Unions of recursive cte queries - regression with merger of pull request #99

[pgammans]

Hi @millerdev,

Two new test case for union of cte queries.
The main one is testcase test_recursive_union_query_with_cte which show the regression caused with the merger of pull request #99 as per the comments by @rgammans the second to ensure we don't break normal unions with CTE models

Unfortunately, this testcase is not the most optimal CTE but shows the issue without adding any new test models.
If pull #99 is reverted this test case then passes, with #99 it cases this testcase to fail raising a OperationalError from the DB driver

[pgammans]

The last commit fixes the issue but i like to look at this in relation to pull #59

[millerdev]

@pgammans thank you for the tests and fix!

with #99 it cases this testcase to fail raising a OperationalError from the DB driver

I did not understand this part. Does that mean there is still a problem with this PR, or are we good to merge this and create a new release?

[pgammans]

@millerdev
This PR was more an example the regression caused by the merge of #99 and not directly a solution.

The test cases in commit e2e70e0 show the regression due to merging #99 this test would pass if the commits in #99 were reverted, then next commit was just to ensure i do not break normal non cte unions.

Commit 54a6788 the 4th commit is just fixed the test case for use on PostgreSQL not SQLite3

This said the commit b7f4b5e is start on fixing the issues so the testcases from #99 and the ones here pass.
After looking at this issue I think the code change in #99 is correct but that when we do the union (and possibly other operations) of a CTE query and another query we end up with the cte on both the CTEQuery attached to the resultant union QuerySet and one of on of the CTEQuerys sub combined_queries the commit b7f4b5e is that start of a possible fix.

with #99 the SQL generated from the test is where before you only have 1 with statement

WITH RECURSIVE "leaf_cte" AS (
     SELECT ....)
WITH RECURSIVE "leaf_cte" AS (
     SELECT ....)
SELECT .... ````

The 1st  come from the query the second comes from the first query in that queries `combined_queries`

[millerdev]

@pgammans, #113 should be a more complete fix. Let me know if you are able to test it. I think we can now close this PR. And thank you for the contribution of test cases and the start to a fix!

[millerdev]

Closing. Please create a new PR if there is anything important here that was not incorporated into #113