Every disk usage analysis utility has its own way of (not) counting hard links. There does not seem to be any universally agreed method of handling hard links, and it is even inconsistent among different versions of ncdu. This section explains what each version of ncdu does.
ncdu 1.5 and below does not support any hard link detection at all: each link is considered a separate inode and its size is counted for every link. This means that the displayed directory sizes are incorrect when analyzing directories which contain hard links.
ncdu 1.6 has basic hard link detection: When a link to a previously encountered inode is detected, the link is considered to have a file size of zero bytes. Its size is not counted again, and the link is indicated in the browser interface with a 'H' mark. The displayed directory sizes are only correct when all links to an inode reside within that directory. When this is not the case, the sizes may or may not be correct, depending on which links were considered as "duplicate" and which as "original". The indicated size of the topmost directory (that is, the one specified on the command line upon starting ncdu) is always correct.
ncdu 1.7 has improved hard link detection. Each file that has more than two links has the "H" mark visible in the browser interface. Each hard link is counted exactly once for every directory it appears in. The indicated size of each directory is therefore, correctly, the sum of the sizes of all unique inodes that can be found in that directory. Note, however, that this may not always be same as the space that will be reclaimed after deleting the directory, as some inodes may still be accessible from hard links outside it.
Directory hard links are not supported. They will not be detected as being hard links, and will thus be scanned and counted multiple times.
Some minor glitches may appear when displaying filenames that contain multibyte or multicolumn characters.
Please report any other bugs you may find at the bug tracker, which can be found on the web site