fix: EXPOSED-731 Timestamp support for SQLite is broken

[obabichevjb]

Description

It's the fix for timestamp() column. That column doesn't support time zones, and should store the values in local timezones.

But it was working different and converting values into UTC timezone. On reading that converted value was recognized as local, and shifted one more time to convert it to UTC.

After fix Exposed will store values with the value in local time zone, and expect that the value in local time zone in the moment of reading.

Tests also read the value from database as string, and check the value, to be sure that the value was not shifted to UTC in the moment of insert, and back in the moment of reading.

---

Type of Change

Please mark the relevant options with an "X":

• Bug fix

Affected databases:

• Mysql5
• Mysql8
• SQLite

Checklist

---

Related Issues

EXPOSED-731 Timestamp support for SQLite is broken

[obabichevjb]

This split into <FORMATTER>_NOTZ and FORMATTER is intended. It solves flaky issue with timezones.

The problem is that lazy caches the value, and that value is cached for the whole test run. And before .withZone() was inside the lazy call, so the first call was caching formatter with particular timezone.

For users it was not a problem, because usually people do not change current timezone in the code. But we do that in tets.

[bog-walk]

Is there any chance these changes fix this TODO, or is that being too hopeful? 😢

[obabichevjb]

It doesn't fix (I'll recheck one more time) the problem for SQLite

I was moving these setDefault() higher when understood that lazy caches the values, and I wanted that even it's cached, the value for the whole test is the same. But actually it doesn't help on run multiple tests, because in this case the first test would init lazy and all other tests use it.

But I was surprised, default value of time zone is reset before every test (even if one gradle job runs multiple tests)

[obabichevjb]

Actually, I found an intersting issue, I can't move that line back under withTables, it failes (with timestamps errors), but error seems to be in threading now...

The test is green if TimeZone.setDefault() is before withTables, and red otherwise. I tried to find the moment when moving TimeZone.setDefault() starts to break the test.

I stopped in the function withConnection:

private suspend fun <T> withConnection(body: suspend Connection.() -> T): T {
    if (localConnection == null) {
        // If the `setDefault()` here, test is **Green**
        TimeZone.setDefault(TimeZone.getTimeZone("Africa/Cairo")) 
        val awaitFirst = connection.awaitFirst()
        // If the `setDefault()` here, test is **Red**
        TimeZone.setDefault(TimeZone.getTimeZone("Africa/Cairo"))

        localConnection = awaitFirst.also {
            // this starts an explicit transaction with autoCommit mode off
            it.beginTransaction().awaitFirstOrNull()
        }
    }
    return localConnection!!.body()
}

I'm not sure why exactly that line breaks the TimeZone. The main hypothesis is that TimeZone() set default changes time zone only or current thread, and it's not friendly to koroutines, and we should expect that time zone is not changing on runtime.