I do not have much background in LZW compression or in C, but I'm reasonably confident this resolves the problem I'm observing. It looks like the GifImageDecoder was not always correctly handling compression codes after the table reached its limit of ~4096. If anyone has suggestions for improvements I'm happy to make adjustments.
Luckily while debugging this: I was able to compare the flawed implementation (GifImageDecoder) with ImageIO's implementation (GIFImageReader) to help identify how the suffix/prefix tables are supposed to work. ImageIO's implementation may have suffered a similar bug (maybe https://bugs.openjdk.org/browse/JDK-7131823 ?), and that appears to have been backported. I have dozens of additional test cases for this problem, but unfortunately I don't have the rights to commit them to the OpenJDK repo. Feel free to email me for additional context/test cases. ------------- Commit messages: - 8351913: changing whitespace - Merge branch 'master' into JDK-8351913 - Merge branch 'master' of https://github.com/mickleness/jdk - 8351913: code cleanup - 8351913: support decoding LZW image data with 4095 entries - Merge branch 'openjdk:master' into master - Merge branch 'openjdk:master' into master - Merge branch 'openjdk:master' into master - Merge pull request #5 from openjdk/master - Merge pull request #4 from openjdk/master - ... and 3 more: https://git.openjdk.org/jdk/compare/9fcb06f9...6179ea22 Changes: https://git.openjdk.org/jdk/pull/24271/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=24271&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8351913 Stats: 249 lines in 4 files changed: 241 ins; 0 del; 8 mod Patch: https://git.openjdk.org/jdk/pull/24271.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/24271/head:pull/24271 PR: https://git.openjdk.org/jdk/pull/24271