Parsing empty string to date

[adkabo]

I expected an exception but got a Date:

julia> parse(Date, "")
0001-01-01

[elextr]

Explicitly documented to behave this way in https://docs.julialang.org/en/v1/stdlib/Dates/

[adkabo]

It looks like the reason this happens is to allow parsing incomplete date strings.

julia> parse(Date, "2000-03")
2000-03-01

In the case where an explicit date format is provided, it's rather unexpected.

julia> parse(Date, "", dateformat"yyyy-mm-dd")
0001-01-01

I often encounter missing data in the form of an empty string, and having it parse to 0001-01-01 instead of failing is very error-prone for my use cases.

Two ideas:

• Change parse(Date, s, fmt) so it throws an error if s doesn't match fmt exactly. This would break backward compatibility. If desired, introduce a PartialDateFormat that supports partial matches.
• Retain the current behavior and introduce a CompleteDateFormat created with completedateformat"yyyy-mm-dd" that would throw an exception here.

[elextr]

Perhaps you need tryparse

[nalimilan]

I agree this is quite dangerous. At least it would make sense to reject empty strings, as then you don't extract any information from the string so it's absolutely useless. But changing it would be breaking so it cannot happen before 2.0.

Rather than introducing a CompleteDateFormat type, maybe we could use non-standard string literal flags like in regexes? For example, dateformat"yyyy-mm-dd"c would require a complete string but dateformat"yyyy-mm-dd"p would allow partial matching?

Another option would be to add an argument to parse.

[anowacki]

See also #28090.

Note that in that discussion I proposed that if any DateFormat is passed to parse or tryparse, the empty string should not parse correctly. I think the non-standard string literal flag idea is okay, but I prefer adding an argument to parse, or perhaps a keyword argument (strict=false as a default?) could be used.

[PallHaraldsson]

Can this be considered a bug and changed in 1.x?

Despite "Explicitly documented", does anyone rely on (the very unhelpful, and illegal date, since in docs: "Both Date and DateTime are based on the ISO 8601 standard", the docs can be changed, and non-default option to allow old way):

julia> parse(Date, "")
0001-01-01

https://en.wikipedia.org/wiki/Wikipedia:Manual_of_Style/Dates_and_numbers#Dates,_months,_and_years

Use yyyy-mm-dd format only with Gregorian dates from 1583 onward
All-numeric yyyy-mm-dd dates might be assumed to follow the ISO 8601 standard, which mandates the Gregorian calendar.

[anowacki]

I certainly would prefer "" to throw an error, even though it was me that wrote the documentation [Edit: about the parsing of empty strings]. I can't think of any circumstances in which "" should correctly parse in reality, and I think the current behaviour is really just a side-effect of the way in which the parsing is implemented, rather than being intended.

That said, I think Julia should be strict about honouring SemVer so I'm uneasy about calling this a bug and changing the default behaviour. Perhaps running PkgEval on a PR for this, and seeing if anything has tests which check for this behaviour, would help? Given this is a parsing issue, I suspect tests will not necessarily catch all the cases where this crops up, even though I also suspect most things which 'work' with the current 'feature' are actually themselves bugs!

[oscardssmith]

I've added a triage tag. To me it seem reasonable that we should match incomplete non-empty dates, but throw on empty dates. I fairly strongly doubt that anyone actually wanted the current behavior.

[anowacki]

I have submitted #47117, which makes parsing empty strings throw an error. I thought it may be useful to help decide whether to go ahead with the change in a minor version bump.

[JeffBezanson]

Triage agrees empty string should be an error. Dropping the partial format parsing might also be good, but that's a separate issue.