public / sequelize

Feature/mssql dialect

findAndCountAll empty include error fixed.

…com/sequelize/sequelize/pull/2697)

There have been many many little changes to the codebase during
the development of the mssql dialect that are no longer needed, as
well as missed imports of changes upstream. This realigns the
disperate trees, and fixes a few related bugs

There was some refactoring on attributes test where mssql dialect
branching was missed. Also added databaseVersion support for mssql
dialect

temporarily disable on update foreign key constraints, mssql dialect
now passes all current test cases

Major overhaul of the createTableQuery to bring it closer in line
with the mysql dialect. This closes a number of failed cases where
things were being created in a strange manner, as well as let's us
do away with SqlGenerator once and for all

These tests must be disabled since MSSQL simply doesn't allow
the updating of these types of fields. Cascading updates are supported
and should be covered by other tests.

Instead of trying to parse the error messages in the connection
manager, forward all errors to the user to deal with on an as-needed
basis

There are more than one error messages related to unique
constraints that we need to catch. This closes one more broken
test, in ancitipation of the crate table refactoring which closes
the other.

There is no need to use the mssql wrapped around tedious, and
switching to the lower level module allows better integration
as well as bypassing the mssql forced connection pooling

Code was merged into master to allow modification of the abstract
dialect to support mssql specific cases. This removes the custom
implementations

This is more of a stylistic change, as no functional behavior differs,
however it makes diagnostics much easier and is theoretically faster
than the previous select code.

MSSQL doesn't support RESTRICT foreign key constraints at all,
so there's no use running any of those tests

This is completely broken anyway, these tests are being used to
conditionally report back false positives for transaction support.

 - modify the returned sql for isolation level
 - remove some commented out code
 - cheat a little by reordering tests (look into this)

  - major deletion of code duplication
  - cleaned up code (removed duplicate documentation)
  - moved a lot of SqlGenerator sql back into QueryGenerator

  - missed a case for addLimitAndOffset for subqueries with a single model
  - formatResults should return undefined if there is no data
  - there are a few possible error messages in an invalid connection scenario

  - bring addIndexQuery in line with abstract dialect
  - use an improved method for showing indices with sp_helpindex
  - actually use handleShowIndexQuery (bringing us closer to the
    abstract dialect for formatResults)
  - temporarily disable tests for a FULLTEXT index

DataTypes.DATE was set to "DATETIME2", which has implications for internal
sequelize code. Instead we should be using a data type mapping, like postgres
support.