library("testthat") # For debugging: force reload of patterns: # rJava::J('org.ohdsi.sql.SqlTranslate')$setReplacementPatterns('inst/csv/replacementPatterns.csv') expect_equal_ignore_spaces <- function(string1, string2) { string1 <- gsub("([;()'+-/|*\n])", " \\1 ", string1) string2 <- gsub("([;()'+-/|*\n])", " \\1 ", string2) string1 <- gsub(" +", " ", string1) string2 <- gsub(" +", " ", string2) expect_equivalent(string1, string2) } test_that("translate sql server -> Big Query select random row using hash", { sql <- translate("SELECT column FROM (SELECT column, ROW_NUMBER() OVER (ORDER BY HASHBYTES('MD5',CAST(person_id AS varchar))) tmp WHERE rn <= 1", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select column from (select column, row_number() over (order by md5(cast(person_id as STRING))) tmp where rn <= 1" ) }) test_that("translate sql server -> BigQuery SELECT CONVERT(VARBINARY, @a, 1)", { sql <- translate("SELECT ROW_NUMBER() OVER CONVERT(VARBINARY, val, 1) rn WHERE rn <= 1", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select row_number() over safe_cast(concat('0x', val) as int64) rn where rn <= 1" ) }) test_that("translate sql server -> bigquery lowercase all but strings and variables", { sql <- translate("SELECT X.Y, 'Mixed Case String' FROM \"MixedCaseTableName.T\" where x.z=@camelCaseVar GROUP BY X.Y", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select x.y, 'Mixed Case String' from `MixedCaseTableName.T` where x.z=@camelCaseVar group by x.y" ) }) test_that("translate sql server -> bigquery common table expression column list", { sql <- translate("with cte(x, y, z) as (select c1, c2 as y, c3 as r from t) select x, y, z from cte;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "with cte as (select c1 as x, c2 as y, c3 as z from t) select x, y, z from cte;" ) }) test_that("translate sql server -> bigquery common table expression column list no from or union", { sql <- translate("WITH data(x) AS (SELECT (CAST(1 AS INT) x)) SELECT x INTO my_table FROM data;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "CREATE TABLE my_table AS WITH data as (select (cast(1 as int64) x) as x) SELECT x FROM data;" ) }) test_that("translate sql server -> bigquery multiple common table expression column list", { sql <- translate("with cte1 as (select 2), cte(x, y, z) as (select c1, c2 as y, c3 as r from t) select x, y, z from cte;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "with cte1 as (select 2), cte as (select c1 as x, c2 as y, c3 as z from t) select x, y, z from cte;" ) }) test_that("translate sql server -> bigquery distinct keyword", { sql <- translate("with cte2 (column1, column2) as (select distinct c1.column1, c1.column2 from cte c1) select column1, column2 into cte2 from cte2", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "with cte2 as (select distinct c1.column1 as column1,c1.column2 as column2 from cte c1) select column1, column2 into cte2 from cte2" ) }) test_that("translate sql server -> bigquery group by function", { sql <- translate("select f(a), count(*) from t group by f(a);", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select f(a), count(*) from t group by 1;") }) test_that("translate sql server -> bigquery group by addition", { sql <- translate("select 100, sum(x), cast(a+b as string) from t group by a+b;", targetDialect = "bigquery" ) expect_equal_ignore_spaces(sql, "select 100, sum(x), cast(a+b as string) from t group by 3;") }) test_that("translate sql server -> bigquery column ref groupby", { sql <- translate("select 100, sum(x), cast(a+b as string) from t group by t.a, t.b;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select 100, sum(x), cast(a+b as string) from t group by t.a, t.b;" ) }) test_that("translate sql server -> bigquery group by without match", { sql <- translate("select 100, sum(x), concat('count = ', c) from t group by a+b;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select 100, sum(x), concat('count = ', c) from t group by a + b;" ) }) test_that("translate sql server -> bigquery group by without final semicolon", { sql <- translate("select f(a) from t group by f(a);", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select f(a) from t group by 1;") }) test_that("translate sql server -> bigquery order by", { sql <- translate("select