[TIKA-2873] Some password protected xlsx files no longer open with password

• Type: Task
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 1.21
• Assignee: Tim Allison
• Reporter: Tim Allison
• Created: 14/May/19 19:08
• Updated: 15/May/19 01:09
• Resolved: 15/May/19 01:09

Description

  A password protected .xlsx file can no longer be read because there's a bug in POI's ChunkedCipherInputStream:

@Override
public int read() throws IOException {
    byte[] b = \{ 0 };
    // FIXME: compare against -1 or 1? (bug 59893)
    return (read(b) == 1) ? -1 : b[0];
}
 

  Reading with a byte[] works fine, but read() individual bytes returns -1 if the stream had something in it. Because of some recent changes...we now wrap the ChunkedCipherInputStream in a RereadableInputStream...whereas we didn't before...this wrapping is now calling read() at some point where the stream used to only be read(byte[]). We can fix this by wrapping the ChunkedCipherInputStream in a TikaInputStream...

I think this is significant enough to do a respin of 1.21-rc1...what do you think?