returns wrong results:
The problem is that in the collation utf8_polish_ci
the letter "O WITH ACUTE" is not equal to "O":
Therefore, returning the records containing accented letter
in the above query is wrong.
Further investigation with EXPLAIN:
tells that MATCH erroneously chooses the non-suitable index "i"
whose collation is NOT equal to the collation of the operation.
If I further drop the index and use BOOLEAN MODE:
it still returns wrong results:
It seems MATCH always uses the collation of the field and does not follow the
standard argument collation aggregation rules.