backport TZ fixes from downstream #209
Conversation
devs-mycroft
added
the
CLA: Yes
There was a problem hiding this comment.
Hey Chance, are naive datetimes being improperly handled, or is there disagreement that we should default to UTC?
Where do we think the majority of naive datetime objects would be coming from? If we assume it to be calls to datetime.datetime.now() then perhaps we should be defaulting to the system timezone and then converting that to the default?
At the end of the day, I still think any manipulation we do of timezone info on naive datetime objects means we are guessing and likely hiding errors for downstream projects. The fact that there is disagreement of what should be the default indicates to me that there is no good universal default.
|
Review changes: Irrelevant funcs removed. Default TZ set to UTC. |
There was a problem hiding this comment.
Thanks for removing those.
I don't think we've yet got to the bottom of the timezone rabbithole. Have you got a project where you actually use this? Some code we can look at to reason out what would happen in different scenarios?
The question in my mind is whether we assume that the majority of tz naive datetime objects are being generated
- In reference to the system clock (such as by
datetime.datetime.now()) or - by something that inherently references the users local timezone like like an API or some manually constructed datetime object.
For case 1 - if I have something that calls datetime.datetime.now() and then try to convert that datetime to local or UTC - replacing it with either UTC or some other default TZ - the result is going to be incorrect depending on what the system timezone is set to. So if case 1 is the most likely then my assumption is that we should be replacing it with the systems timezone info
If case 2 is more likely then replacing it with the configured __default_tz makes the most sense to me.
lingua_franca/time.py
Outdated
| @@ -17,7 +17,7 @@ | |||
| from dateutil.tz import gettz, tzlocal | |||
|
|
|||
|
|
|||
| __default_tz = None | |||
| __default_tz = gettz("UTC") | |||
There was a problem hiding this comment.
Does this make the default_timezone() function obsolete? at least the or tzlocal() seems to be.
| @@ -46,7 +46,7 @@ def now_utc(): | |||
| Returns: | |||
| (datetime): The current time in Universal Time, aka GMT | |||
| """ | |||
| return to_utc(datetime.utcnow()) | |||
| return datetime.now(gettz("UTC")) | |||
| # naive datetimes assumed to be in default timezone already! | ||
| # in the case of datetime.now this corresponds to tzlocal() | ||
| # otherwise timezone is undefined and can not be guessed, we assume | ||
| # the user means "my timezone" and that LN was configured to use it | ||
| # beforehand, if unconfigured default == tzlocal() |
There was a problem hiding this comment.
We'll need to update these comments
|
Having lost the (internal) single source of truth, I'd just as happily clobber naive DTs to the system TZ. The problem arises in the broader ecosystem. The original reason I suggested using LF.time as the source of truth was to ensure that configs and API params would have something to query that only changes on config. If we're comfortable trusting the system clock, and allowing everything to fail in the cases where that fails, alright. That's what all the downstream cores seem to be doing. |
add to_system utility fix extract_datetime_fr timezone bug MycroftAI#218 fix farsi tests (ignoring timezone due to not being rebased before merge) remove unimplemented farsi code (is_fractional and normalize) split english tests into their own files like the other languages
* port timezone fixes for time utils MycroftAI#209 add to_system utility fix extract_datetime_fr timezone bug MycroftAI#218 fix farsi tests (ignoring timezone due to not being rebased before merge) remove unimplemented farsi code (is_fractional and normalize) split english tests into their own files like the other languages + remove solved TODO comment authored-by: jarbasai <[email protected]>
add to_system utility fix extract_datetime_fr timezone bug MycroftAI#218 fix farsi tests (ignoring timezone due to not being rebased before merge) remove unimplemented farsi code (is_fractional and normalize) split english tests into their own files like the other languages
This attempts to fix #203 using portions of HelloChatterbox#32. All tests pass on my computer, but they were already doing that. Paging @forslund.