[R] [C++] Should strptime support a partial format?

[asfimport]

In base R we can use strptime to parse and extract parts from a datetime by passing a restricted format.

> strptime("2012-05-25 18:45:32", format = "%Y-%m-%d")
[1] "2012-05-25 BST"
> strptime("2012-05-25 18:45:32", format = "%Y-%m-%d %H")
[1] "2012-05-25 18:00:00 BST

Reporter: Dragoș Moldovan-Grünfeld / @dragosmg
Watchers: Rok Mihevc / @rok

Related issues:

• [C++] Support parsing of "2020-01" into 2020-01-01 (is related to)
• [C++] Add error handling option to StrptimeOptions (depends upon)

_{Note: This issue was originally created as ARROW-15813. Please see the migration documentation for further details.}

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
There might be a way to implement this directly in R once strptime can return a NULL so we can coalesce through several formats. 

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
@jorisvandenbossche& @rok I'd be curious to know what you think. Would something like this be useful outside R? 

[asfimport]

Rok Mihevc / @rok:
Would you say this is basically a strptime + floor under the hood? If so I would implement it like that in R for now.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
Could be. I haven't thought about it yet, but strptime + floor sounds like a solid direction.

[asfimport]

Joris Van den Bossche / @jorisvandenbossche:
I don't know if strptime + floor is a possible way to implement this, because the strptime will always raise an error (or return null) if the format doesn't match exactly? So you don't get a value that you can floor.

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
I was thinking we could try several formats and then choose the one that parses successfully + then floor to the format that has been passed as an arg (i.e. for strptime(..., format = "%Y-%m-%d %H") coalesce through variations of "%H%M%S" with a variety of separators). 

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
But this approach might come with significant overhead. We would need to do a bit of string pre-processing (which on occasion might be not be needed - e.g. the user passes a format such as "%Y-%m-%d %H" and the data is in that format).

[asfimport]

Jonathan Keane / @jonkeane:
FWIW, in the R docs it describes the process as "Each input string is processed as far as necessary for the format specified: any trailing characters are ignored." which is slightly different from a floor-like action (though agree with Joris: that might not even be possible )

[asfimport]

Rok Mihevc / @rok:
So substring + strptime with appropriate length format string?

[asfimport]

Jonathan Keane / @jonkeane:
Kind of, more like the format string has a .* at the end that eats up and ignores anything that isn't in the format itself

[asfimport]

Rok Mihevc / @rok:
I suppose that can be done from R side now? Do we want this in c++?

[asfimport]

Jonathan Keane / @jonkeane:
I'm skeptical it can be done in R, and almost certainly not in a performant manner (unless format already accepts regex like that?)

[asfimport]

Rok Mihevc / @rok:
Why not just use substr kernel? If we can do strict in strptime I don't see why we need regex.

[asfimport]

Jonathan Keane / @jonkeane:
Because we don't actually know the length of the first bit (or really any other part, though ) of the string:

> times <- c("2012-05-25 18:45:32", "2020-5-6", "20-05-06", "2022-06-07 00:55", "2020-01-01 could be anything at all ROFL")
> strptime(times, format = "%Y-%m-%d")
[1] "2012-05-25 CDT" "2020-05-06 CDT" "0020-05-06 LMT" "2022-06-07 CDT"
[5] "2020-01-01 CST"

Or with %B it's even worse:

> months_plus <- c("2020 January 1 string", "2020 July 1 extra")
> strptime(months_plus, format = "%Y %B %d") 
[1] "2020-01-01 CST" "2020-07-01 CDT"

[asfimport]

Rok Mihevc / @rok:
Ah, I was assuming we're only doing ISO8610 format. Yeah this wouldn't be easy. Thanks for the explanation!