ICU-2134 document .res file format, add alias type description

X-SVN-Rev: 10690

icu4c/source/common/uresdata.h

@@ -39,10 +39,67 @@ typedef uint32_t Resource;
 #define RES_GET_UINT(res) ((res)&0x0fffffff)
 
 /*
+ * File format for .res resource bundle files (formatVersion=1)
+ *
+ * An ICU4C resource bundle file (.res) is a binary, memory-mappable file
+ * with nested, hierarchical data structures.
+ * It physically contains the following:
+ *
+ *   Resource root; -- 32-bit Resource item, root item for this bundle's tree;
+ *                     currently, the root item must be a table resource item
+ *   char keys[]; -- up to 65k of characters for key strings,
+ *                   which consist of invariant characters (ASCII/EBCDIC) and are NUL-terminated;
+ *                   padded to multiple of 4 bytes for 4-alignment of the following data
+ *   data; -- data directly and indirectly indexed by the root item;
+ *            the structure is determined by walking the tree
+ *
+ * Each resource bundle item has a 32-bit Resource handle (see typedef above)
+ * which contains the item type number in its upper 4 bits (31..28) and either
+ * an offset or a direct value in its lower 28 bits (27..0).
+ * The order of items is undefined and only determined by walking the tree.
+ * Leaves of the tree may be stored first or last or anywhere in between,
+ * and it is in theory possible to have unreferenced holes in the file.
+ *
+ * Direct values:
+ * - Empty Unicode strings have an offset value of 0 in the Resource handle itself.
+ * - Integer values are 28-bit values stored in the Resource handle itself;
+ *   the interpretation of unsigned vs. signed integers is up to the application.
+ *
+ * All other types and values use 28-bit offsets to point to the item's data.
+ * The offset is an index to the first 32-bit word of the value, relative to the
+ * start of the resource data (i.e., the root item handle is at offset 0).
+ * To get byte offsets, the offset is multiplied by 4 (or shifted left by 2 bits).
+ * All resource item values are 4-aligned.
+ *
+ * The structures (memory layouts) for the values for each item type are listed
+ * in the table above.
+ *
+ * Nested, hierarchical structures: -------------
+ *
+ * Table items contain key-value pairs where the keys are 16-bit offsets to char * key strings.
+ * Key string offsets are also relative to the start of the resource data (of the root handle),
+ * i.e., the first string has an offset of 4 (after the 4-byte root handle).
+ *
+ * The values of these pairs are Resource handles.
+ *
+ * Array items are simple vectors of Resource handles.
+ *
+ * An alias item is special (and new in ICU 2.4): --------------
+ *
+ * Its memory layout is just like for a UnicodeString, but at runtime it resolves to
+ * another resource bundle's item according to the path in the string.
+ * This is used to share items across bundles that are in different lookup/fallback
+ * chains (e.g., large collation data among zh_TW and zh_HK).
+ * This saves space (for large items) and maintenance effort (less duplication of data).
+ *
+ * --------------------------------------------------------------------------
+ *
  * Resource types:
+ *
  * Most resources have their values stored at four-byte offsets from the start
  * of the resource data. These values are at least 4-aligned.
  * Some resource values are stored directly in the offset field of the Resource itself.
+ * See UResType in unicode/ures.h for enumeration constants for Resource types.
  *
  * Type Name            Memory layout of values
  *                      (in parentheses: scalar, non-offset values)
@@ -52,13 +109,19 @@ typedef uint32_t Resource;
  * 1  Binary:           int32_t length, uint8_t[length], (padding)
  *                      - this value should be 32-aligned -
  * 2  Table:            uint16_t count, uint16_t keyStringOffsets[count], (uint16_t padding), Resource[count]
+ * 3  Alias:            (physically same value layout as string, new in ICU 2.4)
  *
  * 7  Integer:          (28-bit offset is integer value)
  * 8  Array:            int32_t count, Resource[count]
  *
  * 14 Integer Vector:   int32_t length, int32_t[length]
  * 15 Reserved:         This value denotes special purpose resources and is for internal use.
- 
+ *
+ * Note that there are 3 types with data vector values:
+ * - Vectors of 8-bit bytes stored as type Binary.
+ * - Vectors of 16-bit words stored as type Unicode String
+ *                     (no value restrictions, all values 0..ffff allowed!).
+ * - Vectors of 32-bit words stored as type Integer Vector.
  */
 
 /*