Comparing changes

- decode the BrtSSTItem phonetic tail in src/io/xlsb/reader.cpp: the kana is stored once and each PhRun names where its slice starts, so a run's reading ends where the next run's begins and the last runs to the end - treat an empty run array with a non-empty reading as the whole-string case Excel writes when it elides the runs, the same shape <rPh> takes when it spans the entire surface text - emit that tail from src/io/xlsb/sst_writer.cpp behind the RichStr phonetic flag bit, writing the concatenated kana followed by a running ichFirst offset per run - key SstBuilder::intern on the guide as well as the text, so two cells that read the same kanji differently no longer collapse onto a single SST entry - pass cell.phonetic_runs into EmitLiteralCellRecord in cell_writer.cpp so a Text payload carries its guide into the table - add tests/unit/io/xlsb_phonetic_test.cpp over the new xlsb_phonetic fixtures, plus sst_writer cases for the interner key and the tail

- add fm_workbook_set_cell_phonetic_runs plus fm_workbook_get_cell_phonetic_run_count and fm_workbook_get_cell_phonetic_run to src/c_api/formulon_c.h: each <rPh> block travels as an ordered partition of the surface text, spanned in UTF-16 code units - the core has kept one run per <rPh> since 0.11.0 while the bindings carried a single string in each direction, so reading a partially annotated cell and writing it back collapsed every span into one whole-cell annotation - surface the runs as setCellPhoneticRuns / getCellPhoneticRuns on WASM and the Node addon and as set_phonetic_runs / get_phonetic_runs in Python; the flattening getCellPhonetic is unchanged and still returns the readings concatenated - give the native Node addon getCellPhonetic / setCellPhonetic, the last cell-level pair that existed on WASM and Python only, and shrink the WASM-only list in tools/dev/check_binding_drift.py and the npm-native README method counts to match - add fm_styles_set_font, which overwrites an existing font slot in place, and fm_workbook_set_default_font, which declares font 0 -- the record every unstyled cell resolves to, seeded as Calibri 11 -- that fm_styles_add_font could never reach because it only appends - surface those as setFont / setDefaultFont (WASM + Node addon) and set_font / set_default_font (Python); the current default reads back through the existing getFont(0) - extend the C API, WASM, Node and Python test surfaces to cover the new entry points

… font - record the Unreleased Added entries for fm_workbook_set_cell_phonetic_runs and the two run readers, the native Node addon gaining getCellPhonetic / setCellPhonetic, the BrtSSTItem phonetic tail surviving an xlsb round trip, and fm_workbook_set_default_font / fm_styles_set_font - name the binding-level spelling of each entry point alongside the C ABI one, so a reader on any surface can find it

- add PhoneticProperties to src/phonetic.h -- ruby font id, kana form and alignment as ordinals -- hang it off Cell beside phonetic_runs, and give Sheet a set_cell_phonetic_props that is independent of the readings, so editing the kana does not reset how it renders - only the <rPh> runs were carried before, so a guide set to hiragana or to a distributed layout came back as Excel's default halfwidth katakana on the next save - read the element in the DOM, SAX and shared-string paths, and emit it beside every non-empty run list with all three attributes spelled out rather than left to be inferred, which is what Excel writes - an absent <phoneticPr> resolves to halfwidthKatakana / noControl while a bare one falls back per attribute to fullwidthKatakana / left; the two states are different and are now read apart - put the OOXML attribute vocabulary in the new src/io/phonetic_pr.h so the model header stays free of XML; XLSB packs the same ordinals into BrtSSTItem's trailing (ifnt, flags) pair, so the binary path shares the struct without a mapping table - decode that pair in src/io/xlsb/reader.cpp leniently -- a record stopping short of it keeps the defaults rather than failing the load -- and emit the entry's own values from sst_writer.cpp in place of a fixed 0x0030 - key both shared-string interners on the properties as well as the text and the runs, so one reading rendered as hiragana and another as katakana stay separate string items - compare the font record's theme scheme in the xlsb fidelity and cross-format symmetry checks

