Fix block scalar mangling bug #231

[wrouesnel]

The regex based parsing for fixing comments was breaking block scalars. By using the ruyaml round trip handler, instead the comment formatting now can correctly identify block-scalars and avoid mangling them.

Resolves #231

Checklist

• Add test cases to all the changes you introduce
• Update the documentation for the changes

[coveralls]

Pull Request Test Coverage Report for Build 4628587345

• 65 of 66 (98.48%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.2%) to 99.634%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/yamlfix/adapters.py	56	57	98.25%

Totals
Change from base Build 4595856345:	-0.2%
Covered Lines:	545
Relevant Lines:	547

---

💛 - Coveralls

[lyz-code]

Thank you for the quick PR @wrouesnel, I'm a bit concerned with the duplication of work though which will make the code slower.

If you find a way to fix this without that duplication I'll be happy to merge this

[lyz-code]

If fixed_source == source I think it will be cleaner to use result == source. If they are different I was not able to tell them apart :S

[lyz-code]

I don't like the re-running of previous fixers, can't you fix comments before so that we don't have to rerun everything?

I think it is really bad from the point of view of the performance of the tool to do this.

[lyz-code]

Looking at the code, weren't you able to tweak the existing _fix_comments function to correctly handle the case instead of rewriting it?

[lyz-code]

Check my comment on this issue in case it might help simplify the pull request

[wrouesnel]

The problem with solving this issue is that the fixer in question takes a line-oriented approach to processing the files, and can't handle multiple lines.

The specific issue is here -

yamlfix/src/yamlfix/adapters.py

Line 584 in 51410d6

# Comment in the middle of the line, but it's not part of a string

            # Comment in the middle of the line, but it's not part of a string
            if (
                config.comments_min_spaces_from_content > 1
                and " #" in line
                and line[-1] not in ["'", '"']
            ):
                line = re.sub(r"(.+\S)(\s+?)#", rf"\1{comment_start}", line)

This code has no way to know if it's part of a string or not, because it can't know if it's part of a block, which has a number of different possible presentations.

I did try several ideas for detecting block-scalar membership, but when you get right down to it that's just re-implementing a half-baked YAML parser when there's a much better YAML parser already out there.

IMO it's more important for the fixer to be correct (i.e. not change key values) rather then fast, but I can't see a way to handle this otherwise (which isn't just trying to half-bake a YAML parser.

[lyz-code]

I'm sorry but I still feel that this PR is adding a lot of complexity to the code only to fix #231. I don't see myself maintaining this code. Why don't you take a simpler approach to have another fixer run after the comments are run that iterates over the source code and:

• If the line contains a comment that starts and ends with #:
  • Checks if the following lines start and end with #
  • Get the length of the longest line
  • Pad the rest of lines to match the desired length