Idea on datetimes/timezones

[ijlyttle]

Idea on datetimes/timezones

[krlmlr]

Happy to review a PR, but for now I'll focus on plugging the existing code into tibble.

[hadley]

Yeah exactly - and a separate function would be fine.

[krlmlr]

No worries, that's fine.

[hadley]

I don't know of a good (even approximate) way of generating the UTC offset. (Also need to remember that some time zones have partial hour offsets)

[krlmlr]

RcppCCTZ maybe?

[ijlyttle]

Assuming RcppCCTZ meets the needs, I think the partial-hour problem could be overcome by using something like:

<dttm-03:30/-02:30>
2015-06-01 01:00:00

which would just fit.

I think the "range" of the offsets offered could be limited to the date-times being considered in the given column.

On the one hand, this would limit the printing of dual-offset headers to those cases where both offsets appear in the tibble. On the other hand, that would seem to make it harder to format consistently for normal and daylight-saving offsets.

[hadley]

And over long periods of time, I bet only two offsets is insufficient

[ijlyttle]

If I'm reading the conversation correctly it sounds like this feature, though potentially useful, has edge cases that make things sufficiently difficult so as to outweigh the benefits.

I'll close the issue in the hopes that I will be able to do some twiddling myself.

If I come up with something worth talking about, I'll be happy to reopen and make a PR, even if only for the entertainment value ;)

If you (@hadley or @krlmlr) think this is worth keeping open, you'll not offend me by re-opening.

[krlmlr]

I think we could just add the time zone verbatim: Date-times use 19 characters anyway, most time zone descriptors (known to my system) are 14 characters or shorter:

quantile(nchar(OlsonNames()))
#>   0%  25%  50%  75% 100% 
#>    2   12   14   17   32

This means using <dt-...> as column type will display the time zone in full for half of the time zones, and it will be abbreviated automatically (and occupy only 19 spaces) if too wide. Thoughts?

[hadley]

Olson names are probably amenable to human shortening - split in two pieces, manually abbreviate first and then call abbreviate() on the second? (Use ... if there's three pieces?)

[krlmlr]

@ijlyttle: Would you like to take a stab?

[ijlyttle]

@krlmlr: I would like to - first, let me look and see to make sure I can get my level-of-understanding where it needs to be to contribute.

[ijlyttle]

I had a look, and I think I understand the idea - at its core, we would want a function that would take an Olson string, then return an abbreviated string of no more than 14 characters. Something like:

#' Abbreviate an Olson timezone to 14 characters
abbreviate_olson <- function(x) {

}

Then, you could take returned string and embed it into <dt-...>, which would be a maximum of 19 characters.

Alternatively, such a function could return all (max) 19 characters.

I don't know if you will want this in its own function or not, but it will give me a place to start.

Let me see what I can do here.

[ijlyttle]

Just a heads-up that I don't think I can get to this until the end-of-the week (2017-09-16). I am eager to do it, but I don't want to be blocking things either.

[randomgambit]

hello there! I am just wondering, what about DST time changes? I mean, assume I print something between midnight and 5am on a DST day, the column header will be wrong, right? Can you possibly print some additional stuff? Like an asterisk or something that indicates this is a DST day?

[ijlyttle]

As I understand it, the column header will refer to a time zone that is fixed to a geography, i.e. "America/New_York", which will not change if it is DST (or not).

How to address the significance of values for which DST is in effect would be a separate question, IMHO.

[randomgambit]

huuuum the more I think about it, the more printing the offset seems dead wrong.... The header will be likely to be wrong whenever the data encompasses a DST switch - which is very likely. Most useful in my opinion would be to have the Olson name (America-NY) abbreviated if needed.

[krlmlr]

Closing, should be possible to implement with the help of tidyverse/tibble#411 or a similar approach.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.