A number with a thousands separator parses as a much smaller relative date. With two separators it comes back as the base time.
from datetime import datetime
import dateparser
base = datetime(2024, 6, 15, 12, 0, 0)
S = {"RELATIVE_BASE": base}
dateparser.parse("1000 days ago", settings=S) # 2021-09-19 correct
dateparser.parse("1,000 days ago", settings=S) # 2024-06-14 one day
dateparser.parse("12,345 days ago", settings=S) # 2024-06-03 12.345 days
dateparser.parse("1,000,000 days ago", settings=S) # 2024-06-15 12:00, the base, unchanged
The first three follow from float(num.replace(",", ".")) in get_kwargs, and #876 added that on purpose so "1,5 hours" works. That should stay. In a locale where the comma is a decimal mark those readings are right, and there is no locale information at that point to tell the two apart.
The last one is different. Translation rewrites the string first:
"1,000,000 days ago" -> "1 000,000 day ago"
PATTERN.findall -> [(" 000,000", "day")]
float("000.000") -> 0.0
The leading 1 is dropped, not misread. The \s* in ([+-]?\s*\d++[.,]?\d*+), which is there for "+ 5 days", lets the match start partway through the number. No reading of "1,000,000 days ago" gives the current time, so this one is wrong in every locale.
Worth deciding together, since a fix for the second probably settles the first. The options I can see are to use the detected locale for the separator, treat a separator followed by exactly three digits as grouping, or fail instead of returning a truncated number. Happy to send a patch once you say which way you want it.
A number with a thousands separator parses as a much smaller relative date. With two separators it comes back as the base time.
The first three follow from
float(num.replace(",", "."))inget_kwargs, and #876 added that on purpose so"1,5 hours"works. That should stay. In a locale where the comma is a decimal mark those readings are right, and there is no locale information at that point to tell the two apart.The last one is different. Translation rewrites the string first:
The leading
1is dropped, not misread. The\s*in([+-]?\s*\d++[.,]?\d*+), which is there for"+ 5 days", lets the match start partway through the number. No reading of"1,000,000 days ago"gives the current time, so this one is wrong in every locale.Worth deciding together, since a fix for the second probably settles the first. The options I can see are to use the detected locale for the separator, treat a separator followed by exactly three digits as grouping, or fail instead of returning a truncated number. Happy to send a patch once you say which way you want it.