Reveng: Only get NONCLUSTERED and CLUSTERED indexes from sys.indexes

[ErikEJ]

to avoid name being NULL and also ignore other irrelevant index types
https://msdn.microsoft.com/da-dk/library/ms173760.aspx
Fixes some of the issues reported in #3861

[lajones]

@ErikEJ Is this still needed after the fixes in #3934?

[natemcmaster]

I don't believe we support SPATIAL index (and others) in Model. We should log about unsupported index types.

[ErikEJ]

@lajones Yes, in order to avoid spatial and other index types, and to clarify the intent of GetIndexes

[lajones]

@ErikEJ I'm with @natemcmaster I think it would be better to bring them back in the query and issue a warning message along the lines of "Index [{schemaName}].[{tableName}].[{indexName}] has type {typeDescription}. We only support CLUSTERED and NONCLUSTERED. Skipping index." (similar to what we already do for columns with types we don't understand). Note: we already bring back the type_desc column in the query so that should help.

[ErikEJ]

@lajones That would also work, yes, but just looking at the type_desc seems quite brittle (type_desc could be called something else for other DBMS systems http://docs.oracle.com/cd/B19306_01/server.102/b14237/statviews_1069.htm#i1578369 ). I guess it also boils down to what you expect a provider to give you from GetIndexes (currently unclear). If that was documented, then it would probably make sense for a provider to limit the list of "indexes" returned to be actually something useful for reverse engineering

[lajones]

@ErikEJ Yes - this is for SQL Server only i.e. SqlServerDatabaseModelFactory.GetIndexes() would add an appropriate IndexModel to the DatabaseModel if type_desc is good and would log the warning saying why not if not. Other providers would be free to do different things / issue their own warning messages.

Current plans for the IndexModel objects are to support the functionality listed in #3710. Though I agree that when this is finalized it would be good to document what we expect all the *Model objects to contain.

[ErikEJ]

So you prefer to deal with potentially hard to understand warnings for unsupported index types?

[lajones]

Well I hope the warnings will be as helpful as we can make them, but, yes, essentially I'd rather have warnings rather than just ignore them without explanation.