- read <scheme val="major|minor"> into the new FontRecord::scheme and write it back as the last child of <font>, after <family> and <charset>, which is the order Excel emits them in - a ja-JP workbook's Normal font carries scheme="minor", which is what makes Excel show it as the body font and re-resolve it when the theme changes; the element was dropped on read, so re-saving rewrote the font as a literal name - collapse an unrecognised value to 0, which emits no element at all: writing an uninterpretable link back would leave a theme reference Excel cannot resolve - decode BrtFont's bFontScheme into the same field and emit it from the xlsb writer instead of a hardcoded 0; the ordinals match, so the binary path needs no mapping table and the field is no longer listed among the record's unmodelled ones - cover minor, major, an unrecognised value and a font with no link on both the reader and the writer

- add scheme to fm_font_record and carry it through font_to_c and font_from_c, so a read-modify-write of a font over the C ABI no longer unlinks it from the workbook theme - project it as scheme on WASM and the native Node addon and as FontRecord.scheme in Python, defaulting to 0 (no link) on each - the struct grew by one field, so the Python layout expectations for FONT_RECORD and DXF_RECORD and the C ABI offset assertions move with it - state on fm_styles_set_default_font that a workbook Formulon created carries no theme part, so leave scheme at 0 there; a package loaded with its own theme keeps that part, and preserving the link is what keeps Excel showing the font as the theme's body font - exercise the read-modify-write path over the ABI, down to the <scheme> element in the saved styles part

- state in src/print/pagination.cpp that MDW is a property of the Normal font, so a workbook whose Normal style names a different one resolves a different number of points per character - record the measured scope: at 11 pt the family does not move the width (Calibri, Yu Gothic and MS PGothic resolve a 30-unit column alike), while the point size does move it roughly in proportion and the family starts to matter away from 11 pt - note that those observations are Mac Excel's, a different column-geometry regime from the Windows primary oracle the constants come from, so they establish the dependency without supplying the numbers to model it - until a Windows capture over the same sweep exists, a workbook whose Normal font is not 11 pt paginates against the 11 pt geometry

- add an Unreleased Fixed entry for the <scheme> theme link surviving a load and save through both containers and reaching the record projections on WASM, the native Node addon and Python - add the <phoneticPr> entry: the ruby font, kana form and distribution now round-trip, and an absent element is read apart from a bare one

- add fm_workbook_set_cell_phonetic_properties / fm_workbook_get_cell_phonetic_properties, carrying the OOXML <phoneticPr> triple: the font that draws the ruby, the kana form (FM_PHONETIC_TYPE_*) and how the kana is distributed over the characters it covers (FM_PHONETIC_ALIGNMENT_*); scalar arguments rather than a new struct, matching fm_workbook_set_iterative - project it as setCellPhoneticProperties / getCellPhoneticProperties on WASM and the native Node addon, and as set_phonetic_properties / get_phonetic_properties with a PhoneticProperties dataclass on Python - keep it independent of the run entry points in both directions: editing the readings does not reset the rendering, and setting the rendering does not touch the readings; a value-setting call still clears both - reject an out-of-range type, alignment or font_id rather than truncate it, since each of the two enumerations occupies two bits of the XLSB record's trailing flags word and the font id sixteen - report the defaults beside a failure status in the read result instead of dropping the payload keys the return type declares - the core already round-tripped these properties, so this adds only the authoring surface

- add the non-oracle deferred-feature entry pagination_column_width_calibrated_to_one_normal_font - Excel resolves a column's width in points from its stored character-unit width times the Normal font's maximum digit width, so the conversion is a property of that font and pagination carries one calibration for it - measured against Mac Excel 365 16.112.1 by opening workbooks that differ only in font 0: at 11 pt the family is not observable, since Calibri and both Japanese body faces resolve the same width; the point size is, and the family becomes observable away from 11 pt - closing it needs the same sweep captured on the Windows primary oracle, whose column geometry is a different regime from the Mac numbers

- record under Unreleased / Added that how a phonetic guide renders can now be authored, not only round-tripped, naming the C ABI pair and its WASM, Node addon and Python projections - state that it stays separate from the run entry points in both directions and that a value write still clears both

- Bump the version to 0.11.1 across CMake, both npm packages, the Python project and the embedded version header - Close the changelog's unreleased section as 0.11.1 and add its compare link - Index v0.11.1 as the latest release

Navigation Menu

Repository navigation

Comparing changes

Open a pull request

Commits on Aug 22, 2026

This comparison is taking too long to generate.

Uh oh!