| 摘要 |
One embodiment of the present invention provides a technique for automatic diagnosis of regression test failures. Initially, an automatic regression test system provides the test results per configuration for a revision of the device under test. Next, each failing test is analyzed in conjunction with the information in the version control system that is used to manage the device under test in order to conclude what additional revisions that needs to be tested for each test and configuration in order to find the earliest failing revision. Next, a request is issued to the automated regression test system which performs the requested tests and provides back a list of test results. Next, another analysis takes places and if the earliest failing revision cannot be concluded for each failing test and configuration then another request is issues to the automated regression test system. This continues until the earliest failing revision can be concluded for each failing test and configuration. The result presents each faulty revision and the group of tests and configurations that fail as a consequence. Another embodiment of the present invention provides a technique for automatically diagnosing the root cause for a test failure by comparing the difference in outcome between one revision of the device under test that fails with another revision of the same test for which the same test passes. This is done by instrumenting the computer program that constitutes the device under test to extract the variable states that affect the difference in outcome. Next, all parts of the computer program that are proven to have no affect on the difference in outcome is ignored. Finally, the root cause is concluded by identifying the sections of the computer program that are different between the revision for which the test fails and the revision for which test passes. The parts of these sections that have also been proven to be affecting the difference in outcome are concluded to be the root cause. |
