Olson timezone data should be regularly updated in the trunk. If the data is out of date, follow the instruction described in the            readme file.

Once the timezone resource file is generated by the process above, the file should be checked into zoneinfo.txt in the trunk.

ICU4C:

1. Update the API documentation in all header files (.h file) to have correct @draft/@stable/@deprecated labels.
2. Update docmain.h
3. ./configure
4. make docs
5. Follow instructions in readme.txt to generate API status change report.
6. Make sure that ICU headers work with U_HIDE_DRAFT_API and other such switches.
7. Verify that U_DRAFT and U_STABLE match the @draft and @stable tags (same for other such pairs declaration macro vs. tag).

ICU4J:

Update the API documentation to have correct @draft/@stable/@deprecated labels. On ICU4J, run com.ibm.icu.dev.tool.docs.CheckTags (see file for instructions). This requires a Sun jdk with javadoc available. The tool will need to change to reflect the release number to search for.

To check the API status changes, run the ant target "apireport" to gerate the report since the previous official release.

Make sure @provisional is specified along with @draft, and @deprecated with @internal. For example,

 * @draft ICU 4.0
 * @provisional This API might change or be removed in a future release.
          

 * @internal
 * @deprecated This API is ICU internal only.
          

Run the ICU4J versus JDK API comparison tool against the target JDK (anything that will come out before our next release, basically) with the tool com.ibm.icu.dev.tool.docs.ICUJDKCompare and make sure ICU4J adequately covers the JDK API for the classes we replicate.

ICU4C:

The instructions for updating the version number are stored in icu/source/common/unicode/uverdefs.h. Read and follow instructions in this file.

ICU4J:

1. common.properties
   
2. com.ibm.icu.util.VersionInfo.
3. build.properties. (For API change report)

For ICU 3.6, this is the text of the public beta announcement:

Subject: ICU 3.6 beta now available

Dear friends and users of ICU,

We have a "beta" of the ICU release 3.6 available on our public web sites, http://icu-project.org/download/3.6.html and http://icu-project.org/download/
If you plan or consider to use ICU 3.6, then please download and test it at your earliest convenience.

Please help us ensure the quality of this release. Please submit bugs, porting changes and fixes(!) by Wednesday, July 31!

Please use our bug reporting tool and support mailing lists, see http://icu-project.org/

Major features of this reference release include:
- Unicode 5.0
- CLDR 1.4
- and so on...

Please see the release notes on the download pages and the readme for a complete list of features and enhancements.

Sincerely,
markus

If any of the data files in /icu/source/data/ directory has changed MANUALLY, upgrade the version number accordingly as well. If the contents of a resource bundle has changed, then increase the version number (at least the minor version field). The CLDR generated data should have the correct number.

Note from Markus (20090514, ICU 4.2 timeframe): Most data files automatically get their version numbers set by the LDML2ICUConverter, from CLDR version numbers. It is not clear what, if anything, needs to be done for this task.

If the format of a binary data file has changed, upgrade the format version in the UDataInfo header for that file. Better: Change the format version immediately when the format is changed. The change must be made in both the producer/generator and consumer/runtime code.

It is desirable to maintain backward compatibility, but sometimes impractical. Update the major version number for incompatible changes. Runtime code should permit higher minor version numbers for supported major versions.

We rarely use the third and fourth version number fields, except for UTrie (version 1) parameters that don't really change.

Check source and data files, especially newly contributed ones, to make sure the proper copyright notice is in place. For example,

Copyright (c) 2001-2008, International Business Machines Corporation
and others. All rights reserved.

Scan the source code to make sure that every file that was touched recently has the current year in the copyright statement. See the ICU Copyright Scanner page and follow the link to the scripts and the readme.

Scan the source code to include third-party company names in copyright notices if necessary. 

Scan for text files that do not contain the word "Copyright": find_textfiles -nosvn -novc | xargs grep -i -L Copyright (See the find_textfiles Perl script attached to this page.)

