sql.py: add_index(): Don't create index columns for nullable columns, because they require indexes to be created on all columns in order to use a distinct_table() temp table. Also, now that we are no longer using LEFT JOINs, the COALESCE call would only be evaluated once (in the plain JOIN) in the event that PostgreSQL doesn't use an index on a COALESCE expression.

schemas/vegbien.sql: location: Dropped unique constraint on lat/long because it covered only some rows, which interfered with column-based import's selection of different insert methods based on the presence or absence of duplicate keys. (With the constraint, locations with coordinates would have duplicates eliminated, but locations without coordinates would not be able to find which row was added for a particular location because there was no lookup key to join on, and would all just use the first inserted row.) The previous behavior didn't make much sense anyway, because it would assert that two locationevents occurred in the same place just because they had the same coordinates, which may not have been precise enough to make this determination. Asserting that two locationevents occurred in the same place is really part of the secondary validation, not the import process.

sql.py: DbConn: Fixed bug where Exceptions did not have the query appended if the query was not run in cacheable mode, by moving _add_cursor_info() from DbCursor.execute() to run_query() so it would also get called for non-cacheable queries that use a native cursor rather than a wrapper. Fixed bug where non-cacheable queries were not autocommitted, by moving self.do_autocommit() from DbCursor.execute() to run_query() so it would also get called for non-cacheable queries that use a native cursor rather than a wrapper.

sql.py: DbConn._db(): Record that a transaction is already open before setting the search_path so that a query is never run with an _savepoint value less than 1 (manual transactions are not supported yet)

sql.py: DbConn.with_savepoint(): Increment _savepoint before running queries so they don't get autocommitted

sql.py: empty_temp(): Empty temp tables even in debug_temp mode, so that it can be seen which tables have been garbage collected and disk space leaks can be detected. This will not affect the external re-runnability of slow queries in debug_temp mode, as long as the user aborts the debug_temp import while the slow query is still running.

sql_gen.py: ColDict: Use OrderedDict so that order of keys in input dict (if ordered) will be preserved. This should ensure that tempt table unique indexes have their columns in the same order as the output table, so that a merge join can be used.

util.py: dict_subset(): Use OrderedDict so that order of keys in input dict (if ordered) will be preserved

main Makefile: python-Darwin: Added pip installation instructions. python-Linux: Added ordereddict.

sql.py: DbConn.col_info(): cacheable param defaults to True now that callers explicitly turn off cacheable when needed