f(a) from t group by f(a) order by f(a);", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select f(a) from t group by 1 order by 1;") }) test_that("translate sql server -> bigquery nested group by", { sql <- translate("select * from (select 100, cast(a+b as string), max(x) from t group by a+b) dt;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select * from (select 100, cast(a+b as string), max(x) from t group by 2) dt;" ) }) test_that("translate sql server -> bigquery complex group by", { sql <- translate("select 100, 200, cast(floor(date_diff(a, b, day)/30) string string), 300 from t group by floor(date_diff(a, b, day)/30);", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select 100, 200, cast(floor(date_diff(a, b, day)/30) string string), 300 from t group by 3;" ) }) test_that("translate sql server -> bigquery group by having", { sql <- translate(paste( "select cast(stratum_1 as integer) as concept_id, sum(count_value) as count_value ", "from heracles_results ", "where analysis_id in (123) ", "group by cast(stratum_1 as integer) ", "having sum(count_value) > 1;" ), targetDialect = "bigquery") expect_equal_ignore_spaces( sql, paste( "select cast(stratum_1 as INT64) as concept_id, sum(count_value) as count_value ", "from heracles_results ", "where analysis_id in (123) ", "group by 1 ", "having sum(count_value) > 1;" ) ) }) test_that("translate sql server -> bigquery column references", { sql <- translate("select concat(t.a, t.b) from t group by t.a, t.b;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select concat(t.a, t.b) from t group by t.a, t.b;") }) test_that("translate sql server -> bigquery mixed column references", { sql <- translate("select concat(t.a, t.b), x+y+z from t group by t.a, t.b, x+y+z;", targetDialect = "bigquery" ) expect_equal_ignore_spaces(sql, "select concat(t.a, t.b), x+y+z from t group by t.a, t.b, 2;") }) test_that("translate sql server -> bigquery DATEDIFF", { sql <- translate("SELECT DATEDIFF(dd,drug_era_start_date,drug_era_end_date) FROM drug_era;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select DATE_DIFF(IF(SAFE_CAST(drug_era_end_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(drug_era_end_date AS STRING)),SAFE_CAST(drug_era_end_date AS DATE)),IF(SAFE_CAST(drug_era_start_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(drug_era_start_date AS STRING)),SAFE_CAST(drug_era_start_date AS DATE)),DAY)from drug_era;" ) }) test_that("translate sql server -> bigquery DATEDIFF year", { sql <- translate("SELECT DATEDIFF(YEAR,drug_era_start_date,drug_era_end_date) FROM drug_era;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select (EXTRACT(YEAR from IF(SAFE_CAST(drug_era_end_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d', cast(drug_era_end_date AS STRING)),SAFE_CAST(drug_era_end_date AS DATE))) - EXTRACT(YEAR from IF(SAFE_CAST(drug_era_start_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d', cast(drug_era_start_date AS STRING)),SAFE_CAST(drug_era_start_date AS DATE)))) from drug_era;" ) }) test_that("translate sql server -> bigquery DATEADD", { sql <- translate("SELECT DATEADD(dd,30,drug_era_end_date) FROM drug_era;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select DATE_ADD(IF(SAFE_CAST(drug_era_end_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(drug_era_end_date AS STRING)),SAFE_CAST(drug_era_end_date AS DATE)), INTERVAL 30 DAY) from drug_era;" ) }) test_that("translate sql server -> bigquery DATEADD non-integer", { sql <- translate("SELECT DATEADD(dd,30.0,drug_era_end_date) FROM drug_era;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select DATE_ADD(IF(SAFE_CAST(drug_era_end_date AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(drug_era_end_date AS STRING)),SAFE_CAST(drug_era_end_date AS DATE)), INTERVAL 30 DAY) from drug_era;" ) }) test_that("translate sql server -> bigquery GETDATE", { sql <- translate("GETDATE()", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "CURRENT_DATE()") }) test_that("translate sql server -> bigquery STDEV", { sql <- translate("stdev(x)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "STDDEV(x)") }) test_that("translate sql server -> bigquery LEN", { sql <- translate("len('abc')", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "LENGTH('abc')") }) test_that("translate sql server -> bigquery COUNT_BIG", { sql <- translate("COUNT_BIG(x)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "COUNT(x)") }) test_that("translate sql server -> bigquery CAST varchar", { sql <- translate("select cast(x as varchar)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select cast(x as STRING)") }) test_that("translate sql server -> bigquery CAST :float", { sql <- translate("select cast(x as:float)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select CAST(x as float64)") }) test_that("translate sql server -> bigquery DROP TABLE IF EXISTS", { sql <- translate("IF OBJECT_ID('cohort', 'U') IS NOT NULL DROP TABLE cohort;", targetDialect = "bigquery" ) expect_equal_ignore_spaces(sql, "DROP TABLE IF EXISTS cohort;") }) test_that("translate sql server -> bigquery CAST string", { sql <- translate("CAST(x AS VARCHAR(255))", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "cast(x as STRING)") }) test_that("translate sql server -> bigquery LEFT, RIGHT", { sql <- translate("select LEFT(a, 20), RIGHT(b, 30) FROM t;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select SUBSTR(a, 0, 20), SUBSTR(b, -30) from t;") }) test_that("translate sql server -> bigquery cast float", { sql <- translate("cast(a as float)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "cast(a as float64)") }) test_that("translate sql server -> bigquery cast bigint", { sql <- translate("cast(a as bigint)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "cast(a as int64)") }) test_that("translate sql server -> bigquery cast int", { sql <- translate("cast(a as int)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "cast(a as int64)") }) test_that("translate sql server -> bigquery cast date", { sql <- translate("date(d)", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "IF(SAFE_CAST(d AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(d AS STRING)),SAFE_CAST(d AS DATE))" ) }) test_that("translate sql server -> bigquery cast concat string as date", { sql <- translate("cast(concat(a,b) as date)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "parse_date('%Y%m%d', concat(a,b))") }) test_that("translate sql server -> bigquery cast string as date", { sql <- translate("cast(a as date)", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "IF(SAFE_CAST(a AS DATE) IS NULL,PARSE_DATE('%Y%m%d',cast(a AS STRING)),SAFE_CAST(a AS DATE))" ) }) test_that("translate sql server -> bigquery extract year", { sql <- translate("year(d)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "EXTRACT(YEAR from d)") }) test_that("translate sql server -> bigquery extract month", { sql <- translate("month(d)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "EXTRACT(MONTH from d)") }) test_that("translate sql server -> bigquery extract day", { sql <- translate("day(d)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "EXTRACT(DAY from d)") }) test_that("translate sql server -> bigquery union distinct", { sql <- translate("select 1 as x union select 2;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select 1 as x union distinct select 2;") }) test_that("translate sql server -> bigquery intersect distinct", { sql <- translate("SELECT DISTINCT a FROM t INTERSECT SELECT DISTINCT a FROM s;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select distinct a from t INTERSECT DISTINCT select distinct a from s;" ) }) test_that("translate sql server -> bigquery bracketed intersect distinct", { sql <- translate("(SELECT DISTINCT a FROM t INTERSECT SELECT DISTINCT a FROM s)", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "(select distinct a from t INTERSECT DISTINCT select distinct a from s)" ) }) test_that("translate sql server -> bigquery isnull", { sql <- translate("SELECT isnull(x,y) from t;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select IFNULL(x,y) from t;") }) test_that("translate sql server -> bigquery unquote aliases", { sql <- translate("SELECT a as \"b\" from t;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select a as b from t;") }) test_that("translate sql server -> bigquery cast int in coalesce", { sql <- translate("select coalesce(x, 0), coalesce(12, y) from t", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "select coalesce(cast(x as int64), 0), coalesce(12, cast(y as int64)) from t" ) }) test_that("translate sql server -> bigquery cast decimal", { sql <- translate("select cast(x as decimal(18,4)) from t", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select cast(x as float64) from t") }) test_that("translate sql server -> bigquery ISNUMERIC", { sql <- translate("select ISNUMERIC(a) from b", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "select CASE WHEN SAFE_CAST(a AS FLOAT64) IS NULL THEN 0 ELSE 1 END from b" ) sql <- translate("select a FROM table WHERE ISNUMERIC(a) = 1", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "select a from table where CASE WHEN SAFE_CAST(a AS FLOAT64) IS NULL THEN 0 ELSE 1 END = 1" ) sql <- translate("select a FROM table WHERE ISNUMERIC(a) = 0", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "select a from table where CASE WHEN SAFE_CAST(a AS FLOAT64) IS NULL THEN 0 ELSE 1 END = 0" ) }) test_that("translate sql server -> bigquery index not supported", { sql <- translate("CREATE INDEX idx_raw_4000 ON #raw_4000 (cohort_definition_id, subject_id, op_start_date);", targetDialect = "bigquery" ) expect_equal_ignore_spaces(sql, "-- bigquery does not support indexes") }) test_that("translate sql server -> bigquery TRUNCATE TABLE", { sql <- translate("TRUNCATE TABLE cohort;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "DELETE FROM cohort WHERE True;") }) test_that("translate sql server -> bigquery DATEFROMPARTS", { sql <- translate("select DATEFROMPARTS(2019,1,30)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select DATE(2019,1,30)") }) test_that("translate sql server -> bigquery EOMONTH()", { sql <- translate("select eomonth(payer_plan_period_start_date)", targetDialect = "bigquery") expect_equal_ignore_spaces( sql, "select DATE_SUB(DATE_TRUNC(DATE_ADD(payer_plan_period_start_date, INTERVAL 1 MONTH), MONTH), INTERVAL 1 DAY)" ) }) test_that("translate sql server -> bigquery escape chars", { sql <- translate("INSERT INTO t VALUES('some \"string\" ''with escape'' chars')", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "insert into t values(CONCAT('some \\042string\\042 ','\\047','with escape','\\047','chars'))" ) }) test_that("translate SELECT INTO + CTE bigquery", { sql <- translate("WITH data (a,b) AS (SELECT 1, 2 UNION ALL SELECT 3, 4) SELECT a,b INTO test FROM data;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "CREATE TABLE test AS WITH data as (select 1 as a, 2 as b union all select 3, 4) SELECT a,b FROM data;" ) }) test_that("translate sql server -> BigQuery UPDATE STATISTICS", { sql <- translate("UPDATE STATISTICS results_schema.heracles_results;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "-- big query does not support such functionality") }) test_that("translate sql server -> BigQuery modulus", { sql <- translate("SELECT row_number() over (order by cast(person_id % 123 as int))", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select row_number() over (order by CAST(MOD(person_id, 123) AS INT64))" ) sql <- translate("SELECT row_number() over (order by cast((person_id % 123) as int))", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select row_number() over (order by CAST(MOD(person_id, 123) AS INT64))" ) }) test_that("translate sql server -> BigQuery % operator", { sql <- translate("SELECT (CAST(person_id*month(cohort_start_date) AS BIGINT) % 123)*(CAST(year(cohort_start_date)*day(cohort_start_date) AS BIGINT) % 123)) FROM my_table;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select (MOD(cast(person_id*EXTRACT(MONTH from cohort_start_date) as int64), 123))*(MOD(cast(EXTRACT(YEAR from cohort_start_date)*EXTRACT(DAY from cohort_start_date) as int64), 123))) from my_table;" ) }) test_that("translate sql server -> String concatenation", { sql <- translate("SELECT last_name + ', ' + first_name FROM my_table;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select CONCAT(last_name, ', ', first_name) FROM my_table;" ) }) test_that("translate sql server -> String concatenation", { sql <- translate("SELECT first_name + CAST(middle_initial AS VARCHAR) + last_name FROM my_table;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select CONCAT(first_name, CAST(middle_initial AS STRING), last_name) FROM my_table;" ) }) test_that("translate sql server -> String concatenation", { sql <- translate("SELECT first_name + CAST(middle_initial AS VARCHAR(1)) + last_name