Notebooks - Diff indicator shown despite same output

[claudiaregio]

• VS Code Version: 1.62.0-insider
• OS Version: WIndows

Diff color indicators on output that is the same

[rebornix]

@claudiaregio can you please share the modified file content? Also do you see any change when you open the diff editor in text diff?

[claudiaregio]

Original Notebook + env required to run: https://github.com/claudiaregio/data-science
Titanic_mod.zip

This is what I'm seeing for the text diff

[rebornix]

@claudiaregio is the screenshot the only change in the document?

[claudiaregio]

There are other changes in the document, since all I do is run I believe it's flagging the output ID changing

[burkeholland]

Hit this in the Golden Scenario today. I am unable to tell from the diff what has changed and what has not as the output looks identical to me. Perhaps this is a matter of just knowing that the underlying JSON changed, but I found it confusing as I was looking for differences in places where there really is none of material importance.

whats-changed.mp4

[rebornix]

Thanks for the file provided above and the screen recording!

Dig into this and found it interesting, is there an output change? Yes, but it is not anything that one can tell from the Diffing. What are the changes?

• The output contains metadata, which includes the execution count, thus they are different on different runs. Not sure if they are saved to disk, but our diff view doesn't really tell you if it's output metadata change, or output value change. This is something we should improve in the diff editor
• If we then compare the output, the sequence of text/html/`text/plaintext' is flipped. This is when you see no change in the rich diff editor, but you see an entry in SCM cc @roblourens @DonJayamanne

[roblourens]

Maybe they need to be sorted along with other keys in #129435

[roblourens]

We already sort all keys in the notebook when serializing, so the order of the mimetypes should have been sorted. Which order did you see it in? Maybe it started out wrong? In that case Jupyter would have done the same thing.

[tanhakabir]

+1 from golden scenario run by me

[greazer]

Reopening this as it contains a lot of extra info and it's been repro'd and re-entered at least 2 other times. We really should do something to address the problem.

[roblourens]

Could you link the dupe issues?

[greazer]

Basically, looks like the same issue was found 3 separate times during golden scenario runs.

• Diff view shows output changes even when there is no actual change (only change is in the metadata) #141966
• Diff view: output metadata change UX is confusing #137836
• Notebook diff shows false information #121622 (comment) (first example)
• Notebook Diff showing what appears to be wrong information and wrong comparisons in 1.61.0. #133554 (essentially same as above)

[rebornix]

Should be fixed in today's Insiders.