Error while parsing years earlier than 4799BC

Hello! I’m trying to parse a bunch of different dates, and BC ones are among them. And what I’ve noticed, is that if the date is earlier than 4799BC, astropy refuses to parse it and throws this:

Traceback (most recent call last):
  File "main.py", line 4, in <module>
    print(Time('-04800-01-02T03:04:05', format='fits').unix)
  File "/home/runner/.local/share/virtualenvs/python3/lib/python3.7/site-packages/astropy/time/core.py", line 399, in __init__
    precision, in_subfmt, out_subfmt)
  File "/home/runner/.local/share/virtualenvs/python3/lib/python3.7/site-packages/astropy/time/core.py", line 454, in _init_from_vals
    precision, in_subfmt, out_subfmt)
  File "/home/runner/.local/share/virtualenvs/python3/lib/python3.7/site-packages/astropy/time/core.py", line 505, in _get_time_fmt
    raise ValueError('Input values did not match {0}'.format(err_msg))
ValueError: Input values did not match the format class fits

Here is the code and corresponding repl.it

from astropy.time import Time

print(Time('-04799-01-02T03:04:05', format='fits').unix)
print(Time('-04800-01-02T03:04:05', format='fits').unix)

[mhvk]

By precise, I meant that the actual day/month and likely time were known… Easiest may be for early ~year accurate times to use byear or so. Actually, one probably has to check what people actually mean; it may well be the unit of year, which by definition is 365.25 days (i.e., julian year).

My overall feeling is that is it best to special-case those early dates and make an approximation that is clearly described.

[dhomeier]

The error message is a bit trickier to get right, since without an explicit format, the initializer simply tries all unique formats. But at least when one does give an explicit format, the message perhaps could be clearer.

When trying all the formats, the fits case still has to fail in formats.py, probably with a astropy.utils.exceptions.ErfaError: ERFA function "d2dtf" yielded 1 of "unacceptable date (Note 6)" I could not figure out yet exactly where that exception is caught, but if it could instead be passed back up to _get_time_fmt, the caller might check for that specific error. Or at least let an ErfaWarning pass though (although those are probably rather cryptic to the user, too).