The code for sql_column_change used specifically for mssql in db/tools.php deals with looking for any constraints that reference a column and dropping them. Either this function needs to be used or the code being used needs to be re-used when dropping columns, at least for MSSQL. May need some massaging in the case of indexes.

Here is a sample query that will get you all of the indexes in the users table that use the username column:

SELECT DISTINCT ix.name

FROM sys.indexes ix

INNER JOIN sys.index_columns ixc ON ixc.object_id = ix.object_id AND ixc.index_id = ix.index_id

INNER JOIN sys.columns cols ON cols.column_id = ixc.column_id AND cols.object_id = ix.object_id

WHERE ix.object_id = OBJECT_ID('users') AND cols.name = 'username'

It would be a matter of taking each returned index and dropping it before dropping the column. It could be done database side, like it was done in tools.php, but would probably work better looping through the results PHP side.

The MSSQL behaviour kind of makes sense because indexes may be over multiple columns. A user may a) want to recreate the index without the dropped column or b) drop the index completely.

I made a patch which drops the index for us now.
We either added a new index anyway or the column was the only one in the index, so in both cases retaining it without that one column does not make sense

Reopening in order to implement the systematic solution