From 6ed802dbb0568a71afcf84a75470fc6d5cf75b34 Mon Sep 17 00:00:00 2001 From: Markus Scherer Date: Mon, 16 Dec 2002 22:43:08 +0000 Subject: [PATCH] ICU-2134 document .res file format, add alias type description X-SVN-Rev: 10690 --- icu4c/source/common/uresdata.h | 65 +++++++++++++++++++++++++++++++++- 1 file changed, 64 insertions(+), 1 deletion(-) diff --git a/icu4c/source/common/uresdata.h b/icu4c/source/common/uresdata.h index 92acc0f8578..531762b8f28 100644 --- a/icu4c/source/common/uresdata.h +++ b/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. */ /*