Content conversions

@openpresentation/opf/convert (RR-26) holds the pure, deterministic converters behind the editor's "Content type" control, the list level keys, grouping, image-to-design and the slide split and merge. They need no renderer, fonts, DOM, network or model: a function takes OPF JSON and returns new OPF JSON.

Every conversion follows one contract:

• Never invents content. Every word, number and date in the result came from the source. The only generated text is the fixed column headings of a timeline or metric table (When, What, Description, Label, Value, Unit, Delta, Trend, Sentiment), the --- row of a Markdown table, and the title= of a code fence; a heading row can be left off (headings: false).
• Reports what it cannot carry. A result has lossless and loss, a list of names such as text formatting or list nesting levels. lossless is true exactly when loss is empty. Show loss before applying.
• Refuses with a reason. A pair with no mapping, or content that does not fit, throws OPFConversionError (code: "not-convertible") with a message that says what to change. contentConversionTargets() returns the same refusal as available: false and reason.
• Validates its output as OPF. A result that would not validate throws code: "invalid-output".
• Is pure. The input is never changed and the same input gives the same output. Slide-level functions take and return plain slide or presentation objects; the host turns the difference into a patch, so a conversion is one undoable step (the editor does this in convertBlock).

import { contentConversionTargets, convertContent } from "@openpresentation/opf/convert";

const block = { items: ["Plan", { text: "Ship", description: "By June" }] };
contentConversionTargets(block);
// [{ kind: "text", lossless: false, loss: ["list item descriptions (kept as indented lines)"], ... },
//  { kind: "timeline", ... }, { kind: "table", lossless: true, loss: [] }]
const { payload, lossless, loss } = convertContent(block, "table");
// payload: { table: { rows: [["Plan", null], ["Ship", "By June"]] } }

A payload is a block, a slide or region that holds one content field, or a group of metric blocks. id, extensions, type (kept in step with the new kind) and any slide fields (title, design, notes, ...) stay on the result.

Block conversions#

Images, videos and any other group have no content conversion.

Other helpers#

Slide-level functions return the replacement slides, the replaced range (start, deleteCount) and the new presentation; structure functions return the new slide and the path of what they touched.

Decisions#

• Column headings of timeline and metric tables are the schema's own field names. They are the only generated text and can be omitted; reading a table back needs headings (or explicit columns), because a role is never guessed from position.
• Text-to-timeline and text-to-quote parse only unambiguous patterns; the rest stays text. Parsing a pattern is not loss, and a wrong guess would be invented structure.
• Code to text writes a fenced block when the code has a language or file name, so the round trip is lossless. A fence and its title= are the visible cost; fences: "never" writes the bare source and reports the loss.
• Un-paginate needs the pagination mapping. Without it, mergeSlides concatenates blocks and never glues text together, because pagination's exact slice boundaries are not recoverable from the slides.