== Critical ==
[ ] 

== Non-critical
[ ] optimization: already visited dirs, should not be visited again,
	same as with files.
[ ] xmalloc: we abort if we cannot alloc memory, so do not test for null.
[ ] test14: we should not skip hard links. we only need to cache the
	data and use it. Take care to not delete it!
[ ] We nay have also hard linked directories. Deal with it!
[ ] Add --exclude para and use the dev+inode to build a list to exclude dirs.
[ ] Auto run cppcheck!
[ ] 'level' should be int till the end.
[ ] Hard links (t_014) may be not correct. I think we ignore links and
	for matching dirs this may be a problem. Think about.
[ ] Sort dumping by size saved (option)?
[ ] Dump also how much we can save by deleting. Isn't is already in can_save
	and already reported?
[ ] Ignore empty dirs?
[ ] Warn users in man, to never delete some dirs/files that appear on the
	left side. Only from the right.
[ ] dev_ino: use blocks of structs.
[ ] Fix man with regarding the example.
[ ] Add -fno-guess-branch-probability when all braches are tagged.
[ ] Add some progress bar?
[ ] Maybe -v should be the default and -q (--quiet) to be added.
[ ] When big files are hashed, print the file name?
[ ] Lazy compute hash on files. Only if we have a match we will incrementally
	compute the checksum.
[ ] Adapt man file to recent changes: flags etc.
[ ] Ignore empty files. We have min_size. Should be default to 1?
[ ] 1-bit fields are not printed as %hhu!
[ ] Because we ignore !dir and !files, we may not have really identical
	directories. In one of them we may have a socket, for example.
	Add a reporting flag for this situation. Or, do not ignore other
	type of files.
[ ] I must document a high level view over "algorithm". Even myself
	I do not remember what I am doing...
[ ] Use fadvise to not cache data in RAM.
[ ] Set read-ahead when reading full files.
[ ] Use more threads

[ ] Strange case:
	dir1
		dir2
		dir3=dir4
	dir4=dir3

[ ] We could throw away unique files.
[ ] Comparing in O(N*N) sucks! Where?
[ ] Dump in stats also the max memory (peak) used.
[ ] Maybe, for flag 'M', list files that were renamed (cmd line flag).
[ ] Use a cache, specified by command line. Use inode and mtime as key?