On Tue, 09 Nov 2010 18:48:30 -0500, bearophile <[email protected]>
wrote:
Yao G.:
Ugh. The datetime.d file is 1.5 MB? 0_o
So it will be very well tested :) Look, the huge benefit of unit tests
are that they are inline with the code. Either you want that or you want
small files, I don't understand the expectation that the file should be
small and compact as long as the implementation is...
I 100% agree with Jonathan, just ignore the unit tests, and look at the
API. (BTW, this is where a nice doc generator like the one Tango uses
would come in handy -- it shows the implementation if you click on the
function name).
My suggestions:
- Keep only 5-10% of the tests inside the datetime.d module, and move
the other 90-95% in a separated datetime test module.
- Use deterministic (repeatable) random data in a loop to perform some
compact fuzzy testing
- Use alias or string mixing to greatly reduce the amount of text
contained in the tests. Tuples help a lot. All those lines may be
replaced with a single loop that contains something like:
assertEqual(SysTime(DateTime(z0, x1, x2, x3, x4, x5),
FracSec.from!"msecs"(x6)).dayOfGregorianCal, x7)
Where the xn data comes from an array of typecons.Tuples like:
Data5(2010, 11, 1, 23, 59, 59, 999, 734_077)
Well, I'm not certain that this is the reason, but calendars are riddled
with corner cases. I would expect a well-tested date/time library to test
all those corner cases.
I know when writing the Date/Time code for Tango, there were quite a few
odd cases to test.
As long as the corner cases are tested, I think the rest of the fluff
could possibly be reduced. However, on the other hand, unit tests cost
nothing except to make the low-file-size purists cringe :) I say leave
them all in, there are no requirements on the size and style of unit
tests, just that they fully test the library.
-Steve