Consider the following test case:

<pre>
CREATE TABLE t0(c0 COLLATE NOCASE);
INSERT INTO t0(c0) VALUES ('B');
CREATE VIEW v0(c0, c1) AS SELECT DISTINCT t0.c0, 'a' FROM t0;
SELECT v0.c1 >= v0.c0 FROM v0; -- actual: 1, expected: 0
</pre>

I would expect the query to yield TRUE. The condition expectedly yields FALSE when used in a WHERE condition:

<pre>
SELECT * FROM v0 WHERE v0.c1 >= v0.c0; -- no row is fetched
</pre>

This appears to be due to the push-down optimization that pushes the
"C1 >= C0" constraint down into the view, without taking collating
functions into account.  This can be seen in that disabling the push-down
optimization results in different answers:

<blockquote><verbatim>
CREATE TABLE t0(c0 COLLATE NOCASE);
INSERT INTO t0(c0) VALUES ('B');
CREATE VIEW v0(c0, c1) AS SELECT DISTINCT t0.c0, 'a' FROM t0;
SELECT count(*) FROM v0 WHERE c1 >= c0;
.testctrl optimizations 0x1000
SELECT count(*) FROM v0 WHERE c1 >= c0;
</verbatim></blockquote>

Fixed by [18458b1a].