Try: Improve round-trip transformation of text and quote blocks

[westonruter]

With this PR, given a quote block:

Converting it to a text block yields:

This introduces a <br class="citation-break"> element to delineate the original quotation from the citation so it can be converted back into a quote:

There are two opening p tags there. So this is one known issue with the patch. It generates this in the console:

Warning: validateDOMNesting(...): <p> cannot appear as a descendant of <p>.

Fixes #1765.

[youknowriad]

Sorry forgot about this PR :( I should have reviewed a long time ago.

These transforms changed master, a rebase is necessary right now.

[gziolo]

@iseulde or @mcsf is this something we still need to include? Can we help to merge that one?

[mcsf]

Hey, @westonruter. Having a look at this now. I disagree with the proposed solution in that:

• I don't believe we should try to port the intricacies of one block to the one we are transforming to. Sometimes these operations will be destructive—typically when the source block encodes more essential state—and that's fine.

• Arguably, this proposal seeks to make Paragraph blocks conform to a certain shape they weren't intended for. Not that line break isn't supported—it is—but here we are furthermore injecting an invisible mark, effectively entangling user content with implementation details of the editor and the transform stack.

• A merged PR, Fixed quote paragraph transformations. #3802, opens the precedent for some smarter transforms involving quotes. In it, transforming a list block with items A, B, C will yield the quote with text A & B and citation C. Now that multi-block transforms are supported (Add multi-block transform functionality. #3357), we could consider transforming a sequence of paragraphs into a quote with citation, and thus the opposite transform—from quote to paragraphs—would just need to output two blocks, one for each field.

• In parallel, we could look into smarter optimizations for the benefit of the user that don't require encoding application state within the block. For instance, if a type switch is requested from paragraph to quote, Gutenberg can look at whether the paragraph had initially been converted from a quote; if so, and if it hasn't been touched by the user, we can just use the undo buffer to recover the original quote, with no information loss. This would have to be made in a simple way, relying on the existing history subsystem and avoiding adding overhead.

I recommend closing this PR and exploring alternatives.

[youknowriad]

Thanks for working on this, I'm closing per @mcsf comments. Let's try alternatives.