Is green highlighting missing, or am I misunderstanding? about difftastic

[tomjaguarpaw]

I see, thanks. For me an inline view is essential so I guess difft won't be a good fit for me. Still, I'm happy to have tried it out. Thanks for writing it!

[Wilfred]

Why would you expect line 2 to show as green? As far as I can see there are only removals on that line.

[tomjaguarpaw]

Hi Wilfred, thanks for your reply. Perhaps this is an XY problem so let me put it another way.

The standard diff tool on my system has the following property: from the content of test1.hs and the output of diff test1.hs test2.hs I can deduce the content of test2.hs. This seems to me to be the most essential property of a diff tool. Specifically, if test1.hs contains

example cond =
  print (if cond then 0 else 1,
         if cond then 2 else 3
        )

and the output of diff /tmp/test1.hs /tmp/test2.hs is

2,3c2,4
<   print (if cond then 0 else 1,
<          if cond then 2 else 3
---
>   print (if cond
>          then (0, 2)
>          else (1, 3)

then I can deduce the content of test2.hs by replacing test1.hs's lines 2 and 3 with the three lines in the second section of the diff output. However, I don't understand how the same deduction is possible using the output of difft. Perhaps I'm just unaware of the correct interpretation of the output format (or perhaps it's not supposed to be possible). Let's consider (for simplicity) the case without color, that is

difft --display=inline --color=never /tmp/test1.hs /tmp/test2.hs

The output is

/tmp/test2.hs --- Haskell
1    example cond =
2      print (if cond then 0 else 1,
3             if cond then 2 else 3
   3          then (0, 2)
   4          else (1, 3)
   5         )

How can I deduce the content of test2.hs knowing just the content of test1.hs and the output of difft? Put another way, if I were going to write the equivalent of patch for difft, what algorithm would I use to determine the effect of difft's output on the original file? The difficulty here is that I don't know how I could determine that line 2 of test2.hs is

  print (if cond

[Wilfred]

Difftastic does not aim to provide a valid patch. There's a brief discussion in non-goals here: https://github.com/Wilfred/difftastic#non-goals

Difftastic works out the the relationship between the two files, and highlights the parts of the AST that don't match. That's why it considers the following JSON files the same:

[100, 200, 300]

[100,
 200,
 300]

It might be interesting to add a mode that shows a traditional patch with the AST diff elements highlighted. This would be tricky because some line-based patches are not syntactic changes (like the example above).

[tomjaguarpaw]

Sure, I don't actually want to apply the output of difft as a patch. I'm clearly explaining myself poorly. The fact that equivalent ASTs compare as equal is important information for me to take into account, thanks. I'll try again.

I'm asking whether, in principle, knowing the content of test1.hs and the output of difft test1.hs test2.hs, I should be able to deduce the content of test2.hs, modulo AST equivalence? That seems to me to be an essential property of a diffing tool, but it seems to me that difft doesn't have that property (example given above).

If it is a goal, does difft actually satisfy it? If not, then presumably it's a bug. If so, then can you explain how? I don't understand how to deduce the contents of test2.hs in the example above.

[Wilfred]

Yep, you should be able deduce the content of the second file modulo AST equivalence.

The inline display is harder to read, and I rarely use it. This is one of the inline display failure cases: when the line-based diff and AST diffs consider the first change to be in different lines, it's hard to render in a comprehensible way.

Here's the default display:

From this, I can deduce that I can transform the first file into the second by opening a parenthesis before the 0, removing the else 1 to get me up to the comma, and so on.