28.2. Test Evaluation

Some properly installed and fully functional PostgreSQL installations can "fail" some of these regression tests due to platform-specific artifacts such as varying floating-point representation and message wording. The tests are currently evaluated using a simple diff comparison against the outputs generated on a reference system, so the results are sensitive to small system differences. When a test is reported as "failed", always examine the differences between expected and actual results; you may well find that the differences are not significant. Nonetheless, we still strive to maintain accurate reference files across all supported platforms, so it can be expected that all tests pass.

The actual outputs of the regression tests are in files in the src/test/regress/results directory. The test script uses diff to compare each output file against the reference outputs stored in the src/test/regress/expected directory. Any differences are saved for your inspection in src/test/regress/regression.diffs. (Or you can run diff yourself, if you prefer.)

If for some reason a particular platform generates a "failure" for a given test, but inspection of the output convinces you that the result is valid, you can add a new comparison file to silence the failure report in future test runs. See Section 28.3 for details.

28.2.1. Error message differences

Some of the regression tests involve intentional invalid input values. Error messages can come from either the PostgreSQL code or from the host platform system routines. In the latter case, the messages may vary between platforms, but should reflect similar information. These differences in messages will result in a "failed" regression test that can be validated by inspection.

28.2.2. Locale differences

If you run the tests against an already-installed server that was initialized with a collation-order locale other than C, then there may be differences due to sort order and follow-up failures. The regression test suite is set up to handle this problem by providing alternative result files that together are known to handle a large number of locales.

28.2.3. Date and time differences

Most of the date and time results are dependent on the time zone environment. The reference files are generated for time zone PST8PDT (Berkeley, California), and there will be apparent failures if the tests are not run with that time zone setting. The regression test driver sets environment variable PGTZ to PST8PDT, which normally ensures proper results.

28.2.4. Floating-point differences

Some of the tests involve computing 64-bit floating-point numbers (double precision) from table columns. Differences in results involving mathematical functions of double precision columns have been observed. The float8 and geometry tests are particularly prone to small differences across platforms, or even with different compiler optimization options. Human eyeball comparison is needed to determine the real significance of these differences which are usually 10 places to the right of the decimal point.

Some systems display minus zero as -0, while others just show 0.

Some systems signal errors from pow() and exp() differently from the mechanism expected by the current PostgreSQL code.

28.2.5. Row ordering differences

You might see differences in which the same rows are output in a different order than what appears in the expected file. In most cases this is not, strictly speaking, a bug. Most of the regression test scripts are not so pedantic as to use an ORDER BY for every single SELECT, and so their result row orderings are not well-defined according to the letter of the SQL specification. In practice, since we are looking at the same queries being executed on the same data by the same software, we usually get the same result ordering on all platforms, and so the lack of ORDER BY isn't a problem. Some queries do exhibit cross-platform ordering differences, however. When testing against an already-installed server, ordering differences can also be caused by non-C locale settings or non-default parameter settings, such as custom values of work_mem or the planner cost parameters.

Therefore, if you see an ordering difference, it's not something to worry about, unless the query does have an ORDER BY that your result is violating. But please report it anyway, so that we can add an ORDER BY to that particular query and thereby eliminate the bogus "failure" in future releases.

You might wonder why we don't order all the regression test queries explicitly to get rid of this issue once and for all. The reason is that that would make the regression tests less useful, not more, since they'd tend to exercise query plan types that produce ordered results to the exclusion of those that don't.

28.2.6. Insufficient stack depth

If the errors test results in a server crash at the select infinite_recurse() command, it means that the platform's limit on process stack size is smaller than the max_stack_depth parameter indicates. This can be fixed by running the server under a higher stack size limit (4MB is recommended with the default value of max_stack_depth). If you are unable to do that, an alternative is to reduce the value of max_stack_depth.

28.2.7. The "random" test

The random test script is intended to produce random results. In rare cases, this causes the random regression test to fail. Typing

diff results/random.out expected/random.out

should produce only one or a few lines of differences. You need not worry unless the random test fails repeatedly.