Make sure that you use (c), not ©, because non-ASCII characters in source code can break the build!
(See ICU's check-in policy for more details.)

Check ICU, Unicode and other license terms. Make sure these files are up to date. The Unicode data and software license term is updated anually (usually year number only).

>java com.ibm.icu.dev.tools.misc.NonAsciiFileDetector <icu4c|icu4j source root path>
          

Verify the following for library code (common, i18n, layout, ustdio). The requirement is for ICU's memory management to be customizable by changing cmemory.h and the common base class.

• No raw malloc/free/realloc but their uprv_ versions.
• All C++ classes must inherit the common base class UObject or UMemory
  • But not for mixin/interface classes (pure virtual, no data members, no constructors) because that would break the single-inheritance model.
  • Also not for pure-static classes (used for name scoping) declare but don't implement a private default constructor to prevent instantiation.
  • Simple struct-like C++ classes (and structs) that do not have constructors, destructors, and virtual methods, need not inherit the base class but must then be allocated with uprv_malloc.
• All simple types (UChar, int32_t, pointers(!), ...) must be allocated with uprv_malloc and released with uprv_free.
• For Testing that this is the case, on Linux
  • run the Perl script ICUMemCheck.pl. Follow the instructions in the script. The script is in in the ICU4C sources at icu/tools/memcheck/ICUMemCheck.pl.
• For testing that this is the case, on Windows:
  • Don't bother, as of Nov, 2004. Failures appear in many dozens of files from the mixin class destructors. Do the check on Linux. But, for reference, here are the old instructions.
    • Make sure that in uobject.h UObject::new and delete are defined by default. Currently, this means to grep to see that U_OVERRIDE_CXX_ALLOCATION is defined to 1 (in pwin32.h for Windows).
    • Check the *.obj's for linkage to the global (non-UObject::) operators new/delete; see uobject.cpp for details.
    • Global new must never be imported. Global delete will be imported and used by empty-virtual destructors in interface/mixin classes. However, they are not called because implementation classes always derive from UMemory. No other functions must use global delete.
    • There are now (2002-dec-17) definitions in utypes.h for global new/delete, with inline implementations that will always crash. These global new/delete operators are only defined for code inside the ICU4C libraries (but must be there for all of those). See Jitterbug 2581.
    • If a global new/delete is used somewhere, then change the class inheritance and/or use uprv_malloc/free until no global new/delete are used in the libraries (and the tests still pass...). See the User Guide Coding Guidelines for details.

• Build ICU with debug information. On Linux,
     runConfigureICU --enable-debug --disable-release Linux
• Run all of the standard tests under valgrind.  For intltest, for example
     cd <where ever>/source/test/intltest
     LD_LIBRARY_PATH=../../lib:../../stubdata:../../tools/ctestfw:$LD_LIBRARY_PATH  valgrind  ./intltest
  You can grab the command line for running the tests from the output from "make check", and then just insert "valgrind" before the executable.
  

• Apply the latest tz update patch to the JRE.
• Run the ICU4J test cases using ant target "jdktzCheck"
  

1. Make sure the Eclipse ICU4J plug-in binaries are successfully produced.
2. Run the ICU4J plug-in test cases.
3. Update license files, build version strings for the new release.

ICU4C:

ICU data should be removed, so that tests cannot access it. Both cintltst and intltest should be run with -w option and they should not crash. Every crash should be investigated and fixed.

ICU4J:

ICU4J has the test target for this, but does not work as designed for now. For now, this task is not required for ICU4J.

You need a working snapshot of the ICU build first. To do so, you need to check out the trunk and build it on all the reference platforms as stated in the readme.html.

Once the branch is created, only changes necessary for the target release are taken from the trunk.

