Are you working to eliminate similar/duplicate code from your files? Do you have a suspicion that chunks of code are copy-pasted, but are slightly different s.t. normal diff methods don't work? Are you tired of visually going through and inspecting your code for repeating chunks?

If so, this might be the tool for you!

• Finds duplicate code slices
• Finds similar-enough code slices
• JSON reporting for jq integeration
• Fast enough (00:01:12 for a 17k LOC with block size 10 and Levenshtein threshold 10)
• Can check for duplicate code across multiple files
• Vim integration!
• Multithreaded

• Not instantaneous for large files
• Multithreading only applies for multiple files (no difference when running on a single file)

$ superdiff -l 1 -b 5 examples/really-bad-code.py
=== MATCH ===
File: examples/really-bad-code.py
Lines: [5, 11]
Size: 5
$ find src vim-superdiff -type f | superdiff -l 1 -b 7 --worker-threads 4
=== MATCH ===
File: src/types.rs
Lines: [87, 185]
Size: 7
=== MATCH ===
File: vim-superdiff/autoload/superdiff.vim
Lines: [136, 152]
Size: 7

Say you have some file examples/really-bad-code.py that you want to inspect.

#!/usr/bin/env python
class SomeClass:
 def __init__(self):
 self.alpha = 12
 self.beta = 14
 self.gamma = 16
 self.is_bad = True
 def reset(self):
 self.alpha = 12
 self.beta = 14
 self.gamma = 16
 self.is_bad = True
 def do_something(self):
 d = {}
 import random
 for i in range(20):
 if i % 3 == 0: continue
 d[i] = random.randrange(1, 1001)
 d[i ** 2] = d[i] ** 2
 d[d[i]] = i
 def do_something_else(self):
 d = {}
 import random
 for i in range(21):
 if i % 3 == 1: continue
 d[i] = random.randrange(1, 1001)
 d[i ** 2] = d[i]
 d[d[i]] = i
inst = SomeClass()
inst.reset()

You have a feeling that it might be bad, so you use the tool.

$ superdiff -b 4 examples/really-bad-code.py
=== MATCH ===
File: "examples/really-bad-code.py"
Lines: [5, 11]
Size: 5
A total of 1 unique match(es) were found in the 1 file(s).

Wow! That's pretty nice that you found that! But maybe there are places in the file that aren't exact copies, but are similar enough.

$ superdiff -b 4 -t 5 examples/really-bad-code.py
=== MATCH ===
File: "examples/really-bad-code.py"
Lines: [16, 26]
Size: 10
=== MATCH ===
File: "examples/really-bad-code.py"
Lines: [5, 11]
Size: 5

Huh, apparently there is a duplicate function that are pretty similar! And now (assuming that the output of the function is pretty long and not laughably short), you want to know if line 30 is involved in duplicate code, so you do the following:

$ superdiff --reporting-mode json -b 5 -t 5 examples/really-bad-code.py > output.json
$ cat output.json | jq
{
 "files": {
 "examples/really-bad-code.py": {
 "count_blocks": 4
 }
 },
 "matches": [
 {
 "blocks": {
 "examples/really-bad-code.py": [
 {
 "block_length": 5,
 "starting_line": 11
 },
 {
 "block_length": 5,
 "starting_line": 5
 }
 ]
 },
 "files": {
 "examples/really-bad-code.py": {
 "count_blocks": 2
 }
 }
 },
 {
 "blocks": {
 "examples/really-bad-code.py": [
 {
 "block_length": 10,
 "starting_line": 26
 },
 {
 "block_length": 10,
 "starting_line": 16
 }
 ]
 },
 "files": {
 "examples/really-bad-code.py": {
 "count_blocks": 2
 }
 }
 }
 ],
 "version": "2.1.2"
}
$ cat output.json | jq '.matches | map(select((.blocks."examples/really-bad-code.py" | any(.starting_line <= 30 and .starting_line + .block_length >= 30))))'
[
 {
 "files": {
 "examples/really-bad-code.py": {
 "count_blocks": 2
 }
 },
 "blocks": {
 "examples/really-bad-code.py": [
 {
 "starting_line": 16,
 "block_length": 10
 },
 {
 "starting_line": 26,
 "block_length": 10
 }
 ]
 }
 }
]

Note: If anyone finds a better way of making the jq query, please make a pull request and/or let me know.

It's kind of work-in-progress at the moment, but here's what we have:

• Load JSON with :SDLoad (make sure you are in the same directory you ran superdiff)
• Open a file to edit
• Run :SDLocal to highlight matching blocks of code
• Run :SDQuery on a matching block to find other blocks of similar code

Check the vimdocs for more options and commands.

asciicast

These numbers are here to give a ballpark estimate of how long something would take given some files. It is not scientific. I simply appended time to the beginning and copied the real time.

The data used for this benchmark can be found in scripts/populate-data.sh.