From: Beman Dawes (bdawes_at_[hidden])
Date: 2002-07-15 15:54:32
In trying to improve regression test status reporting, I'm finding the need
to test the reporting process itself more explicitly.
We currently have these test types:
(and may need a "lib-build", too, but that's in the future a bit.)
A test program (.cpp file) might exhibit these characteristics.
compile, link, and run all report success, no warnings.
compile and link report success, no warnings, run reports failure.
compile reports success, no warnings, but link reports failure.
compile reports failure.
compile, link, and run all report success, but compile has warnings.
compile and link report success, run reports failure, but compile
compile reports success, link reports failure, compile has warnings.
So we've got something like 6 * 7 = 42 tests we'd like to be sure work
correctly. That number grows a bit if library build tests and/or failures
are considered separate cases.
My initial reaction is to create a directory with seven little programs to
be used as test cases, and its own Jamfile to test all the 42 combinations?
Is this a good idea?
Does anyone already have such a set of programs? Are they in the CVS tree
Where should such test programs go in the Boost directory structure?
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk