Regression: SELECT ... UNION ... with inconsistent column names fails

A SELECT from a subquery containing a UNION with inconsistent result column names fails with an ill-fitting error message, but only when used with a WHERE clause.
This is a heavily striped down version of a query that fails:

select * from (

    select 1 x

    union all

    select sum(1) y

  ) t where x>0

yields: ERROR 1054 (42S22): Unknown column 'x' in 'order clause'
which is strange, because there is no order clause.

This slightly different statement, however, works fine:

select * from (

    select 1 x

    union all

    select 2 y

  ) t  where x>0

(Note: second branch of UNION has no aggregate function)
and yields, as expected:

+---+

| x |

+---+

| 1 |

| 2 |

+---+

Omitting the WHERE clause from the first query allows it to succeed, too.

This seems to be a regression. My original query (much longer) used to work with an older 10.x version, but I cannot reconstruct any more which version that was.
On an old MySQL server 5.7.9, the problem does NOT come up. SQLite also has NO problems with the query, so I assume the query is legal as such, even given the inconsistent naming of columns.

Of course, a workaround is easy: use consistent aliases:

select * from (

    select 1 x

    union all

    select sum(1) x

  ) t where x>0