Hi,
RandomAccessCharacterFile.java needed fixing. I've modified the version in abcl-1.3.2 and attached it. It makes my original test case pass. It also makes Stas's shorter test case pass.
I was able to make Stas's test case pass before mine, so they aren't identical.
I've also attached Java versions of those two test cases.
I haven't run any of abcl's existing tests or the ANSI test suite against my version of RandomAccessCharacterFile.java to check for regressions. But my own program that does a lot of IO runs correctly against it.
Vibhu