Use and record writer time zone in ORC files

[electrum]

Early versions of the Apache ORC writer made the mistake of recording
timestamps from an epoch that was relative to the time zone of the
writer. This was fixed in later versions by recording the writer time
zone in the stripe footer. Hive 3.1 always writes using UTC.

Presto used a global configuration for the writer time zone, which
was needed to handle old files, but was never updated to use the time
zone from the stripe footer.

On read, Presto now uses the stripe value if present, otherwise it
uses the configured value. On write, Presto continues to write
timestamps using the configured time zone, but now records this value
when writing files.

[findepi]

Hive 3.1 always writes using UTC. ..
On write, Presto continues to write
timestamps using the configured time zone, but now records this value
when writing files.

Can you share rationale for not using UTC when writing?

For timestamp values, using some other zone than UTC may mean that some values cannot be represented, right?

[sopel39]

For timestamp values, using some other zone than UTC may mean that some values cannot be represented, right?

Could you give an example? I don't think there are gaps in time in terms of missing milliseconds from relative epoch.

[electrum]

Can you share rationale for not using UTC when writing?

We should switch in the future, but if we did it now, older versions of Presto would return the wrong answer for these files (if hive.time-zone is not UTC). Maybe switch this in a year, after which we hope no one is using mixed versions over the same set of files.

For timestamp values, using some other zone than UTC may mean that some values cannot be represented, right?

The time zone is for the epoch, not the timestamp itself. The only potential (already existing) problem is if the epoch (2015-01-01 00:00:00) did not exist in that time zone.

[dain]

One comment, but otherwise looks good