Actually, we get correct key condition Key condition: (column 0 in [4294967280, +inf)), (column 0 in (-inf, 0]), and And it's always false for all ranges, but for some reason mayBeTrueAfter https://github.com/clickhouse/ClickHouse/blob/master/src/Storages/MergeTree/KeyCondition.cpp#L1372-L1378 return true.

Actually our code works badly with impossible KeyConditions. In this case, we have condition:

(column 0 in [4294967280, +inf)), (column 0 in (-inf, 0]), and
            >>>>>first<<<<<                  >>>>>second<<<<<

Which is obviously always false for any interval. But for example, for interval [0, +inf) our code will consider that it intersects first interval (on [4294967280, +inf)) and second interval in point 0. After that, we will evaluate AND (https://github.com/clickhouse/ClickHouse/blob/master/src/Storages/MergeTree/KeyCondition.cpp#L1304-L1311)[link] and yes, in both ranges our key condition "can be true" and we will try to read something from disk (actually full column).

I think we can simplify KeyCondition's RPN after it was parsed from WHERE expression, and get rid of similar bugs. For example, if we have OR expression and left interval contains the right interval than we can replace this element with the left interval. Or if a left interval of AND expression doesn't intersect the right interval than we can replace AND with ALWAYS_FALSE (this case). But it can be quite complicated for a multidimensional KeyConditions.

Closed in favor to #14225.