Hash calculation seems buggy (but consistent)

[csachs]

I was trying to recreate imagej-updater's hashing in Python, and while I had success with many files, I stumbled upon some very peculiar behaving files. Six hours of debugging later, I found two non-obvious bugs in imagej-updater's code.
While this seems to be two bugs, it doesn't really affect the updater, since the hashes are calculated consistently wrong ;) … or at least contrary to what taking a brief look at the code might suggest.

First: Contrary to what keepOnlyMainClass parameter (the default) suggests, the FilterManifest filter will never show the main class:

imagej-updater/src/main/java/net/imagej/updater/util/FilterManifest.java

Line 90 in cd8ba1c

if (line.toUpperCase().startsWith("Main-Class:")) builder.append(line).append('\n');

After the .toUpperCase() call, it can never .startsWith() the mixed case text fragment ... hence just an empty manifest is passed.

Second: A missing super(). turns this simple supposedly-iterative (I presume) function into a rather unpredictable recursive affair ... contrary to what SkipHashedLines suggests, some lines starting with a # will hence remain in the resulting stream.

imagej-updater/src/main/java/net/imagej/updater/util/SkipHashedLines.java

Line 65 in cd8ba1c

while ((ch = read()) != '\n' && ch != -1)