Failing date_time "default classic" tests

Steven Solie wrote:

I'm getting the following output from one of the date_time tests with Boost 1.33.0 on AmigaOS:

execute-test ../../../bin/boost/libs/date_time/test/testdate_facet_new.test/gcc/debug/testdate_facet_new.run ====== BEGIN OUTPUT ====== Pass :: to_string & default formats match Pass :: to_string & default formats match FAIL :: default classic date FAIL :: default classic date period Pass :: custom date facet date period ....

I've traced it down to the %x date format specifier using the default locale while the tests themselves expect something else.

The test looks like this: date_facet* datefacet = new date_facet(); datefacet->format(date_facet::standard_format_specifier); teststreaming("default classic date", d, std::string("10/13/04"), std::locale(std::locale::classic(), datefacet)); AFAICT, this should produce a date formatted for the classic locale, i.e. the C locale. But isn't the output of the C local itself implementation defined?

According to the ANSI K&R C book this is correct behaviour because C++ classic locale uses the same format specifiers as the ANSI C strftime() function.

The C++ classic locale's result is implementation defined, see 22.2.5.3.2 of the C++ std. OTOH, the std says that "Implementations are encouraged to refer to other standards (such as POSIX) for these definitions.", so it might be a reasonable assumption to expect the POSIX output format for this.

So am I correct in assuming these two test cases are broken?

If the result of formatting with the classic locale really is implementation defined, then yes, you are right. I'm sure Jeff can shed some more light on this. Markus