ydiff

ydiff is a language-aware tool for comparing programs. It supports multiple languages and has an extensible design. It is not yet ready for general release, although it is already an indispensible tool to me.

ydiff is an open source project. You can follow its development on github: yinwang0/ydiff or get its source code from there. If you want to know more about it, welcome to read on and play with the interactive demos provided at the end of this page.

Why ydiff?

We very often want to know how our code is modified. This is normally done by using the diff program. While effective for finding differences between close revisions, the result from diff can be very hard to read for revisions over a long time span. The worst case is when the code is massively rearranged or refactored, where diff considers every line to have been changed.

ydiff can help in this kind of situation. Different from line-based or character-based diff, ydiff is "language-based". It parses the programs and then performs structural comparison on the parse trees. In addition to the usual functionalities of diff, it has the following properties:

• Language-awareness. ydiff truly understands language constructs and will not make non-sensical comparisons. For example it will never compare a string "10000" with an integer 10000 even though they look similar. Also, it tries to find definitions with the same name before it attempts to destruct and compare definitions of different names.
• Refactor detection. It can find renamed, moved, reordered, wrapped, lifted, combined or fragmented code. Refactored code can be detected however deep they are into the structures.
• Format insensitivity. The comparison result will not be affected by linebreaks or whitespaces.
• Comprehensible output. The output of ydiff is designed for human understanding. The interactive UI helps the user navigate and understand changes efficiently.

These properties make ydiff helpful for understanding changes. It may also be possibly used for detecting plagiarism in programming classes or copyright infringement of code.

Why "ydiff"?

I wanted to use the name "λdiff", but having noticed how troublesome it is to type it from the command line, I decided to flip its first letter upside down.

Supported Languages

Currently ydiff supports the following languages:

• Scheme
• Emacs Lisp
• Python
• Javascript
• C++

I will be adding more languages.

License

GPL v3

Authors

Yin Wang (yinwang0@gmail.com)

Demos

You are welcome to have a look at the demos. They are fun to play with. The demos are interactive HTML files. All program structures are in framed nodes, whose colors mean:

• inserted
• deleted
• changed
• moved
• unchanged

The left and right windows are always locked in their relative position. A mouse click on changed, moved or unchanged nodes will highlight the matched nodes and scroll the other window to match. After that, the windows will be locked into their new relative position for browsing.

Okay, here are the demos:

• Scheme

  Comparison of the original miniKanren from Professor Dan Friendman and the version I modified in order to support condc, a "negation operator", which fascinated me for quite some time. Pay attention to unify, whose major part is moved into unify-good, the righteous opponent of unify-evil.

• Emacs Lisp

  Comparison of two versions (v20 and v22) of Taylor Campbell's paredit-mode, a wonderful tool for editing Lisp programs.

• Arbitrary S-expressions

  Trying to find a bug in an optimizing pass of my Scheme compiler by comparing the IRs.

• Python

  Comparison of two implementations of a small list library that I wrote, which implements Lisp-style lists. The first implementation uses recursive procedures while the second uses Python's generator syntax and is iterative. Pay some attention to append, whose code is moved inside another function appendAll.

• Javascript

  Comparison between two major revisions of the UI code of ydiff itself.

• C++ demo1 and C++ demo2

  There are two demos for C++. The first demo compares two versions of the d8 Javascript debugger from the V8 project (v3404 from 2009 and the current v8424). The second demo compares V8's simulators for two different processors (MIPS and ARM).

  The d8 demo is especially interesting because by clicking on the lines of the method Shell::Initialize in the old version, it can be clearly observed that its body has been distributed into several procedures in the new version:

  Shell::Initialize
  Shell::CreateGlobalTemplate
  Shell::RenewEvaluationContext
  Shell::InstallUtilityScript
  

  Also the major part of Shell::Main is moved into the helper Shell::RunMain.