FROM my_table;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select CONCAT(first_name, CAST(middle_initial AS STRING), last_name) FROM my_table;" ) }) test_that("translate sql server -> String concatenation", { sql <- translate("SELECT subgroup_id, 'Persons aged ' + cast(age_low as varchar) + ' to ' + cast(age_high as varchar) + ' with gender = ' + gender_name FROM subgroups;", targetDialect = "bigquery" ) expect_equal_ignore_spaces( sql, "select subgroup_id, CONCAT('Persons aged ', CAST(age_low AS STRING), CONCAT('to ', cast(age_high as STRING)), 'with gender = ', gender_name ) FROM subgroups;" ) }) test_that("translate sql server -> bigquery DROP TABLE IF EXISTS", { sql <- translate("DROP TABLE IF EXISTS test;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "drop table if exists test;") }) test_that("translate sql server -> bigquery nullable field", { sql <- translate("CREATE TABLE test (x NUMERIC NULL, y NUMERIC NOT NULL);", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "create table test (x NUMERIC, y numeric not null);") }) test_that("translate sql server -> bigquery NEWID()", { sql <- translate("SELECT *, NEWID() FROM my_table;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select *, GENERATE_UUID() from my_table;") }) test_that("translate sql server -> bigquery DBPLYR alias collision", { sql <- translate("SELECT * FROM (SELECT *, ROW_NUMBER() OVER (ORDER BY RAND()) AS q01 FROM cdmv5.dbo.person) q01 WHERE (q01 <= 10)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select * from (select *, row_number() over (order by rand()) AS val_q01\n from cdmv5.dbo.person) q01 where (val_q01 <= 10)") }) test_that("translate sql server -> bigquery IIF", { sql <- translate("SELECT IIF(a>b, 1, b) AS max_val FROM table;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select CASE WHEN a>b THEN 1 ELSE b END as max_val from table ;") }) test_that("translate sql server -> bigquery drvd()", { sql <- translate("SELECT TRY_CAST(name AS VARCHAR(MAX)) AS name, TRY_CAST(speed AS FLOAT) AS speed FROM ( VALUES ('A', 1.0), ('B', 2.0)) AS drvd(name, speed);", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select\n CAST(name as STRING) as name,\n cast(speed as float64) as speed\n FROM (SELECT NULL AS name, NULL AS speed UNION ALL SELECT 'A', 1.0 UNION ALL SELECT 'B', 2.0 LIMIT 999999 OFFSET 1) AS values_table;") }) test_that("translate sql server -> bigquery temp table field ref", { sql <- translate("SELECT #tmp.name FROM #tmp;", targetDialect = "bigquery", tempEmulationSchema = "ts") expect_equal_ignore_spaces(sql, sprintf("select %stmp.name from ts.%stmp;", getTempTablePrefix(), getTempTablePrefix())) }) test_that("translate sql server -> bigquery temp dplyr ... pattern", { sql <- translate("SELECT * FROM table...1;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select * from tablexxx1;") }) test_that("translate sql server -> bigquery quotes", { sql <- translate("SELECT \"a\" from t;", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "select `a` from t;") }) test_that("translate sql server -> bigquery RIGHT with implicit concat", { sql <- translate("RIGHT('0' + CAST(p.month_of_birth AS VARCHAR), 2)", targetDialect = "bigquery") expect_equal_ignore_spaces(sql, "SUBSTR(CONCAT('0', cast(p.month_of_birth as STRING)),-2)") }) test_that("translate sql server -> bigquery create temp table", { sql <- translate("CREATE TABLE #temp (x INT);", targetDialect = "bigquery", tempEmulationSchema = "ts") expect_equal_ignore_spaces(sql, sprintf("DROP TABLE IF EXISTS ts.%stemp;\nCREATE TABLE ts.%stemp (x INT64);", getTempTablePrefix(), getTempTablePrefix())) }) test_that("translate sql server -> bigquery select into temp table", { sql <- translate("SELECT * INTO #temp FROM my_table;", targetDialect = "bigquery", tempEmulationSchema = "ts") expect_equal_ignore_spaces(sql, sprintf("DROP TABLE IF EXISTS ts.%stemp;\nCREATE TABLE ts.%stemp AS\nSELECT\n* \nFROM\nmy_table;", getTempTablePrefix(), getTempTablePrefix())) }) test_that("translate sql server -> bigquery create temp table if not exists", { sql <- translate("CREATE TABLE IF NOT EXISTS #temp (x INT);", targetDialect = "bigquery", tempEmulationSchema = "ts") expect_equal_ignore_spaces(sql, sprintf("create table if not exists ts.%stemp (x INT64);", getTempTablePrefix())) })