[boost] Boost.Test: Regressions on Unix

From: Jens Maurer (Jens.Maurer_at_[hidden])
Date: 2003-07-20 16:22:40

There's a mild compatibility problem with "errors_handling_test"
on Unix vs. Win32.

There's a reference pattern file checked in that was apparently
generated on Windows. Windows passes the test.

On Unix, there are 228 test failures because of differences
like these:

-Exception in "bad_function_param": memory access violation
+Exception in "bad_function_param": signal: memory access violation

-Exception in "bad_test::test": integer divide by zero
+Exception in "bad_test::test": signal: SIGFPE (arithmetic exception)

That means that, on Windows, a structured exceptions indicates
a null pointer dereference or somesuch, while on Unix, you get
a SIGBUS or SIGSEGV, which decodes to a slightly different string.
Similarly, on Linux at least, a SIGFPE indicates an integer
divide by zero.

I believe that at least Linux with gcc 3.3 is handling
"errors_handling_test" just fine and the failures are
random noise, i.e. named "implementation defined" or
"unspecified" behaviour in the C++ standard.  What do we
do to get rid of the ugly red "Fail" in the HTML test
summary?

Jens Maurer