Howard Chu wrote:

markus.widmer@daasi.de wrote:

...

We could reproduce this with 2.4.42, 2.4.44 and 2.4.48. We hope you can reproduce this as well to see what is happening here.

Yes, can reproduce this. The function that converts a component-wise time into a timet is referencing time since 1970. The date in 1956 would yield a negative timet value but the fields are unsigned ints, so instead it's treated as 17,000 years in the future. We can probably change this to handle signed timestamps but need to consider this further.

I believe the best way forward would be to allow signed values, and also to switch our epoch reference from 1970-01-01 to 0000-01-01 (i.e., use Gregorian Proleptic calendar). Year zero would be 1 BCE in this calendar, and anything earlier would be a negative year.

Changing these functions will require regenerating any indices. Looks like something we'll rewrite for 2.5 but leave 2.4 alone.