Travis Vitek wrote:
[...]
+ TEST_N (data.bool_, bool, data.value_ != 0);
+ TEST_N (data.long_, long, data.value_);
+ TEST_N (data.ulong_, unsigned long, data.value_);
I note you've changed the test from invoking one num_put member
per iteration to invoking all members each iteration. I'm curious
about your rationale for the change? (I don't necessarily disagree
with the approach, just wondering what if anything led you to make
the change.) FWIW, the one advantage to sticking with the original
testing strategy is that it would let us eliminate the call to
pubsetp() and (possibly) do away with the macros.
I'm missing something here. I don't see how we can eliminate the
pubsetp() calls. If we did, then the next insertion operation will
insert after the previous one. Instead of writing to the front of the
buffer every time, we'd be appending to it. At some point an insertion
will fail because the put pointer is at the end. Am I wrong?
You're right. I meant that if called put() only once each iteration
like we did before we would only need to call pubsetp() once as well.
Like so:
for (opt_nloops) {
buf.pubsetp (...);
switch (overload_to_call) {
case bool_overload: np.put ((bool)value); break;
case int_overload: np.put ((int)value); break;
case dbl_overload: np.put ((double)value); break;
case ptr_overload: np.put ((void*)value); break;
}
RW_ASSERT (postconditions);
}
Martin
