jacktengg opened a new pull request, #70204:
URL: https://github.com/apache/doris/pull/70204
### What problem does this PR solve?
Issue Number: N/A
Related PR: N/A
Problem Summary:
TIMESTAMPTZ input in a time zone gap loses its position within that gap. For
example, New York skips from 01:59:59 to 03:00:00 on 2024-03-10. Doris converts
local 02:30:00 to 03:00:00. The required result is 03:30:00, or 07:30:00 UTC.
BE uses `cctz::convert`, which selects the transition instant for skipped
local time. FE constant folding uses the same policy. Use the offset before the
transition for TIMESTAMPTZ input instead. This moves the input forward by the
actual gap length and keeps its fraction.
Apply this policy to strict and fallback string parsing, session-local
strings, and DATETIMEV2/TIMESTAMP_NS casts. Use one FE helper for these
TIMESTAMPTZ paths. Round the local value before the zone lookup so a carry that
leaves the gap selects the correct offset.
Minimal reproduction:
```sql
SET time_zone = 'America/New_York';
SET debug_skip_fold_constant = true;
SELECT CAST('2024-03-10 02:30:00' AS TIMESTAMPTZ(6));
-- Result: 2024-03-10 03:30:00.000000-04:00
```
The regression test uses constant queries and inserts no rows. It covers gap
boundaries, fractions, a 30-minute gap on Lord Howe, fixed offsets, repeated
local time, and FE/BE agreement. Its output file was generated by the
regression script.
### Release note
TIMESTAMPTZ conversion now moves local input in a time zone gap forward by
the actual gap length. For example, New York local 2024-03-10 02:30:00 now
becomes 03:30:00-04:00.
### Check List (For Author)
- Test:
- Unit Test: 18 BE tests passed under ASAN. FE tests: 52 passed, 2
existing tests skipped, no failures.
- Regression test: 7 suites passed. These cover DST gaps and overlaps,
casts, UTC functions, historical offsets, output boundaries, and TIMESTAMP_NS
named-zone parsing.
- Manual test: the reported query returned `03:30:00.000000-04:00`,
`1800`, and `0`.
- Build: `DISABLE_BE_JAVA_EXTENSIONS=ON ./build.sh --be -j32` and
`./build.sh --fe -j16` passed.
- Checks: clang-format 16, build hygiene, clang-tidy, and FE Checkstyle
passed.
- Behavior changed: Yes. TIMESTAMPTZ gap input retains its position within
the gap.
- Does this need documentation: No. The release note describes the corrected
conversion policy.
Validation commands:
```sh
./run-be-ut.sh --run --filter='TimeStampTzValueTest.*:CastTimeStampTzTest.*'
-j32
./run-fe-ut.sh --run
'org.apache.doris.nereids.trees.expressions.literal.TimestampTzLiteralTest,org.apache.doris.nereids.trees.expressions.literal.TimeStampNsLiteralTest,org.apache.doris.nereids.trees.expressions.literal.DateTimeLiteralTest'
./run-regression-test.sh --run -d datatype_p0/timestamptz -s
test_timestamptz_dst_gap -forceGenOut
./run-regression-test.sh --run -d
datatype_p0/timestamptz,datatype_p0/timestamp_ns -s
test_timestamptz_dst_gap,test_timestamptz_dst_fold,test_timestamptz_cast,test_timestamptz_utc_functions,test_timestamptz_historical_offset,test_timestamptz_output_boundary,test_timestamp_ns_named_zone_gap
-parallel 2
```
Local build note: a temporary change to the ccache version print avoided
SIGPIPE in `env.sh`. That change and test port settings were restored and are
excluded from this PR.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
