Travis Vitek wrote:
Martin Sebor wrote:
Travis, this test looks too good -- there's nothing for me to pick
on, not even formatting! ;-)
I'm _finally_ getting used to it.
Glad to hear that! :)
I'm running it on a 4 CPU dual core Xeon against a locally modified
trunk and so far so good. Has it ever failed in your tests?
I have yet to see a failure in the num_get.mt test. I've only tested on
win32/vc8 and linux/gcc though. That brings up an important question
that isn't covered on the patch submission rules page. Should I be
testing on all platforms available, or should I just do enough testing
to convince myself that the test is portable and is working correctly?
Verifying new tests on a couple of platforms with different
compilers should be good enough. The bigger the differences
between the platforms and compilers the better (e.g., little
endian vs big endian, 32 vs 64 bit, etc.) Not-trivial library
changes and extensive changes to existing tests or examples
should be tested on more than two platforms. Small changes
can be tested on just one.
IMO, it's a judgment call that will depend on the nature of
the changes and how comfortable you are with the area you're
working on.
Martin
