# Text variables Text variable definitions are objects of class 0xCAB4. Instances placed in text are objects of class 0xCA64, owned by a U+0018 character in the story (`objects.md`, owned items). Implemented in `src/model/variable.rs`. IDML writes each definition as a `TextVariable` element in `designmap.xml`, between the preference elements and `idPkg:Tags`, as the schema orders them. Its `Self` is `dTextVariablen` followed by the name; it has no UID, so definitions were paired with INDD objects by name. ## Evidence All 135 corpus pairs, including those whose IDML is from an older version. They have 1,347 IDML `TextVariable` elements. Each has exactly one class 0xCAB4 object with the same name, and no class 0xCAB4 object lacks an IDML element. The layout below also holds for all 2,496 class 0xCAB4 objects in the 251 distinct little-endian files. ## Definition (chunk 0x28BD) | Field | Contents | |---|---| | u32 *n*, segments | Name: *n* UTF-16 code units as text segments (`objects.md`) | | u32 *n*, segments | Text: the date format of date variables (below) | | u32 | Type code | | u32 *a*, u32 *b*, u32 *c* | Settings (below) | | u32 *n*, segments | `TextBefore` | | u32 *n*, segments | `TextAfter` | The "20 bytes" of the definitions of the earlier pairs are this layout with empty texts. The running header and chapter number definitions of the trustworthy pairs of the corpus of 2026-10 parse to the end of their chunk with it (their settings are all reproduced, below). **Type codes** (each code with its `VariableType` in every pair): | Code | `VariableType` | Pairs | IDML settings element | |---|---|---|---| | 0xCAA1 | `OutputDateType` | 135 | `DateVariablePreference` | | 0xCAA3 | `CustomTextType` | 2 | `CustomTextVariablePreference` | | 0xCAA6 | `FileNameType` | 135 | `FileNameVariablePreference` | | 0xCAA8 | `LastPageNumberType` | 135 | `PageNumberVariablePreference` | | 0xCAA9 | `ChapterNumberType` | 135 | `ChapterNumberVariablePreference` | | 0xCAAA | `MatchParagraphStyleType` | 135 | `MatchParagraphStylePreference` | | 0xCAAB | `ModificationDateType` | 135 | `DateVariablePreference` | | 0xCAAC | `CreationDateType` | 135 | `DateVariablePreference` | | 0xCAB5 | `XrefPageNumberType` | 135 | none | | 0xCAB6 | `XrefChapterNumberType` | 135 | none | | 0xCAC0 | `LiveCaptionType` | 130 | `CaptionMetadataVariablePreference` | **Name.** Equal to IDML `Name` in 1,347 of 1,347. The names of the two cross-reference variables start with U+001B, which IDML writes as `` inside the attribute value (270 of 270). **Order.** IDML lists the variables sorted by name in code point order (135 of 135 pairs; U+001B sorts first). The UID order differs in 113. **Date format.** For the three date types, the text field is IDML `Format` (405 of 405, 9 different formats such as `dd.MM.yy` and `d MMMM yyyy, H:mm`). **Running header style.** For type 0xCAAA, the u32 at offset 4 of the settings is a paragraph style UID. It differs between files, but in all 135 pairs it is the style that IDML writes as `AppliedParagraphStyle="ParagraphStyle/$ID/NormalParagraphStyle"`. ## Settings - `TextBefore` and `TextAfter` are the two texts; IDML writes U+3000 as `^(`. With them the converter reproduces every `TextBefore` and `TextAfter` of the running header and chapter number settings of the trustworthy pairs (1,287 and 1,227 elements), among them chapter numbers with text before or after the number. - Running header (0xCAAA): *b* is the paragraph style; *c* is `SearchStrategy`, 0 `FirstOnPage`, 1 `LastOnPage`; *a* is `ChangeCase`, 0 or 0xCAC9 `None`, 0xCAD0 `Sentencecase`. 