Roman Vasin

> Backwards incompatible change? Can you change behavior based on server setting? This is no options on server settings to change range back to 1925 ... 2283 because I removed...

While technically it's possible to implement an option like support_old_ext_range to support old (current) range 1925..2283 in current approach (without using the mask). But this will add many check conditions...

Could someone please assign a label to PR? So I could see the results of CI tests.

> @rvasin the tests need to be updated: https://s3.amazonaws.com/clickhouse-test-reports/39425/0b102c6d1ffeeeae79c3f50a522288367a567a46/fast_test__actions_.html Yes, I know. I am working on the tests right now.

@alexey-milovidov Could you please comment on results of Performance Comparison tests? What I see there is no slowdown in tests related to date/time functions. In fact: Performance Comparison [3/4] `SELECT...

@ilejn for this query taken from the performance test: `SELECT count() FROM numbers(50000000) WHERE NOT ignore(toDate('2017-01-01') + number % 1000 + rand() % 10 AS t, toRelativeDayNum(t))` I built CH...

@alexey-milovidov I am on vacation from 1st to 7th August. I also don't think that it's related to L2 cache. I discussed with @ilejn which "plan of actions" could be...

@alexey-milovidov and @ilejn I found the reason of performance degradation of some tests. It's not related at all to L2 cache or new LUT size. In fact in this PR...

@alexey-milovidov In the latest commit I removed the clamping and corrected one functional test. I see on the performance tests there is no slowdown of date/time related functions/queries. There is...

> I've tried to benchmark normalizeDayNum via quickbench https://quick-bench.com/q/gSrrRikIlfAo4S2Nb9EzrqLWqKE . Totally pointless: there is no difference. But funny. Still don't understand the performance difference between runs, branch predictor hardly can...