream(FileInputStream(file))) // no output, no error
• ImageIO.createImageInputStream(FileCacheImageInputStream(FileInputStream(file), File("someDir"))) // no output, no error

Using setCache(false) for 80kB losless webp:

• ImageIO.createImageInputStream(FileInputStream(file).buffered(bufferSize)) // 45ms

Using setCache(true) does make it worse.

[haraldk]

New stream implementation out in 3.9.0 release, feel free to give it a try and report back! 😀