1,287 of 1,287 values each in the trustworthy pairs. - Running header `DeleteEndPunctuation`: no field holds it. The settings parse to the end of chunk 0x28BD and the object has no other chunk (107 of 107 definitions checked), so no field is left for it. All 3,568 `MatchParagraphStylePreference` elements in the corpus IDMLs that have the attribute have `false` (2 have none); in the trustworthy pairs, 1,288 of 1,288. The converter writes `false` as a default (`idml-values.md`, defaults) for every running header that matches the sample pattern below or whose settings parse with *c* 0 or 1. Over 5,869 running header definitions in 5,504 little-endian files, *c* is 0 or 1 except in 2 definitions, in files without IDML, where it is 0x40. That may be a flag bit for this attribute; which one is open, so those definitions get no `DeleteEndPunctuation`. - Chapter number (0xCAA9): *a* is `Format`, 0 `Current`; 0x4C15 `Arabic` (one definition, in a trustworthy pair). ## Settings with one value in the corpus The other IDML settings have the same value in every pair: | Element | Attributes | |---|---| | all settings elements | `TextBefore=""`, `TextAfter=""` | | `PageNumberVariablePreference` | `Format="Current"`, `Scope="SectionScope"` | | `ChapterNumberVariablePreference` | `Format="Current"` | | `FileNameVariablePreference` | `IncludePath="false"`, `IncludeExtension="false"` | | `MatchParagraphStylePreference` | `SearchStrategy="FirstOnPage"`, `ChangeCase="None"`, `DeleteEndPunctuation="false"` (`DeleteEndPunctuation`: settings, above) | | `CaptionMetadataVariablePreference` | `MetadataProviderName="$ID/#LinkInfoNameStr"` | | `CustomTextVariablePreference` | an empty `Contents` string | Their INDD fields cannot be located, because nothing varies. The INDD side has one pattern per type in all 2,496 objects of the little-endian files (with or without IDML): | Types | Text field | 20 settings bytes | |---|---|---| | Dates | the format | u32 0 or 0xCAB3, then zeros | | 0xCAC0 | empty | u32 0x8C64, then zeros | | 0xCAAA | empty | u32 0, u32 style, then zeros | | Others | empty | zeros | The exception is custom text variables (0xCAA3): in files without IDML, 5 of 7 have a non-empty text field (for example a person's name or a date), probably the variable's contents, and one has 0xCAB3 in the first u32. The two in the pairs have an empty text field and an empty `Contents`. The converter writes the settings above only for a definition whose fields match its type's pattern. Then the INDD data equals that of the samples, so the IDML settings are taken to be equal too. For any other definition it writes the date format and the running header style (the proven fields) and leaves out the rest, except `DeleteEndPunctuation` (settings, above). A custom text variable with a non-empty text field gets no `CustomTextVariablePreference`: that the text is its contents is not proven. ## Instances (class 0xCA64) | Chunk | Contents | |---|---| | 0x100B | u8 (1 in every sample), then the name of the variable as an in-object string | | 0x288B | u32 UID, not used | IDML writes an instance as a `TextVariableInstance` element in place of the U+0018, with `Self` the instance's UID, `Name` the variable's name and `AssociatedTextVariable` the variable's `Self`. The pairs have 8 instances, in four copies of one document; all 8 match on all three. In the 251 little-endian files, all 25 instance names name a definition of the same file. **Result text.** IDML `ResultText` is the text the variable displays. The INDD does not store it: the four pairs' modification date text (`October 31, 2023 4:06 PM`) does not occur in the INDD files, either as single bytes or as UTF-16. The file name instances (4 of 4) show the INDD file name without its extension, as their settings (`IncludePath="false"`, `IncludeExtension="false"`) say. In the corpus of 2026-10 two more kinds of instance show their text: - A running header on a master page shows `TextBefore`, `<`, the variable's name, `>` and `TextAfter`. - A chapter number in the `Current` format shows `TextBefore`, the document's chapter number (`ChapterNumberPreference`) and `TextAfter`. The converter writes `ResultText` for these and for the file name variables above, and leaves it out (it is optional in the schema) for all others: page numbers, dates and cross-reference page numbers depend on the layout or the export time. Over the trustworthy pairs it reproduces 80 of the 280 `ResultText` values, with 1 different.