This repository provides tools and procedures for verifying that an implementation is working correctly according to the data-based specifications. The tests are implemented on several platforms including NodeJS (JavaScript), ICU4X (RUST), ICU4C, etc. Additional programming platforms may be added to use the test driver framework.
The goal of this work is an easy-to-use framework for verifying that an implementation of ICU functions agrees with the required behavior. When a DDT tet passes, it a strong indication that output is consistent across platforms.
Data Driven Test (DDT) focuses on functions that accept data input such as numbers, date/time data, and other basic information. The specifications indicate the expected output from implementations when given the data and argument settings for each of the many individual data items.
Note that these tests are only part of the testing required for ICU-compliant libraries. Many additional tests are implemented in the
- !!! TODO: reference to data specifications
Each ICU test program is built with a specific version of ICU & CLDR data. These versions are updated periodically. For each ICU version, there is a specific CLDR version, e.g., ICU73 uses data from CLDR 43, although multiple ICU releases may depend on the same CLDR data.
For this reason, specifying a particular ICU version for test data or test executor or both
Each part of Data Driven Testing is designed to handle a specific ICU version.
-
Data generation uses specifications starting with ICU versions 70, 71, etc. For each ICU release, these data should be updated.
-
Test execution allows setting the data version explicitly with a command line argument --icuversion that points to the indicated test data. The ICU version of the test executor platform is requested from each platform at the start of the test driver. Output directories are created under the platform for the test results running a particular ICU version, e.g., testOutput/node/icu73.
-
Test verification uses ICU version information in the test output files for matching with the corresponding expected results. Verification output appears in the testResults subdirectory for each node, e.g. testOutput/rust/icu71.
Conceptually, there are three main functional units of the DDT implementation:
Utilizes Unicode (UTS-35) specifications, CLDR data, and existing ICU test data. Existing ICU test data has the advantage of being already structured towards data driven testing, is in many cases formatted in a way to simplify adding new tests, and contains edge and error cases.
Data generation creates two files:
- Test data instance: a JSON file containing the type of test and additional information on the environment and version of data.
The test type is indicated with the "Test scenario" field.
Individual data tests are stored as an array of items, each with a label and parameters to be set for computing a result.
Example line for collation_short:
{
"description": "UCA conformance test. Compare the first data\n string with the second and with strength = identical level\n (using S3.10). If the second string is greater than the first\n string, then stop with an error.",
"Test scenario": "collation_short",
"tests": [
{
"label": "0000000",
"string1": "\u0009!",
"string2": "\u0009?"
},
-
A required test result file (JSON) containing the expected results from each of the inputs. This could be called the “golden data”.
Sample verify data:
{"Test scenario": "collation_short", "verifications": [ { "label": "0000000", "verify": "True" },
Test execution consists of a Test Driver script and implementation-specific executables. The test driver executes each of the configured test implementation executables, specifying the input test data and the location for storing results. STDIN and STDOUT are the defaults.
Each test executor platform contains a main routine that accepts a test request from the test driver, calling the tests based on the request data.
Each executor parses the data line sent by the test driver, extracting elements to set up the function call the the particular test.
For each test, the needed functions and other objects are created and the test is executed. Results are saved to a JSON output file.
See executors/README for more details
Each test is matched with the corresponding data from the required test results. A report of the test results is generated. Several kinds of status values are possible for each test item:
- Success: the actual result agrees with expected results
- Failure: a result is generated, but the result is not the same as the expected value.
- No test run: The test was not executed by the test implementation for the data item
- Error: the test resulted in an exception or other behavior not anticipated for the test case
-
What should be done if the test driver fails to complete? How can this be determined?
- Proposal: each test execution shall output a completion message, indicating that the test driver finished its execution normally, i.e., did not crash.
In its first implementation, Data Driven Test uses data files formatted with JSON structures describing tests and parameters. The data directory string is set up as follows:
- Test data files for each type of test, e.g., collation, numberformat, displaynames, etc. Each file contains tests with a label, input, and parameters.
- Verify files for each test type. Each contains a list of test labels and expected results from the corresponding tests.
This contains a subdirectory for each executor. The output file from each test is stored in the appropriate subdirectory. Each test result contains the label of the test and the result of the test. This may be a boolean or a formatted string.
The results file contains information identifying the test environment as well
as the result from each test. As an example, collation test results from the
testOutput/node file are shown here:
{
"platform": {
"platform": "NodeJS",
"platformVersion": "v18.7.0",
"icuVersion": "71.1"
},
"test_environment": {
"test_language": "nodejs",
"executor": "/usr/bin/nodejs ../executors/nodejs/executor.js",
"test_type": "collation_short",
"datetime": "10/07/2022, 16:19:00",
"timestamp": "1665184740.2130146",
"inputfile": "/usr/local/google/home/ccornelius/DDT_DATA/testData/icu73/collation_testt.json",
"resultfile": "/usr/local/google/home/ccornelius/DDT_DATA/testOutputs/node/icu73/collation_test.json",
"icu_version": "ICUVersion.ICU71",
"cldr_version": "CLDRVersion.CLDR41",
"test_count": "192707"
},
"tests": [
{
"label": "0000000",
"result": "True"
},
{
"label": "0000001",
"result": "True"
},
...
]
}
This directory stores summary results from verifying the tests performed by each executor. Included in the testReports directory are:
-
index.html: shows all tests run and verified for all executors and versions. Requires a webserver to display this properly. -
exec_summary.json: contains summarized results for each pair (executor, icu version) in a graphical form. Contains links to details for each test pair. -
subdirectory for each executor, each containing verification of the tested icu versions, e.g.,
node/,rust/, etc.
Under each executor, one or more ICU version files are created, each containing:
-
verfier_test_report.html- for showing results to a user via a web server -
verfier_test_report.json- containing verifier output for programmatic use -
failing_tests.json- a list of all failing tests with input values -
pass.json- list of test cases that match their expected results -
test_errors.json- list of test cases where the executor reported an error -
unsupported.json- list of test cases that are not expected to be supported in this version
The verifier_test_report.json file contains information on tests run and comparison with the expected results. At a minimum, each report contains:
- The executor and test type
- Date and time of the test
- Execution information, from the testResults directory
- Total number of tests executed
- Total number of tests failing
- Total number of tests succeeding
- Number of exceptions identified in the test execution. This may include information on tests that could not be executed, along with the reasons for the problems.
- Analysis of test failures, if available. This may include summaries of string differences such as missing or extra characters or substitutions found in output data.
Requirements to run Data Driven Testing code locally:
- Install the Python package
jsonschema- In a standard Python environment, you can run
pip install jsonschema - Some operating systems (ex: Debian) might prefer that you install
the OS package that encapsulates the Python package
sudo apt-get install python-jsonschema
- In a standard Python environment, you can run
- Install the minimum version supported by ICU4X
- The latest minimum supported supported Rust version ("MSRV") can be found in the
rust-toolchain.tomlfile - To view your current default Rust version (and other locally installed Rust versions):
rustup show - To update to the latest Rust version:
rustup update
- Install
logrotatesudo apt-get install logrotate
- The latest minimum supported supported Rust version ("MSRV") can be found in the
Data Driven Test was initiated in 2022 at Google. The first release of the package was delivered in October, 2022.
Copyright © 2022-2024 Unicode, Inc. Unicode and the Unicode Logo are registered trademarks of Unicode, Inc. in the United States and other countries.
The project is released under LICENSE.
A CLA is required to contribute to this project - please refer to the CONTRIBUTING.md file (or start a Pull Request) for more information.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent