how to specify columns with loadCSV?

[fundef1]

Hi, I'm trying to load a CSV and manually specify the columns.
it's a bit of a degenerate case as I only have a single column, but I can't figure out the correct format:

const csvopts = {
	auto_detect: false, 
	all_varchar: true,
	columns: "{'line': 'VARCHAR'}", 
	header: false
};

vg.coordinator().exec(vg.loadCSV("lines", `${window.location}data/lines.csv`,csvopts));

results in

Uncaught (in promise) Error: Parser Error: syntax error at or near "line"
LINE 1: ..._size=-1, all_varchar=true, columns='{'line': 'VARCHAR'}', header=false,...

the quoting of columns looks off and I've tried different variants of quoting the columns option but just can't get it to work.

any hints? thanks!

[jheer]

The loadCSV method supports a select option, as described in the docs here: https://uwdata.github.io/mosaic/api/sql/data-loading.html#loadcsv

For example: vg.loadCSV(table, filename, { select: ['ColumnA', 'ColumnB'] })

[fundef1]

Right, but that's not what I'm after. I do not want to select a subset of columns.
DuckDB requires that you specify the columns with auto_detect false:

Additional CSV-specific options, see the [DuckDB CSV documentation](https://duckdb.org/docs/data/csv/overview).

and in the table with csv options from duckdb:

columns	A struct that specifies the column names and column types contained within the CSV file (e.g., {'col1': 'INTEGER', 'col2': 'VARCHAR'}).	struct	(empty)

it's this struct that I need to pass to duckdb.

[jheer]

Apologies, I triaged too quickly and misunderstood. This is a shortcoming, as input string values will always be quoted. We need to expand the option processor to support objects (-> structs) and arrays as well. I will create an issue for this.

[jheer]

Fixed by #183