Format:ICU data file (icudt)
| Magic | DA 27 at offset 2; dataFormat 'CmnD' (43 6D 6E 44) in the UDataInfo |
|---|---|
| Extensions | .dat, .icu |
| Endianness | either, flagged in the header; file name letter l little, b big, e EBCDIC |
| License | Unicode License |
| Website | https://unicode-org.github.io/icu/userguide/icu_data/ |
| On this wiki | Facts · Code · Notes · Topics |
ICU's common data package (icudt<NN><l|b|e>.dat, icudtl.dat in Chromium/Android/Qt WebEngine builds): a data header with headerSize, magic bytes DA 27 and a UDataInfo block (endianness, charset, data format 'CmnD', versions), then a table of contents of named items (collation, break rules, converters, locale resources).
Overview
ICU (International Components for Unicode) keeps locale, collation, normalization, break-iterator and charset-conversion data in 'data files'. Applications link ICU data as a DLL/shared object, or load a common package file at run time; it is the biggest part of ICU and appears in Chromium-based apps (Library:Chromium Embedded Framework, Library:Electron) as icudtl.dat.
Name and version
The package name is "icudt" U_ICU_VERSION_SHORT U_ICUDATA_TYPE_LETTER: the version digits without dots (major, or major and minor for old releases, so ICU 74 is icudt74) and one letter: l little-endian ASCII, b big-endian ASCII, e big-endian EBCDIC. So icudt74l.dat is ICU 74 data for x86/ARM. The file name (and the entry names inside, icudt74l/ucadata.icu) therefore identify the ICU version a program was built with; the ICU library exports are suffixed the same way (u_strlen_74).
Header
Every ICU data item starts with a DataHeader: uint16 headerSize (in the file's byte order), uint8 magic1 = 0xDA, uint8 magic2 = 0x27, then a UDataInfo: uint16 size, uint16 reservedWord, uint8 isBigEndian, uint8 charsetFamily (0 ASCII), uint8 sizeofUChar (2), uint8 reservedByte, uint8 dataFormat[4], uint8 formatVersion[4], uint8 dataVersion[4]. For a common package the dataFormat is CmnD ('C' 0x43, 'm' 0x6D, 'n' 0x6E, 'D' 0x44, formatVersion[0] = 1); an older pointer-table package uses ToCP. Individual items have other formats (ResB resource bundle, Norm, Brk , UCol ...) with their own dataFormat. See Fact:ICU data file (icudt)/struct/DataHeader.
Table of contents (CmnD)
At headerSize after the start: uint32 count, then count pairs uint32 nameOffset, uint32 dataOffset (offsets relative to the start of the package), sorted by name so lookup is a binary search; each item itself begins with its own DataHeader. Names are like en.res, root.res under the package prefix.
Code & repositories
- https://github.com/unicode-org/icu @
ad484f3d6c036ba06193ac7c3bc2233a53ab68b8· license: Unicode-3.0 (declared by the contributor for the files used; GitHub reports Unicode-3.0 (read from LICENSE; GitHub's detector did not classify it) for the repository)
ICU sources; icu4c/source/common ucmndata.h/.cpp and unicode/udata.h define the data header and common package layout.
Tested on: source read, not run: source read · checked 2026-10-18; last activity 2026-10-18; Note: the latest release ships binaries; the wiki links only the source, never the binaries
Pitfalls
- A reader rejects the file when
isBigEndianorcharsetFamilydiffers from the build (U_INVALID_FORMAT_ERROR), so a big-endian .dat does not work on x86; the 'l' in the name is not enforced by the loader. - Unverified: that headerSize is 32 in current packages (24 bytes of fields rounded up); read the field instead of assuming it.
Evidence & references
- https://github.com/unicode-org/icu at commit ad484f3d6c036ba06193ac7c3bc2233a53ab68b8 (Unicode License):
icu4c/source/common/ucmndata.h(DataHeader, MappedData, UDataOffsetTOC),ucmndata.cpplines 331-362 (udata_checkCommonData: 0xda 0x27, 'CmnD', 'ToCP'),unicode/udata.h(UDataInfo),unicode/utypes.hlines 112-156 (U_ICUDATA_NAME, type letters).