ICU4C:

Build the API documentation pages for the new release. Run Doxygen to create the javadoc files. Create icu4c-X_X_X-docs.zip

ICU4J:

Build the API documentation pages for the new release. "ant docsJar"

Testing external dependencies in header files:

(on Unixes) Prerequisite: Configure with --prefix (../trunk/source/runConfigureICU Linux --prefix=/some/temp/folder) and do 'make install'. Then set the PATH so that the installed icu-config script can be found. (export PATH=/some/temp/folder/bin:$PATH)

Then go to the 'icu/source/test/hdrtst' directory and do 'make check'. This will attempt to compile against each header file individually on C and C++, to make sure there aren't any hidden include ordering problems. Output looks like this, the zero indicates a successful test. If a C++ file fails to compile as a C file, add it to the 'cxxfiles.txt' located in the hdrtst directory. Run this test with all the uconfig.h variations (see below).

unicode/bidi.h        - 0  (0 means no error)
unicode/chariter.h    - 0
unicode/convert.h     - 0

Test ICU completely, and run the header test (above) with:

1. none of the 'UCONFIG_NO_XXX' switches turned on (i.e., the normal case)
2. all of the 'UCONFIG_NO_XXX' switches turned on (set to 1)
3. For each switch, test once with just that one switch on.

Update urename.h to catch all possible ICU4C library exports (especially on non-Windows systems that tend to ignore export qualifiers). See icu/source/tools/genren/README.

Diff the new one with the previous one; there are typically a few bad #defines in there.

1. make doc
2. cd source/tools/gendraft ; make install-headers
3. Double check the modified files in <icu>/source/common/unicode folder and commit.

• Convert icu/source/test/testdata/ra.txt to XLIFF

genrb -s icu/source/test/testdata -d icu/source/test/testdata/ -x -l en ra.txt

• Verify that the ra.xlf produced is identical to the one in CVS HEAD (except for generation date)
• Convert icu/source/test/testdata/ra.xlf back to ICU format

java -cp icu4j/classes com.ibm.icu.dev.tool.localeconverter.XLIFF2ICUConverter
-d . -t ra ra.xlf

• Verify that the ra.txt produced is identical to the one in CVS HEAD (except for generation date)
• Go through the steps given in http://icu.sourceforge.net/docs/papers/localize_with_XLIFF_and_ICU.pdf

Build and run all of the sample and demo apps that are included with ICU, on each of the reference platforms. A list of them is in the readme. Also see the build system.

Another worthy test: Test suites and demos from the previous release should also compile and run with the libraries of the current release, at least when certain #defines are set (unless they test APIs that are deprecated and have been removed since)!

1. Build ICU4C
2. Temporarily modify the data makefile: s/usePoolBundle/writePoolBundle/
3. Rebuild the data
4. Copy $icu4c_out/data/out/build/icudt.../pool.res to $icu4c_root/source/data/locales/pool.res
5. Revert the data makefile change
6. Rebuild the data (make sure it actually gets rebuilt, e.g., by first deleting all of the data/out files)
7. Run ICU4C tests
8. Rebuild the ICU4J data .jar files

Test if the data portability (under common endianness & charset family) is ok. On the ICU build server, you would use the "Build from source .dat archive" option. When it's not available, you would do the following:

1. Build ICU4C on Win32.
2. Copy the icu/source/data/out/icudt<data_version>l.dat file into icu/source/data/in
3. Delete non-essential directories from icu/source/data.
4. Package up a clean copy of ICU for a non-Windows machine, like Linux on x86.
5. Build ICU on the non-Windows machine from the newly created package.
6. Run all tests on that non-Windows machine.

Run environmentTest.sh on a Linux machine This test verifies that the ICU test suite will work regardless of a user's default locale and timezone. This test should be run on a fast machine with several CPU cores. This will take a long time to run. Here are the steps to run the test.

1. cd tools/release/c/
2. ./environmentTest.sh
3. Wait a while for the tests to finish. The logs will be written in each test directory. e.g. icu/source/test/intltest/intltest-*.txt. A message will be printed from each spawned test script when it finishes.
4. grep the logs for any test failures when the tests are done.

