Phonetic guides (furigana) survive a read and a save

Cluster: xlsx-io

Scenario

Japanese Excel records the reading of text typed through the input method beside the text itself: a string holding 漢字 also carries the furigana かんじ. The reading is what Japanese Excel sorts such text by and what the PHONETIC() function returns, and a user can show it above the cell. A workbook of names or addresses read by the library and saved again comes back with its text intact but every reading gone.

In the package the reading is part of the string, not of the cell. A CT_Rst (a shared string's <si>, an inline string's <is>, a note's <text>) holds, after its base text or runs, one <rPh> per annotated span, sb and eb giving the span's start and end in the base text and a <t> its reading, then a <phoneticPr> naming the font the reading is drawn in (fontId, an index into the stylesheet's fonts), its script (type) and its alignment. Whether the reading is shown is a separate flag: ph on a <c> and on a <row>, phonetic on a <col>. A binary workbook carries the same data in the phonetic tail of a RichStr.

Spec note, not a corpus case: the reader already keeps a reading out of the text it annotates (phonetic-run-not-read-as-cell-text locks that), so nothing is corrupted. What is missing is a model for the reading, which is an API decision with no recorded demand yet; the open questions below are that decision. It becomes a corpus case once a reading has somewhere to live, against a fixture Excel saves with readings set through Range.Phonetics and shown through Range.Phonetic.Visible.

Desired behavior

• A string read with phonetic runs keeps them, each span and its reading, and its <phoneticPr>, and a save writes them back in the same container they were read from.
• The show-reading flags on cells, rows and columns survive a save.
• Two cells with the same text but different readings do not share one shared-string entry, and two cells with the same text and reading still do.
• Changing a cell's text does not leave a reading describing text the cell no longer holds.
• A reading's font survives the stylesheet being re-interned on write, since fontId is an index the writer renumbers.

Open questions

• Where a reading lives. A field on the string value makes every furigana-bearing cell in a Japanese workbook something other than a plain string, which every consumer testing typeof value would trip over. A facet on the cell, beside note, keeps the value a string but has to say what happens to the reading when the value is reassigned: dropped, as the fourth bullet above wants, or kept.
• Whether a rich-text value's reading rides on the value (it is already an object) while a plain string's rides on the cell, or both on the cell.
• How <phoneticPr fontId> is modelled: resolved to a Font on read and interned on write, or kept as an index into the preserved font table.
• Whether a note's body, which is plain text by model, keeps a reading at all.
• The streaming writer, which interns strings row by row, and the binary reader, which leaves the RichStr phonetic tail unread today.

Related: phonetic-run-not-read-as-cell-text.