These are defects in the current format, not opinions. Each is stated with what it would break.
| # | Hazard | Consequence |
|---|---|---|
| H1 | No magic bytes. The file is detected by extension only (MADocument.m:459). |
Windows has no meaningful .mad association, and a renamed file is unidentifiable. See §7. |
| H2 | WAL and SHM are shipped. analyitics-wal is 313 KB against a 516 KB database. |
A reader that ignores the WAL reads stale data. A CLI must either replay the WAL or require a checkpoint before archiving. |
| H3 | Core Data schema inside. | Ties every reader to Core Data, which does not exist off Apple platforms. The format cannot be called portable until this is addressed. |
| H4 | Hidden files are skipped (MADArchiveManager.m:111). |
Any future hidden member is silently dropped from the archive. |
| H5 | analyitics is misspelled on disk. |
Cannot be corrected without breaking existing documents. |
| H6 | version is a double in the writer, an integer in the samples. |
Consumers must accept both. |
| H7 | sqlite compile flags are not pinned. build_sqlite3.sh declares 20 defines; the MADSqlite3 target applies one. |
The engine that writes .maddb uses a 4096-byte page size where the script asks 8192, and is missing FTS5 and STAT4. The page size is written into the file header, so the format is defined by accident. Now fixed: the defines live in Compressor/sqlite3.xcconfig, which both the target and the script read. Caches written since have 8192-byte pages; older ones still open, because SQLite reads the page size from the header. See docs/project/LIBRARIES.md §4. |