Build the icuapps module following the README's. Update code and/or docs as needed. "Reference" platforms for icuapps are: RedHat Linux and win32. On Linux, icuapps is built against the "make install "'ed ICU. So, run ICU4C's configure with --prefix=/some/where pointing to where ICU4C should be installed, and also follow icuapps's README.

Install the new locale explorer and other demos/samples onto the public demo hosting site.

For ICU4C:

1. Update the API documentation in all header files (.h file) to have correct @draft/@stable/@deprecated labels.
2. Update docmain.h
3. ./configure
4. make docs
5. Folow instructions in tools/release/java/readme.txt.

For ICU4J:

Run the ICU4C API comparison tools (build.xml targets gatherapi, apireport) to generate new api and report files (build.xml strings will need to change). Check in the API file. Rename the report file to ${root}/APIChangeReport.html and replace the existing one and check it in. Make sure the readme.html file links to it (no change should be required).

If there are any last second changes. Make sure to document especially items that affect software that is currently compiled with the previous version of ICU. Update build/installation instructions as necessary.

Make sure that, aside from download pages, homepages, news items, feature lists and feature comparisons, etc. are updated. Upload the new API references. Update the User Guide.

Post the update to the icu-announce list, the icu-support list, the icu-design list.

Put a News item on the ICU project site. Both ICU4C and ICU4J are announced in the same announcement.

For freshmeat.net

1. http://www.freshmeat.net/
2. Login under the correct account.
3. I usually search for: icu unicode
4. Click on the project returned
5. Under the 'project' drop down, choose 'Add release'
6. We have (thankfully) only one branch of development- choose Default
7. Release info
   • release focus: Choose the one which best fits
   • Version
   • Changes: be VERY brief, and include links if needed.
     Please read the other ICU4C/ICU4J updates to get an idea for how we are wording these. DO NOT use bulleted lists. DO use short 'sentences'. Remember they can go to the release list and/or bug DB for the full scoop!
     Put the URL to the 'release notes' (i.e. http://oss.software.ibm.com/icu/download/(version)/ ) in the changes list.
   • tar/gz, zip: Put the FULL URLs to the actual files, here
   • Changelog: Leave BLANK. This is for a single page of ALL release changes (Actually: http://oss.software.ibm.com/icu/download/ fits this fairly well.)
   • License - leave at MIT/X
   • mailing list: I set this to the main mailing list page. (http://oss.software.ibm.com/icu/archives/)
   Click 'go on..'
8. PREVIEW.
   Very important- Check:
   • SPELLING
   • links!!!!! Make sure they go where we want
   • accuracy - version number is OK?
   Remember that this is an important PR face for ICU4C and ICU4J.
9. Submit!

That's it. Submissions take some time (<24 hrs) to go on. I guess they check the accuracy as well- not sure here. (Steven Loomis, 2001-06-14)

Other sites:

ICU sourceforge News - login here, go to News, Admin, and submit a news item as an administrator.

LanguageTech Net News (tm)

"LanguageTech Net News" features the latest news for language technology and multilingual computing. It is prepared from materials compiled and maintained by the research editors at "MultiLingual Computing & Technology" and the Language Technology Research Center.

Press releases and submissions are always welcomed and encouraged. Please send e-mail to news@multilingual.com, faxes to 208-263-6310 and postal mail to MultiLingual Computing, Inc., 319 North First Avenue, Sandpoint, ID 83864 USA, or call 208-263-8178.

To view archived issues of "LanguageTech Net News" visit http://www.multilingual.com/.

To subscribe to "LanguageTech Net News" go to http://www.multilingual.com/multilingualpress/ or send an e-mail to news-l-subscribe@multilingual.com.

Cleanup the milestone in the ICU Trac. Move left over items to future milestones. Close the milestone.

Look for TODO comments in the source code and file new tickets as required.