1. Looks like you failed 721 tests of 2186.

With Columnstore 1.2.3 and InnoDB as comparing engine.

This appears to only break on the edge, such as i1 = max(type) – Columnstore max, that is.

In the following, you can see using i1 results in NULL, while using a constant gives the correct result.
MariaDB [test]> select i1, -126 | i1 from fact where i1 = 126;
---------------+

---------------+

---------------+
1 row in set (0.027 sec)

MariaDB [test]> select i1, -126 | 126 from fact where i1 = 126;
--------------------------+

--------------------------+

--------------------------+
1 row in set (0.022 sec)

A goodly number of these are caused by maxing out the integer field. Any value that happens to be our magic NULL number gets reported as NULL, rather than that value.

Other issues include:

1) Not handling string constants containing numbers correctly. Corrected.
2) Decimals with > 9 decimal digits cause integer overflow in an internal calculation. Corrected.
3) Decimal Constants > Columnstore's decimal limit. Not corrected.
4) For certain float values with large exponents, it gets a different answer than InnoDB. Not corrected.

Fixed some things, but not all. This can be closed, but I don't know how to handle a partial fix.

Changes won't compile

Fixed compile errors

1) For Decimal Constants too large, this will be handled by MCOL-641 Full DECIMAL support.
2) Fixing float inaccuracies is difficult to impossible. These micro variations are considered acceptable.

Build verified: 1.2.6-1

engine commit:
d4173ef

Verified the fix for NULL output.

Micro variance for float are expected.