In Postgres I store data given to me by a user with:
Column | Type | Collation | Nullable | Default
------------+--------------------------+-----------+----------+---------
id | uuid | | not null |
value | numeric | | |
date | timestamp with time zone | | |
Now I'm presented with the requirement of maintaining the original timezone in which the data was produced. The timestamp with timezone is normalized to the database's timezone and the original timezone is lost, so I must manually restore date back from the normalized timezone before serving it back to the user.
Most solutions suggest adding an extra column to the table and storing the original timezone information together with the timestamp:
Column | Type | Collation | Nullable | Default
------------+--------------------------+-----------+----------+---------
id | uuid | | not null |
value | numeric | | |
date | timestamp with time zone | | |
tz | text | | |
So given that I'm using Go, which information should I extract from time.Time to store in tz for the most precise and seamless restoration?
date.Location().String() doesn't seem right as it might return the value Local which is relative.
And how should I restore the information from tz back into to time.Time?
Is the result of time.LoadLocation(tz) good enough?
Upon save, I would obtain the zone name and offset using Time.Zone():
func (t Time) Zone() (name string, offset int)
Then when querying such a timestamp from the database, you can construct a time.Location using time.FixedZone():
func FixedZone(name string, offset int) *Location
And switch to this location using Time.In().
Word of caution! This will restore you a timestamp with "seemingly" in the same time zone, but if you need to apply operations on it (such as adding days to it), the results might not be the same. The reason for this is because time.FixedZone() returns a time zone with a fixed offset, which does not know anything about daylight savings for example, while the original timestamp you saved might have a time.Location which does know about these things.
Here's an example of such a deviation. There is a daylight saving day in March, so we'll use a timestamp pointing to March 1, and add 1 month to it, which results in a timestamp being after the daylight saving.
cet, err := time.LoadLocation("CET")
if err != nil {
panic(err)
}
t11 := time.Date(2019, time.March, 1, 12, 0, 0, 0, cet)
t12 := t11.AddDate(0, 1, 0)
fmt.Println(t11, t12)
name, offset := t11.Zone()
cet2 := time.FixedZone(name, offset)
t21 := t11.UTC().In(cet2)
t22 := t21.AddDate(0, 1, 0)
fmt.Println(t21, t22)
now := time.Date(2019, time.April, 2, 0, 0, 0, 0, time.UTC)
fmt.Println("Time since t11:", now.Sub(t11))
fmt.Println("Time since t21:", now.Sub(t21))
fmt.Println("Time since t12:", now.Sub(t12))
fmt.Println("Time since t22:", now.Sub(t22))
This will output (try it on the Go Playground):
2019-03-01 12:00:00 +0100 CET 2019-04-01 12:00:00 +0200 CEST
2019-03-01 12:00:00 +0100 CET 2019-04-01 12:00:00 +0100 CET
Time since t11: 757h0m0s
Time since t21: 757h0m0s
Time since t12: 14h0m0s
Time since t22: 13h0m0s
As you can see, the output time after the 1-month addition is the same, but the zone offset is different, so they designate a different time instant in time (which is proven by showing the time difference with an arbitrary time). The original has 2-hour offset, because it knows about the daylight saving that happened in the 1 month we skipped, while the "restored" timestamp's zone doesn't know about that, so the result has the same 1-hour offset. In the timestamp after addition, even the zone name changes in real life: from CET to CEST, and again, the restored timestamp's zone doesn't know about this either.
A more wasteful, and still prone to error, but still valid solution will be to also store the original timestamp in ISO 8601 format like 2019-05-2T17:24:37+01:00 in a separate column datetext:
Column | Type | Collation | Nullable | Default
------------+--------------------------+-----------+----------+---------
id | uuid | | not null |
value | numeric | | |
date | timestamp with time zone | | |
datetext | text | | |
Then query using date for the strength of the native timestamp column, and return to the user datetext which is the exact value that was originally sent.