Integration Tests


Run ICU4J Locale Service Provider tests

JDK 6 introduced the locale service provider interface. ICU4J release include a special jar file (icu4j-localespi.jar) implementing the locale service provider interface.

The test case for the ICU4J locale service provider requires JRE 6 or later version. Otherwise, there is no extra settings except for the standard ant set up ( necessary. To run the test cases, use ant with the top-level build.xml with target "localespiCheck"

$ ant localespiCheck

You should get the output like below -

...
build:

_runLocalespiCheck:
     [java] TestAll {
     [java]   BreakIteratorTest {
     [java]     TestGetInstance (7.581s) Passed
     [java]     TestICUEquivalent (0.832s) Passed
     [java]   } (8.669s) Passed
     [java]   CollatorTest {
     [java]     TestGetInstance (0.900s) Passed
     [java]     TestICUEquivalent (0.040s) Passed
     [java]   } (0.966s) Passed
     [java]   CurrencyNameTest {
     [java]     TestCurrencySymbols (2.682s) Passed
     [java]   } (2.683s) Passed
     [java]   DateFormatSymbolsTest {
     [java]     TestGetInstance (0.348s) Passed
     [java]     TestICUEquivalent (0.207s) Passed
     [java]     TestNynorsk (0.000s) Passed
     [java]     TestSetSymbols (0.071s) Passed
     [java]   } (0.631s) Passed
     [java]   DateFormatTest {
     [java]     TestGetInstance (0.456s) Passed
     [java]     TestICUEquivalent (0.113s) Passed
     [java]     TestThaiDigit (0.000s) Passed
     [java]   } (0.571s) Passed
     [java]   DecimalFormatSymbolsTest {
     [java]     TestGetInstance (0.098s) Passed
     [java]     TestICUEquivalent (0.000s) Passed
     [java]     TestSetSymbols (0.000s) Passed
     [java]   } (0.099s) Passed
     [java]   LocaleNameTest {
     [java]     TestCountryNames (2.500s) Passed
     [java]     TestLanguageNames (14.262s) Passed
     [java]     TestVariantNames (8.638s) Passed
     [java]   } (25.402s) Passed
     [java]   NumberFormatTest {
     [java]     TestGetInstance (0.266s) Passed
     [java]     TestICUEquivalent (0.209s) Passed
     [java]   } (0.475s) Passed
     [java]   TimeZoneNameTest {
     [java]     TestTimeZoneNames (17.766s) Passed
     [java]   } (17.766s) Passed
     [java] } (57.268s) Passed
     [java]
     [java] Test cases taking excessive time (>10s):
     [java] TestAll/LocaleNameTest/TestLanguageNames (14.262s)
     [java] TestAll/TimeZoneNameTest/TestTimeZoneNames (17.766s)
     [java]

BUILD SUCCESSFUL
Total time: 1 minute 46 seconds

Run ICU4J Test cases with JDK TimeZone

ICU4J provides an option to alter TimeZone implementation to use the JRE's own implementation. The test cases used for this is the regular ICU4J unit test suite, except forcing ICU4J to use JRE TimeZone by special system property - com.ibm.icu.util.TimeZone.DefaultTimeZoneType=JDK.

To run the test, you have to apply the latest time zone patch from a JRE vendor, because some test cases are sensitive to actual time zone transitions. For Oracle JRE, you should go to the J2SE download page http://www.oracle.com/technetwork/java/javase/downloads/index.html and download the latest JDK DST Timezone Update Tool and apply the patch to your local JRE.

To run the test case, you just need to invoke the ant target "jdktzCheck".

$ ant jdktzCheck

Note: You might not be able to get the update tool matching the tzdata version used by ICU. In this case, some test cases may reports failures. Unfortunately, you have to walk though the failures to see if they are expected or not manually in this case.

Verify the Eclipse ICU4J plug-in

  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.

Run Tests Without ICU Data

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.

To do this, build and test normally, then replace the ICU data shared library with the stubdata library and run the tests again with the -w option.

On Linux (adjust the version number, 60.1 in this example, as required)

cd icu4c/source
cp stubdata/libicudata.so.60.1 lib/
cd test/intltest
INTLTEST_OPTS=-w make check
cd ../cintltst
CINTLTST_OPTS=-w make check

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.

Verify that ICU4C tests pass without collation rule strings

ICU4C 53 and later

Background: http://bugs.icu-project.org/trac/ticket/10636

Rebuild ICU data without collation rule strings. This should work: make GENRBOPTS='-k --omitCollationRules'  (-k is --strict)

For ICU 54..63, I went into the build output folder and did:

cd data
ICUDT=icudt63l
rm out/build/$ICUDT/coll/*.res
make GENRBOPTS='-k --omitCollationRules'

If this does not work, then add this option to the configure'd data/Makefile, see ticket #10636.

Run the tests with data-errors-as-warnings. INTLTEST_OPTS=-w CINTLTST_OPTS=-w make -j5 check

See that they pass, or fix them to pass. See ticket #10636 test code changes for examples.

Test ICU4J with only little-endian ICU4C data

Only available since ICU 54.

