I am storing date/times in the database as UTC and computing them inside my application back to local time based on the specific timezone. Say for example I have the following date/time:
01/04/2010 00:00
Say it is for a country e.g. UK which observes DST (Daylight Savings Time) and at this particular time we are in daylight savings. When I convert this date to UTC and store it in the database it is actually stored as:
31/03/2010 开发者_如何学Python23:00
As the date would be adjusted -1 hours for DST. This works fine when your observing DST at time of submission. However, what happens when the clock is adjusted back? When I pull that date from the database and convert it to local time that particular datetime would be seen as 31/03/2010 23:00
when in reality it was processed as 01/04/2010 00:00
.
Correct me if I am wrong but isn't this a bit of a flaw when storing times as UTC?
Example of Timezone conversion
Basically what I am doing is storing the date/times of when information is being submitted to my system in order to allow users to do a range report. Here is how I am storing the date/times:
public DateTime LocalDateTime(string timeZoneId)
{
var tzi = TimeZoneInfo.FindSystemTimeZoneById(timeZoneId);
return TimeZoneInfo.ConvertTimeFromUtc(DateTime.UtcNow, tzi).ToUniversalTime().ToLocalTime();
}
Storing as UTC:
var localDateTime = LocalDateTime("AUS Eastern Standard Time");
WriteToDB(localDateTime.ToUniversalTime());
You don't adjust the date for DST changes based on whether you're currently observing them - you adjust it based on whether DST is observed at the instant you're describing. So in the case of January, you wouldn't apply the adjustment.
There is a problem, however - some local times are ambiguous. For example, 1:30am on October 31st 2010 in the UK can either represent UTC 01:30 or UTC 02:30, because the clocks go back from 2am to 1am. You can get from any instant represented in UTC to the local time which would be displayed at that instant, but the operation isn't reversible.
Likewise it's very possible for you to have a local time which never occurs - 1:30am on March 28th 2010 didn't happen in the UK, for example - because at 1am the clocks jumped forward to 2am.
The long and the short of it is that if you're trying to represent an instant in time, you can use UTC and get an unambiguous representation. If you're trying to represent a time in a particular time zone, you'll need the time zone itself (e.g. Europe/London) and either the UTC representation of the instant or the local date and time with the offset at that particular time (to disambiguate around DST transitions). Another alternative is to only store UTC and the offset from it; that allows you to tell the local time at that instant, but it means you can't predict what the local time would be a minute later, as you don't really know the time zone. (This is what DateTimeOffset
stores, basically.)
We're hoping to make this reasonably easy to handle in Noda Time, but you'll still need to be aware of it as a possibility.
EDIT:
The code you've shown is incorrect. Here's why. I've changed the structure of the code to make it easier to see, but you'll see it's performing the same calls.
var tzi = TimeZoneInfo.FindSystemTimeZoneById("AUS Eastern Standard Time");
var aussieTime = TimeZoneInfo.ConvertTimeFromUtc(DateTime.UtcNow, tzi);
var serverLocalTime = aussieTime.ToLocalTime();
var utcTime = serverLocalTime.ToUniversalTime();
So, let's think about right now - which is 13:38 in my local time (UTC+1, in London), 12:38 UTC, 22:39 in Sydney.
Your code will give:
aussieTime = 22:39 (correct)
serverLocalTime = 23:39 (*not* correct)
utcTime = 22:39 (*not* correct)
You should not call ToLocalTime
on the result of TimeZoneInfo.ConvertTimeFromUtc
- it will assume that it's being called on a UTC DateTime
(unless it's actually got a kind of DateTimeKind.Local
, which it won't in this case).
So if you're accurately saving 22:39 in this case, you aren't accurately saving the current time in UTC.
It's good that you are attempting to store the dates and times as UTC. It is generally best and easiest to think of UTC as the actual date and time and local times are just pseudonyms for that. And UTC is absolutely critical if you need to do any math on the date/time values to get timespans. I generally manipulate dates internally as UTC, and only convert to local time when displaying the value to the user (if it's necessary).
The bug that you are experiencing is that you are incorrectly assigning the local time zone to the date/time values. In January in the UK it is incorrect to interpret a local time as being in a Summertime time zone. You should use the time zone that was in effect at the time and location that the time value represents.
Translating the time back for display depends entirely on the requirements of the system. You could either display the times as the user's local time or as the source time for the data. But either way, Daylight Saving/Summertime adjustments should be applied appropriately for the target time zone and time.
You could work around this by also storing the particular offset used when converting to UTC. In your example, you'd store the date as something like
31/12/2009 23:00 +0100
When displaying this to the user, you can use that same offset to convert, or their current local offset, as you choose.
This approach also comes with its own problems. Time is a messy thing.
The TimeZoneInfo.ConvertTimeFromUtc() method will solve your problem:
using System;
class Program {
static void Main(string[] args) {
DateTime dt1 = new DateTime(2009, 12, 31, 23, 0, 0, DateTimeKind.Utc);
TimeZoneInfo tz = TimeZoneInfo.FindSystemTimeZoneById("GMT Standard Time");
Console.WriteLine(TimeZoneInfo.ConvertTimeFromUtc(dt1, tz));
DateTime dt2 = new DateTime(2010, 4, 1, 23, 0, 0, DateTimeKind.Utc);
Console.WriteLine(TimeZoneInfo.ConvertTimeFromUtc(dt2, tz));
Console.ReadLine();
}
}
Output:
12/31/2009 11:00:00 PM
4/2/2010 12:00:00 AM
You'll need .NET 3.5 or better and run on an operating system that keeps historical daylight saving time changes (Vista, Win7 or Win2008).
Correct me if I am wrong but isn't this a bit of a flaw when storing times as UTC?
Yes it is. Also, days of the adjustment will have either 23 or 25 hours so the idiom of the prior day at the same time being local time - 24 hours is wrong 2 days a year.
The fix is picking one standard and sticking with it. Storing dates as UTC and displaying as local is pretty standard. Just don't use a shortcut of doing calculations local (+- somthing) = new time and you are OK.
This is a huge flaw but it isn't a flaw of storing times in UTC (because that is the only reasonable thing to do -- storing local times is always a disaster). This is a flaw is the concept of daylight savings time. The real problem is that the time zone information changes. The DST rules are dynamic and historic. They time when DST starting in USA in 2010 is not the same when it started in 2000. Until recently Windows did not even contain this historic data, so it was essentially impossible to do things correctly. You had to use the tz database to get it right. Now I just googled it and it appears that .NET 3.5 and Vista (I assume Windows 2008 too) has done some improvement and the System.TimeZoneInfo actually handles historic data. Take a look at this.
But basically DST must go.
精彩评论