[vagenas]

Defines whitespace handling similarly to HTML: outside <content>, whitespace collapses to one space, also across inline tags. So far the spec left whitespace to applications, and producers and consumers disagreed, especially between raw text and nested elements.

Spec

• Terms: a structural rule for when an element that allows raw text is inline or block-level. A virtual <text> is block-level, and the spoken text of a transcript turn is handled like a block-level body.
• Whitespace handling: four rules (ignore whitespace outside block-level bodies and at their start and end, read the rest as a single space across inline tags, attribute the space to the element holding its first character), plus <content> read literally. Includes an example.
• Tracks: the "crossing the hour mark" example is retitled, since its two lines now read as one unattributed turn.
• Examples whose meaning would change are fixed (head and body, cross-references, hyperlinks), and code and formula <content> no longer starts with a line break and indentation.
• Annex C: recommendation on line breaks, indentation and the minimized canonical form.

Migration between 0.8 and 0.9

• doclang migrate --to 0.9 reads 0.8 documents as 0.8 guaranteed: only <content> keeps whitespace, every other text is stripped and the pieces joined directly. It removes the whitespace that 0.9 would read as a space (next to inline tags) and keeps what 0.9 ignores (element head, body edges, list/cell delimiters, elements without raw text) and whitespace between words, so layout stays as far as possible.
• <content> that is no longer needed (a single space next to an inline tag) is unwrapped where 0.9 reads the plain text the same.
• doclang migrate --to 0.8 keeps the reading under 0.8's rules: each space that 0.8 would strip (whitespace at the edge of a text, next to an inline tag) becomes <content> </content>, where its first character is, so it keeps its formatting. Whitespace that 0.8 ignores and whitespace between words stays. In 0.8, <field_heading> cannot hold <content>, so there the space goes into the neighbouring formatting element (recorded as mapped).
• Tests for how transcript turns read (edges, cue block and <voice> boundaries, line breaks).
• Test cases in pretty, minified and hybrid layouts, in both directions; a test reader of the 0.8 rules checks that downgrading keeps the reading of the cases and fixtures, and upgrading again removes the added <content>; a test reader of the 0.9 rules checks that unwrapping never changes the reading of the cases and fixtures.

BEHAVIOR CHANGE: whitespace in raw text, previously left to applications, now collapses to a single space as in HTML; <content> is read literally.