For me, the most annoying thing about Python is that out of the box, AFAICT, you cannot get a TZ-aware datetime. In fact, while tzinfo is part of the standard library, I don't know of any included solution for making TZ-aware datetime, short of writing your own subclasses for each timezone.
EDIT: datetime.utcnow() is TZ-aware, I meant any other timezone.
I agree that dateutil is excellent. I think it's also worth mentioning how powerful dateutil's `parse()` is. You can pass virtually any string representation of a date to it and it will return a Python datetime.
dateutil includes tzinfo implementations 'tzutc' (whose offset is always zero), 'tzoffset' (whose offset is passed as a parameter) and 'tzlocal' (whose offset is the value of time.timezone). All these could easily and safely be implemented in the stdlib without having to push out updates every time the Olson timezone database has a new release.
It also contains tzinfo implementations that read timezone definitions from a system-wide copy of the Olson database (for most POSIX OSs) or the registry (for Windows-based OSs). These could also be added to the standard library without requiring regular Python updates, because the OS's usual timezone update system will take care of things.
Perhaps the stdlib can't possibly solve timezone issues in all possible cases, but it could be a lot more useful than it currently is.
dateutil is really great; j2labs is quick to point out `parse()` which I also think is worth mentioning on its own.
I think this project still fills a nice niche. datetime.now(los_angeles).astimezone(minsk).strftime(fmt) is long enough that you're going to have to find the StackOverflow thread in question to implement this, and maybe go to the stdlib. This library makes that a lot less painful.
If you go to the source, you'll see how tiny it is! It's nowhere near the complexity of dateutil. I'd love to see some of dateutil's darker corners wrapped into this library.
This is good, but for my app, we just decided to do all of the presentation on the client side using javascript. This prevents us from having to know or store the users tz at all. That way, whatever tz they are in is the correct one. =)
We thought this might be an issue for things like emailing purchase receipts to people since that is something that can't just be 'rendered' on the client, but we ended up just asking ourselves "What would amazon do?" and looked at an emailed Amazon receipt.
You will notice that there isn't a single time in the receipt other than when the email was sent. Looked at a receipt from Apple. Same thing. Problem solved by just removing the timestamp from the receipt.
Thought that was an interesting tidbit that I hadn't thought of before.
I recently was evaluating JS and Python for handling time in my app. The goal was not to localize to a specific timezone for the client, but rather to inform the client of where in the world a current time is.
I ended up picking the Python backend solution because it was easier to reason about with the more abstract `datetime` primitives.
I don't think its fair to say one should "never work with local times". I frequently work with "floating time", the opposite of "fixed time".
You can accept user input as (year, month, day, hour, minute, second), convert to a common timezone, and store as seconds since epoch. To display, convert back to the common timezone, but omit the timezone identifier. This is useful for events which are local for the user, since they don't care about their own timezone.
See the iCalendar spec,
They are used to represent the same hour, minute, and second value regardless of which time zone is currently being observed. For example, an event can be defined that indicates that an individual will be busy from 11:00 AM to 1:00 PM every day, no matter which time zone the person is in. In these cases, a local time can be specified.
I don't think "2012-02-03 09:30:00+0100" is valid ISO 8601 because of the space; it should be a "T" to be a proper "combined date and time" representation.
The date and time representations may appear in proximity to each other, often separated by a space or sometimes by other characters. In these cases they occupy two separate fields in a data system, rather than a single combined representation. This is usually done for human readability. Unlike the previous examples, "2007-04-05 14:30" is considered two separate, but acceptable, representations—one for date and the other for time. It is then left to the reader to interpret the two separate representations as meaning a single time point based on the context.
Great initiative. Dealing with timezone is a pain and I totally agree with you that storing universal time is the way to go. It's a little bit like the 'validating input problem': Strongly validating and converting input/output so that the entire application can be safe. I.e. it's not to the sqrt function to validate its parameters. (Although it'd throw an exception if it fails.)
This solves one of my problems with datetime and timezones in Python. The other being that it is totally awkward and bafflingly hard to add and substract from datetime objects.
Yup, works fine if you're doing things only locally. However, if you're using timezone aware datetime objects you're out of luck, because then datetime.timedelta doesn't immediately apply.
Just creating timezone aware datetime objects is really too difficult with the standard library anyway. You need to install `pytz` or create your own `datetime.tzinfo` classes just to be able to represent timezones.
To top it off, if you have datetime objects with timezones and datetime objects without timezone information (*even created using utcfromtimezone()!), you can't compare them.
Months and years are tricky because they can consist of a variable number of days. To me, it's a better way of handling things without being ambiguous. For example, is one month from now exactly 30 days from today, or is it the next month on the same month day? It's pretty simple to do either with the datetime library.
When taking into account DST, you should first convert from local time to UTC (pytz does this) before doing your calculations, then convert it back to local time if needed.
And yes, you do get a delta when subtracting two dates.
EDIT: datetime.utcnow() is TZ-aware, I meant any other timezone.