use nu_test_support::fs::Stub::FileWithContentToBeTrimmed; use nu_test_support::playground::Playground; use nu_test_support::{nu, pipeline}; #[test] fn flatten_nested_tables_with_columns() { let actual = nu!( cwd: ".", pipeline( r#" echo [[origin, people]; [Ecuador, ('Andres' | wrap name)]] [[origin, people]; [Nu, ('nuno' | wrap name)]] | flatten --all | flatten --all | get name | str join ',' "# )); assert_eq!(actual.out, "Andres,nuno"); } #[test] fn flatten_nested_tables_that_have_many_columns() { let actual = nu!( cwd: ".", pipeline( r#" echo [[origin, people]; [Ecuador, (echo [[name, meal]; ['Andres', 'arepa']])]] [[origin, people]; [USA, (echo [[name, meal]; ['Katz', 'nurepa']])]] | flatten --all | flatten --all | get meal | str join ',' "# )); assert_eq!(actual.out, "arepa,nurepa"); } #[test] fn flatten_nested_tables() { let actual = nu!( cwd: ".", pipeline( r#" echo [[Andrés, Nicolás, Robalino]] | flatten | get 1 "# )); assert_eq!(actual.out, "Nicolás"); } #[test] fn flatten_row_column_explicitly() { Playground::setup("flatten_test_1", |dirs, sandbox| { sandbox.with_files(vec![FileWithContentToBeTrimmed( "katz.json", r#" [ { "people": { "name": "Andres", "meal": "arepa" } }, { "people": { "name": "Katz", "meal": "nurepa" } } ] "#, )]); let actual = nu!( cwd: dirs.test(), "open katz.json | flatten people --all | where name == Andres | length" ); assert_eq!(actual.out, "1"); }) } #[test] fn flatten_row_columns_having_same_column_names_flats_separately() { Playground::setup("flatten_test_2", |dirs, sandbox| { sandbox.with_files(vec![FileWithContentToBeTrimmed( "katz.json", r#" [ { "people": { "name": "Andres", "meal": "arepa" }, "city": [{"name": "Guayaquil"}, {"name": "Samborondón"}] }, { "people": { "name": "Katz", "meal": "nurepa" }, "city": [{"name": "Oregon"}, {"name": "Brooklin"}] } ] "#, )]); let actual = nu!( cwd: dirs.test(), "open katz.json | flatten --all | flatten people city | get city_name | length" ); assert_eq!(actual.out, "4"); }) } #[test] fn flatten_table_columns_explicitly() { Playground::setup("flatten_test_3", |dirs, sandbox| { sandbox.with_files(vec![FileWithContentToBeTrimmed( "katz.json", r#" [ { "people": { "name": "Andres", "meal": "arepa" }, "city": ["Guayaquil", "Samborondón"] }, { "people": { "name": "Katz", "meal": "nurepa" }, "city": ["Oregon", "Brooklin"] } ] "#, )]); let actual = nu!( cwd: dirs.test(), "open katz.json | flatten city --all | where people.name == Katz | length" ); assert_eq!(actual.out, "2"); }) } #[test] fn flatten_more_than_one_column_that_are_subtables_not_supported() { Playground::setup("flatten_test_4", |dirs, sandbox| { sandbox.with_files(vec![FileWithContentToBeTrimmed( "katz.json", r#" [ { "people": { "name": "Andres", "meal": "arepa" } "tags": ["carbohydrate", "corn", "maiz"], "city": ["Guayaquil", "Samborondón"] }, { "people": { "name": "Katz", "meal": "nurepa" }, "tags": ["carbohydrate", "shell food", "amigos flavor"], "city": ["Oregon", "Brooklin"] } ] "#, )]); let actual = nu!( cwd: dirs.test(), "open katz.json | flatten tags city --all" ); assert!(actual.err.contains("tried flattening")); assert!(actual.err.contains("but is flattened already")); }) }