Feature request: Diff mode

[turion]

I frequently have to compare large csv files where only a few fields in moderately many rows have changed. It would be cool to have a diff mode that shows the cell-wise diff of two csv files.

[BurntSushi]

Could you please provide more details? An example with inputs and outputs would help.

[turion]

An example with inputs and outputs would help.

Good idea.

Here is a rough sketch:

$ cat a.csv
foo,bar
a,23
b,42
[...many lines]

$ cat b.csv
foo,bar
a,100
b,42
[...many lines]
c,0

$ xsv --diff a.csv b.csv
@@ -1,bar +1,bar @@ foo,bar
  a,-100+23
@@ -1234 +1234 @@ foo,bar
- c,0

$ xsv table --diff a.csv b.csv
  foo bar
  a   -100+23
+ c   0

[turion]

I guess there are other interesting interactions. E.g. xsv stats --diff could show the number of changed rows and cells. xsv select --diff could limit the diff on certain columns.

[Yomguithereal]

Hello @turion, do you know daff? Reading your initial question I remembered about this tool that seem to do the job you need. It can also be easily integrated with git if I remember correctly.

[turion]

@Yomguithereal that sounds cool! Yes, that's sort of the feature set I'd like to see.

[nicoburns]

I think a daff-style diff is the way to go for this feature. Daff actually has a spec: http://paulfitz.github.io/daff-doc/spec.html, and the codebase (written in Haxe) is MIT licensed.

[kevinji]

The simplest version of this that would be useful for me would contain:

• A way to select a subset of columns as a key. If there are multiple rows with the same key, raise an error.
• In the diffed csv, some way to tell what was added/edited/removed/unchanged.

Something like this is useful if you have a job that snapshots state periodically and you need to figure out what changed. Here, the format rarely changes but the contents often do.