issue_comments

I confirmed and it's possible to have a SQLite column with a hyphen at the start, confirmed using: ``` % sqlite-utils create-table demo.db demo -- id integer name text -blah integer % sqlite-utils tables --schema demo.db -t table schema

demo CREATE TABLE [demo] ( [id] INTEGER, [name] TEXT, [-blah] INTEGER ) ```

This doesn't work: sqlite-utils create-index mydb.db dogs "-age" name But this does: sqlite-utils create-index mydb.db dogs -- -age name

It's weird having to use Database.DescIndex - I'm going to put DescIndex in sqlite_utils.db directly and let people import it.

PRAGMA index_xinfo(table) DOES return that data: (Pdb) [c[0] for c in fresh_db.execute("PRAGMA index_xinfo('idx_dogs_age_name')").description] ['seqno', 'cid', 'name', 'desc', 'coll', 'key'] (Pdb) fresh_db.execute("PRAGMA index_xinfo('idx_dogs_age_name')").fetchall() [(0, 2, 'age', 1, 'BINARY', 1), (1, 0, 'name', 0, 'BINARY', 1), (2, -1, None, 0, 'BINARY', 0)] See https://sqlite.org/pragma.html#pragma_index_xinfo

Example output: https://covid-19.datasettes.com/covid?sql=select+*+from+pragma_index_xinfo%28%27idx_ny_times_us_counties_date%27%29

Annoyingly the table.indexes property won't indicate if an index is in regular or reverse order - because the SQLite PRAGMA index_info(table) statement doesn't indicate that either. You have to look at the sqlite_master index definition to tell if any of the columns are in reverse order: (Pdb) fresh_db.execute("select * from sqlite_master where type = 'index'").fetchall() [('index', 'idx_dogs_age_name', 'dogs', 3, 'CREATE INDEX [idx_dogs_age_name]\n ON [dogs] ([age] desc, [name])')] (Pdb) fresh_db.execute("PRAGMA index_info('idx_dogs_age_name')").fetchall() [(0, 2, 'age'), (1, 0, 'name')] (Pdb) fresh_db.execute("PRAGMA index_info('idx_dogs_age_name')").description (('seqno', None, None, None, None, None, None), ('cid', None, None, None, None, None, None), ('name', None, None, None, None, None, None)) (Pdb) dogs.indexes [Index(seq=0, name='idx_dogs_age_name', unique=0, origin='c', partial=0, columns=['age', 'name'])]

Decisions: for the Python API I'm going with db.DescIndex("column") as the way to do this.

For the CLI I'm going to do the "-age" thing.

For the CLI version I could say that you can use a - prefix to specify reverse direction: sh sqlite-utils create-index mydb.db dogs -age name No, that doesn't work - it could get confused with a command-line flag. I guess you could do this: sqlite-utils create-index mydb.db dogs "-age" name This does mean that if any of your column names begin with a hyphen you can't use the CLI to add indexes to them. Is that an acceptable limitation? Users can always use sqlite-utils mydb.db "create index ..." in that case.

Need to solve this for the CLI tool too. Currently that works like this: https://sqlite-utils.datasette.io/en/stable/cli.html#creating-indexes sh sqlite-utils create-index mydb.db mytable col1 [col2...] Even harder to decide how to add a descending option to this. Maybe like this? sh sqlite-utils create-index mydb.db mytable --direction col1 asc --direction col2 desc It's a bit gross though! We're saying here that if a single one of the columns you are creating an index for is in reverse direction you have to use --direction to specify each end every other index.

I could use db.desc_index("age") to match SQLite SQL syntax, which uses desc and not descending.

Maybe: python db["dogs"].create_index([db.descending_index("age"), "name"]) It's a little verbose but it's for a relatively rare activity and it does make it very clear what is going on.

A few options: ```python db["dogs"].create_index([("age", "desc"), "name"])

db["dogs"].create_index([desc("age"), "name"])

db["dogs"].create_index([db.desc("age"), "name"]) `` The first option uses an optional tuple. The second two use adesc()` function - the question is where should that live?

sqlite_utils.desc(column) or db.desc(column) are both options.

I don't like using the term desc() for "descending index" though - it feels like it should mean something more broad.

The problem here is differentiating between a column with the name date desc and wanting to create a descending index on a column called date.

This won't work: python db["ny_times_us_counties"].create_index(["date desc"], desc=True) Because we need to be able to create compound indexes with columns with different directions - for example: sql create index idx_age_desc_name on dogs (age desc, name)