ModernJsonInVBA version 3.0.0

Changelog

All notable changes to ModernJsonInVBA are recorded here. The format follows
Keep a Changelog, and the project uses
semantic versioning.

3.0.0 - 2026-07-05

This release reorganizes the library into eleven modules split by concern and
rewrites the parser and Excel ingestion path for speed. The public API is
unchanged, so existing calling code runs without edits. The module split is
the only breaking change: you import eleven files instead of pasting one.

Breaking

• The library ships as eleven .bas modules (in vba_source/) instead of the
  single zz_ModernJsonInVBA module. Remove the old module and import all
  eleven. Function names, arguments, return shapes, and error numbers are
  unchanged, so calling code does not change.

Added

• json_payloads/: a seeded payload generator (generate_payloads.py) and a
  workbook macro suite (Run_JsonPerfSuite, Run_JsonPerfMatrix) that print
  Markdown timing reports to the Immediate window.
• PERFORMANCE.md: a measured timing table across the JSON parser and Excel
  ListObject surface, regenerable from the workbook.
• vba_source/: the eleven modules as individual files for import and version
  control.

Changed

• Module layout by concern: Json_Common (shared plumbing), Json_Parser,
  Json_Serializer, Json_Model, Json_Transforms, Json_Tables,
  Json_Coalesce, Json_Csv, Json_Xml, Json_Excel (table ingestion), and
  Json_Excel_Export (table and range export). The README lists what each
  module owns.
• Rewrote the module and README comments and removed dead procedures left over
  from the single-module layout.

Fixed

• The 32-bit FNV hash used LongLong, which does not compile on 32-bit Office.
  It is replaced with a Long-only rolling hash, so every module now compiles on
  both 32-bit and 64-bit Office.
• CSV conversion escapes control characters, so CsvTextToJson output always
  parses back through Json_Parse. Previously a tab or other control byte in a
  field produced JSON that the parser rejected.

Performance

Measured against the previous single-module release on a Ryzen 7 9800X3D with
64-bit Excel. Times are hardware dependent; see PERFORMANCE.md for the full
table and method.

• Table ingestion streams JSON text directly into a 2D array for the common
  tableRoot = "$" case and does not build the intermediate object model. A
  per-row key cache reuses the column layout from one row to the next.
• Large table writes go to the sheet in one block. The former 50,000-cell
  chunking is removed; a block-write fallback remains for memory-constrained
  hosts.
• Character scanning reads UTF-16 code units from a byte snapshot of the input
  rather than allocating a one-character string per position.
• JSON parsing is about 4x faster. Table-row extraction and CSV conversion,
  which previously scaled quadratically on large inputs, are about an order of
  magnitude faster.
• A 500,000-row, 110 MB document loads into a ListObject in about 18 seconds on
  the benchmark machine.