Julia functions for finding and removing duplicate files.



Build Status Coverage Status

** This software deletes files. If it is used incorrectly, or if it has bugs, it may delete the wrong ones, causing data loss and/or rendering your system unusable. Make sure you have backed up all your data before proceeding. **

This julia module helps with finding and deleting duplicate files. There are already numerous pieces of software, both free and commercial, that do this. Most of them have nice-looking graphical user interfaces. This module does not have a graphical user interface, or even a command-line interface. It can only be used from within julia.

Instead of answering questions about which files to delete (which can be a very time-consuming process when there are lots of files to de-duplicate) the user writes a function that tells the software when one of two identical files should be deleted.


Let's assume that we want to find all duplicate files under the directory foo, and for each set of duplicates, keep one file. (We don't care which one.)

We're not allowed to use a decision function that simply returns true for any pair of files. It is required to be consistent in its choice of files to delete. So we use alphabetical ordering of the file paths as a tie breaker.

list = deduplicate_files("foo", (a,b)->A.realpath > B.realpath, verbose=true, dry_run=true)

We then examine the returned list. If all looks good, we re-run the command without the dry_run argument.


The typical way to use the software is to first call the function list = deduplicate_files(startdirs, dfun, dry_run=true), where startdirs is an array of directory paths, and dfun is a "decision function" that tells the software which file(s) to delete from a set up duplicates. The directories will be searched recursively and the duplicate files that would be deleted will be stored in a list. If the list looks correct, then the function can be called again, without dry_run=true causing the files to actually be deleted. Otherwise, the decision function must be re-written, and a new dry run performed.

The decision function dfun(A,B) must satisfy the following criteria:

  • It takes as input arguments two DeduplicationFile structs.
  • It should return true if the file that A points to should be deleted after confirming that it is identical to the one that B points to. (And false otherwise.)
  • It must behave like an isless function, establishing a consistent order. For example, dfun(x,y) and dfun(y,x) may not both be true. (But they may both be false if neither x nor y should be deleted.)
  • It must not assume that the given files are identical. The software is allowed to call the decision function on suspected duplicates, deferring the byte-for-byte comparison until just before a duplicate is deleted.
  • It should NOT delete the duplicate file. (This can have disastrous consequences if the user only intended to do a dry run.)

The DeduplicationFile descriptors provided in the arguments to the decision function have the following fields:

  • start - The starting point given.
  • realstart - The absolute path of start, after expanding symbolic links.
  • relpath - The relative path from start to the file in question.
  • realpath - The absolute path to the file, after expanding symbolic links.
  • dirname - The directory part of realpath.
  • dirinode - The inode number of the directory in which the file resides.
  • basename - The filename.
  • stat - The StatStruct for the file. (See the documentation for stat.)

Another example

Let's assume that foo and bar are directories. We want to delete files under foo that have identical copies somewhere under bar, but only if they are larger than one kilobyte, the filenames are identical, and they are not hard-linked to the same inode. (Deleting hard-linked copies doesn't free up much disk space, so we keep them.)

We write a decision function for this:

const p1 = "/path/to/foo"
const p2 = "/path/to/bar"
dfun(a,b) = a.start == p1 && b.start == p2 &&
            a.stat.size > 1024 && a.basename == b.basename &&
            a.stat.inode != b.stat.inode

Then we test our decision function by doing a dry run.

list = deduplicate_files([p1, p2], dfun, dry_run=true)

We might store the returned list to disk in order to keep a record of the files that were deleted and where their identical copies were found.


** This software is provided as-is, without any warranty of any kind. **

I wrote it in my spare time, because I needed it myself, and I open-sourced it in case somebody else might find it useful.

I haven't done any extensive testing, other than using it for the specific task that I wrote it for. In particular, I've never tested this on a file-system with hard-linked directories. (Why would you use hard-linked directories?)

If using this software breaks your system, I will not be able to help you in any way, except to tell you to restore from backups. You did make backups, didn't you?

First Commit


Last Touched

16 days ago


4 commits


Used By: