Boost Test Policies and Protocols
The Boost libraries are intended to be both reliable and portable.
Every experienced programmer knows that means each library must be tested against a suitable number of test cases, on a wide range of platforms,
and then tested again (regression tested) every time a change is made and before
every release.
"Quality assurance based on a wide range of targeted tests" as one
of the key answers to C.A.R
Hoare's question "How
did software get so reliable without proof."
Regression test
Boost uses an automatic regression test suite which generates HTML compiler
status tables.
Test Policy
Required
- Every Boost library should supply one or more suitable test programs to be
exercised by the Boost regression test suite. In addition to
the usual compile-link-run tests expecting successful completion,
compile-only or compile-and-link-only tests may be performed, and success
for the test may be defined as failure of the steps.
- Test program execution must report errors by returning a non-zero value. They
may also write to stdout or stderr, but that output should be relatively
brief. Regardless of other output, a non-zero return value is the only
way the regression test framework will recognize an error has
occurred. Note that test programs to be included in the status tables must
compile, link, and run quickly since the tests are executed many, many,
times.
- Libraries with time consuming tests should be divided into a
fast-execution basic test program for the status tables, and a separate
full-coverage test program for exhaustive test cases. The basic test
should concentrate on compilation issues so that the status tables
accurately reflect the library's likelihood of correct compilation on a
platform.
- If for any reason the usual test policies do not apply to a particular
library, an alternate test strategy must be implemented.
Optional
- Use the Boost Test Library for the construction of simple tests.
- [Planned] Use the Boost Unit Test Library for the construction of more complex tests
and test suites.
- The Boost Program Execution Library is also available for very simple
tests, but the Boost Test Library is much preferred even for very simple
tests.
Suggested Protocol for Fixing Bugs or Adding Features.
- First, add regression test cases that detects the bug or tests the
feature. Sometimes adding one case suggests similar untested cases, and they
are added too.
- Second, for bugs, run the regression test and verify that the bug is now
detected.
- Third, then, and only then, fix the bug or add the feature.
- Finally, rerun the full regression tests - sometimes the change breaks
something else.
History
The current regression test program was developed by Jens Maurer in December,
2000. The original regression test concept and program, including the automatic
generation of HTML compiler status tables, was developed by Beman Dawes in June,
2000.
Acknowledgements
Written by Beman Dawes. Jens Maurer, Paul Moore, Gary Powell and Jeremy Siek contributed helpful suggestions.
Revised 17 August, 2001
