Fix auto-indent on paste, second line and after

[eliot-akira]

This PR fixes #32320, fixes #31255.

In the method AutoIndentOnPaste.trigger, there's a logic to re-indent pasted lines based on editor tab/indent size (and optionally editor.detectIndentation). At the end of that calculation is a loop that re-indents the second line and after: it was adding unnecessary indents if any of the original lines had no indentation.

To illustrate, pasting the following:

<div>
  <i></i>
</div>

..was resulting in:

<div>
    <i></i>
  </div>

The only way for the user to fix this when it happened, was to undo once after pasting.

This PR adds a check to handle that case, by not adding indent if a pasted line had no indent to start with. In the above example, that's the third line.

It was somewhat tricky to reproduce the issue - although many people have reported variations of the symptoms in HTML, CSS, JS, etc. The common precondition seems to be when the tab/indent sizes detected for the document are different from the ones detected from the pasted lines. That leads to the loop that re-indents the second line and after.

[eliot-akira]

The failing integration tests are due to an unrelated network issue in the CI pipeline itself, it seems.

[eliot-akira]

I've been trying to find a reliable way to reproduce the issue that this PR attempts to fix.

On a clean install with default settings (editor.autoIndent and .detectIndentation set true, and .tabSize at 4):

• Create new file

• Enter

  <div>
    test
  </div>

• Save as test.html, close and reopen.

  Sometimes this step is not necessary(?), especially when .detectIndentation is false

• Copy & paste results in:

  <div>
      test
    </div>

• A single undo operation corrects the indentation back to original

Notably, in this example the editor's default tab size 4 is different from the pasted block's detected indent 2. I believe this is related to the issue.

At the beginning of AutoIndentOnPaste.trigger, there's a call to model.getOptions (which leads to textModel.resolveOptions) that gets the "starting" indent size. At the end of the auto-indent logic, since the pasted block's indent is different, it goes into the "reindent second line and after" loop - which adds unnecessary indent (the last line of the pasted block in the example).

With the proposed change in this PR, the paste produces the expected result:

<div>
    test
</div>

I hope that helps clarify the issue and the motivation/aim of the PR. For more reported examples, please see issue #32320.

---

EDIT: I found that the proposed change in this PR is not the right solution (yet).

If the pasted block is itself indented, the change will remove it.. This:

    <div>
        test
    </div>

..when pasted becomes:

<div>
    test
</div>

So, I'll need to rethink this - I had a feeling this "solution" was too simple to be true. :) It seems the check needs to be relative to the starting indent size of the pasted block.

OK, I believe I got it - please see next commit.

[eliot-akira]

Nope, those commits are not correct either. OK, I'll continue trying to solve this.

[rebornix]

@eliot-akira thanks for your contribution and great effort! I love reading your thoughts and analysis into this problem and indeed it's really hard. May I ask what's the current state of this pr now?

[eliot-akira]

@rebornix Thank you for reading over this prior effort. To be honest, during the last few commits I wasn't sure anymore if I was going in the right direction. I think this PR is too out of date to pursue further, and can be closed.

But, I believe it was getting closer to the crux of the issue (at least a variant of it), that it's related to the logic of "reindent second line and after".