With ICU 55 and later:

  • Rebuild ICU4C with make INCLUDE_UNI_CORE_DATA=1 check or similar, and provide a path only for the .dat file.
    • Should be little-endian for coverage
    • Find the .data file in the build output area, e.g., data/out/tmp/icudt59l.dat
    • Create a temporary directory such as mkdir -p /tmp/icu/build/data/out/tmp
    • Copy the .dat file to the new directory.
  • Build and test ICU4J without its own data: ant clean && ant -Dicu4c.data.path=/tmp/icu/build/data/out/tmp check
    • The configuration option sets the ICUConfig.properties data path
    • Verify that all tests pass.

ICU 54 method:

In icu4j-core/src/com/ibm/icu/ICUConfig.properties set com.ibm.icu.impl.ICUBinary.dataPath to a list of paths with all of the ICU4C data (should be little-endian for coverage), with a path to where the .dat file is and a path to the source/data/in files for data that is hardcoded in ICU4C and therefore not in the .dat file (e.g., uprops.icu).

Change com.ibm.icu.impl.ICUData.logBinaryDataFromInputStream to true, maybe set a breakpoint where such a message is logged.

Run all of the ICU4J tests, maybe in the debugger for the breakpoint, or look for logger output to the console.

Revert your config changes.

Build and run testmap

Build and run the source/test/testmap project. (There is currently no Windows project defined for it.)

$ cd <root of your ICU build tree>
$ CONFIG_FILES=test/testmap/Makefile CONFIG_HEADERS= ./config.status
$ cd test/testmap
$ make check

Verify XLIFF conversion

Note: The following instruction does not work.  Please read the comments with orange background. There are some issues in the current ICU XLIFF tools and the test case below.  See the comments in ticket#6383.

Instructions for verifying the XLIFF conversion tools.

  • 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

-d icu/source/test/testdata/ overwrite the existing ra.xlf. Specify another directory.

ra.txt has the top level item "ra", which is supposed to be the content language. Thus, with -l en, you'll get a warning -
"The top level tag in the resource and language specified are not the same. Please check the input."
We should use "-l ra" here.
  • Verify that the ra.xlf produced is identical to the one in CVS HEAD (except for generation date)

    If you use "-l ra" above, you'll get <file .... source-language = "ra" .... />, which is different from ra.xlf in the repository.
    Also, new line codes is broken for imported contents.

  • 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

    The option "-t ra" does not work, because ra.xlf does not contain target language data. Use "-c ra" instead.
  • Verify that the ra.txt produced is identical to the one in CVS HEAD (except for generation date)

    You cannot expect the generated ra.txt exactly matches the original one because of table item re-ordering, new line code changes,
    and explicit resource types (e.g. "ra {" vs. "ra:table {").


  • Go through the steps given in http://icu.sourceforge.net/docs/papers/localize_with_XLIFF_and_ICU.pdf

Test sample and demo programs

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)!

Test data portability

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 the environment tests

Run environmentTest.sh on a Linux machine which has many (all possible?) POSIX locales installed. 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 icu4c/source/
  2. ./runConfigureICU Linux
  3. make check
  4. cd ../../tools/release/c/
  5. ./environmentTest.sh
  6. Wait a while for the tests to finish. The logs will be written in each test directory. e.g. icu4c/source/test/intltest/intltest-*.txt. A message will be printed from each spawned test script when it finishes.
  7. grep the logs for any test failures when the tests are done.


Run Thread Sanitizer tests

Thread sanitizer testing is one of the standard Travis builds. If it is passing there, nothing further is required.
To run manually, on a Linux system with clang,

cd icu4c/source
CPPFLAGS=-fsanitize=thread LDFLAGS=-fsanitize=thread ./runConfigureICU --enable-debug --disable-release Linux
make clean
make -j6 check

Errors are displayed at the point they occur, and stop further testing.


Run Address Sanitizer tests

Address sanitizer testing is included in the standard Linux with Clang Travis builds. If it is passing there, nothing further is required.
To run manually, on a Linux system with clang,

cd icu4c/source
CPPFLAGS=-fsanitize=address LDFLAGS=-fsanitize=address ./runConfigureICU --enable-debug --disable-release Linux
make clean
make -j6 check

Memory leaks are summarized at the end. Other errors are displayed at the point they occur, and stop further testing. 


ICU4J Serialization Compatibility Test Data


The regular ICU4J unit test includes serialization compatibility tests. The test case creates an ICU service object from serialized data created by a former version of ICU. When we prepare a new release, the serialization compatibility test data should be created and checked in for future testing. This task is usually done just before publishing release candidate.

  1. Run regular ICU4J unit tests -  ant check
  2. Make sure the unit tests pass successfully.
  3. Run - ant serialTestData
  4. Copy a folder with ICU version (e.g. ICU_61.1) generated under <icu4j root>/out/serialTestData to <icu4j root>/main/tests/core/src/com/ibm/icu/dev/test/serializable/data.
  5. You may delete older serialization test data from the directory (but keep the oldest one - ICU_3.6).
  6. Run  ant check  again before committing the changes.

Comments