These changes will presumably break some working code in ways that are very hard to detect, which carries a cost. The second suggestion in particular seems completely arbitrary to me. What is the benefit of making these changes?

The benefit would be to get a consistent modification of data set by the user, to avoid unexpected results at working with data related to dates.

I have no intuition that "2010-01-00" should be "2010-01-01". In fact to me it seems intuitive that "2010-01-00" is "the before 2010-01-01", and so "2009-12-31" seems correct.

The issue is that it changes three data.

Instead of be handled with "before of", it could be handled thinking in both minimum and maximum values for a field:

• Whether the minimum of a field (year, month, day) is one, then whatever value lesser to 1 could be converted to 1.
• And for maximum values (9999 for year, 12 for month, 29/30/31 for day), then the values greater than those, should be changed to the maximum value.

I think that suggested change would be strictly worse than the current behavior, because it would make it much harder to write code like time.Date(t.Year(), t.Month(), t.Day() - 1, 0, 0, 0, 0, t.Location()) to get the day before.

In my opinion the current time.Date() function is useful and clearly documented. Its usefulness can be seen for instance in how time.AddDate() is implemented. Changing this will break existing code and reduce the usefulness, while the suggested prenormalization can easily be done in a wrapper function by the user if really needed.

The current behavior is documented and works and is not clearly wrong. Changing it will break existing programs for a very small benefit. We are not going to do this.