std.typeset.classifyBreaks

BreakKind.CjkCluster() constructs the variant for the existing CJK and East Asian cluster classification. BreakKind.KoreanCluster() is absent and rejected.

Renaming this variant preserves the classification predicates and the effect of the selected Unicode, locale, custom punctuation, and shaping-boundary policies.

A CjkCluster category does not itself permit a break. classifyBreaks returns the opportunities admitted by its supplied policy.

For the same ordered input values, policies, calculation grid, and product release, classifyBreaks and breakParagraph produce the same result. This includes boundary order, selected breaks, line widths, costs, and the returned error variant.

Planning reads measured text and policy values without changing caller arrays. It does not shape text, load fonts, or write output. Ordinary argument evaluation finishes before the operation begins.

The selected cost model, admission rules, and search bounds define plan quality. The two reported plans expose their own costs for inspection.

classifyBreaks reads measured clusters in array order and returns Result<Array<BreakOpportunity>, TypesetError>. It does not shape text, normalize strings, split a cluster, or change any supplied record or array.

Each BreakCluster.text is the scalar string for one indivisible measured unit. Empty text still occupies one cluster index. advance includes trackingAfter. A span width sums its advances and removes the trackingAfter of its final cluster.

advance and advance minus trackingAfter must be nonnegative point lengths. Zero advance and signed trackingAfter are accepted. Invalid extents return TypesetError.InvalidInput. Length arithmetic follows the common Length contract.

classifyBreaks sums stored point ticks and compares them with the stored point targets. It has no DistanceGrid argument and does not quantize to a paragraph grid. Exact equality with a target counts as fitting.

boundaryAllowed has exactly one entry for each cluster. Entry i governs the boundary after cluster i. The last entry must be true for nonempty input. A wrong mask length or a false last entry returns TypesetError.InvalidInput.

Validation checks mask length and terminal permission, word targets, forbidBefore, forbidAfter, generated hyphen indices, and measured extents in that order. Array checks visit entries in input order. Unicode mandatory crossings are checked before boundary conflicts are visited in increasing cluster order. An empty result follows successful structural and policy validation.

An empty cluster array with a valid empty boundary mask returns Ok with an empty opportunity array. It does not create a synthetic unit or terminal opportunity.

Successful nonempty results contain increasing, unique unitEnd values from one through the cluster count. unitEnd is an exclusive cluster index rather than a byte or scalar offset. The final opportunity has unitEnd equal to the cluster count and kind ForcedEnd.

Repeated calls with equal cluster values and equal policies produce the same ordered opportunities and categories. Validation and classification do not read host locale, perform I/O, invoke caller callbacks, or mutate input values. Argument expressions use the ordinary language evaluation order.

A BreakOpportunity carries a boundary and its category. It does not carry a penalty width or BreakRule. A caller preserving mandatory boundaries maps every ForcedEnd opportunity to a Forced penalty before calling breakParagraph.