Fix the problem, and the tests.

packages/backend-core/src/sql/sql.ts

@@ -122,7 +122,7 @@ function generateSelectStatement(
 
   const schema = meta.table.schema
   return resource.fields.map(field => {
-    const [table, column, ..._rest] = field.split(/\./g)
+    const [table, column, ...rest] = field.split(/\./g)
     if (
       client === SqlClient.POSTGRES &&
       schema[column].externalType?.includes("money")
@@ -138,13 +138,21 @@ function generateSelectStatement(
       // HH:mm format
       return knex.raw(`CONVERT(varchar, ${field}, 108) as "${field}"`)
     }
-    return `${field} as ${field}`
-    // return knex.raw(
-    //   `${quote(client, table)}.${quote(client, column)} as ${quote(
-    //     client,
-    //     field
-    //   )}`
-    // )
+
+    // There's at least two edge cases being handled in the expression below.
+    //  1. The column name could start/end with a space, and in that case we
+    //     want to preseve that space.
+    //  2. Almost all column names are specified in the form table.column, except
+    //     in the case of relationships, where it's table.doc1.column. In that
+    //     case, we want to split it into `table`.`doc1.column` for reasons that
+    //     aren't actually clear to me, but `table`.`doc1` breaks things with the
+    //     sample data tests.
+    return knex.raw(
+      `${quote(client, table)}.${quote(
+        client,
+        [column, ...rest].join(".")
+      )} as ${quote(client, field)}`
+    )
   })
 }
 

packages/server/src/api/routes/tests/search.spec.ts

@@ -38,18 +38,18 @@ import { structures } from "@budibase/backend-core/tests"
 import { DEFAULT_EMPLOYEE_TABLE_SCHEMA } from "../../../db/defaultData/datasource_bb_default"
 
 describe.each([
-  // ["in-memory", undefined],
-  // ["lucene", undefined],
+  ["in-memory", undefined],
+  ["lucene", undefined],
   ["sqs", undefined],
-  // [DatabaseName.POSTGRES, getDatasource(DatabaseName.POSTGRES)],
-  // [DatabaseName.MYSQL, getDatasource(DatabaseName.MYSQL)],
-  // [DatabaseName.SQL_SERVER, getDatasource(DatabaseName.SQL_SERVER)],
-  // [DatabaseName.MARIADB, getDatasource(DatabaseName.MARIADB)],
+  [DatabaseName.POSTGRES, getDatasource(DatabaseName.POSTGRES)],
+  [DatabaseName.MYSQL, getDatasource(DatabaseName.MYSQL)],
+  [DatabaseName.SQL_SERVER, getDatasource(DatabaseName.SQL_SERVER)],
+  [DatabaseName.MARIADB, getDatasource(DatabaseName.MARIADB)],
 ])("search (%s)", (name, dsProvider) => {
   const isSqs = name === "sqs"
   const isLucene = name === "lucene"
   const isInMemory = name === "in-memory"
-  const isInternal = isSqs || isLucene
+  const isInternal = isSqs || isLucene || isInMemory
   const config = setup.getConfig()
 
   let envCleanup: (() => void) | undefined
@@ -115,10 +115,7 @@ describe.each([
       if (isInMemory) {
         return dataFilters.search(_.cloneDeep(rows), this.query)
       } else {
-        return config.api.row.search(table._id!, {
-          ...this.query,
-          tableId: table._id!,
-        })
+        return config.api.row.search(this.query.tableId, this.query)
       }
     }
 
@@ -187,6 +184,15 @@ describe.each([
       return row
     }
 
+    // By default, the global `table` variable is used to search against. If you
+    // need to, though, you can specify a different table to search against. You
+    // may wish to do this in a nested describe block that defines a different
+    // table.
+    againstTable(table: Table) {
+      this.query.tableId = table._id!
+      return this
+    }
+
     // Asserts that the query returns rows matching exactly the set of rows
     // passed in. The order of the rows matters. Rows returned in an order
     // different to the one passed in will cause the assertion to fail.  Extra
@@ -735,29 +741,6 @@ describe.each([
           query: {},
         }).toHaveLength(1)
       })
-
-      isInternal &&
-        describe("space at end of column name", () => {
-          beforeAll(async () => {
-            table = await createTable({
-              "name ": {
-                name: "name ",
-                type: FieldType.STRING,
-              },
-            })
-            await createRows([{ ["name "]: "foo" }, { ["name "]: "bar" }])
-          })
-
-          it("should be able to query a column that starts with a space", async () => {
-            await expectSearch({
-              query: {
-                string: {
-                  "1:name ": "foo",
-                },
-              },
-            }).toContainExactly([{ ["name "]: "foo" }])
-          })
-        })
     })
 
     describe("equal", () => {
@@ -2205,8 +2188,7 @@ describe.each([
         }).toContainExactly([{ name: "baz", productCat: undefined }])
       })
     })
-
-  isInternal &&
+  ;(isSqs || isLucene) &&
     describe("relations to same table", () => {
       let relatedTable: Table, relatedRows: Row[]
 
@@ -2394,6 +2376,7 @@ describe.each([
       beforeAll(async () => {
         await config.api.application.addSampleData(config.appId!)
         table = DEFAULT_EMPLOYEE_TABLE_SCHEMA
+        rows = await config.api.row.fetch(table._id!)
       })
 
       it("should be able to search sample data", async () => {
@@ -2478,4 +2461,76 @@ describe.each([
       }).toContainExactly([{ [name]: "a" }])
     })
   })
+
+  // This is currently not supported in external datasources, it produces SQL
+  // errors at time of writing. We supported it (potentially by accident) in
+  // Lucene, though, so we need to make sure it's supported in SQS as well. We
+  // found real cases in production of column names ending in a space.
+  isInternal &&
+    describe("space at end of column name", () => {
+      beforeAll(async () => {
+        table = await createTable({
+          "name ": {
+            name: "name ",
+            type: FieldType.STRING,
+          },
+        })
+        await createRows([{ ["name "]: "foo" }, { ["name "]: "bar" }])
+      })
+
+      it("should be able to query a column that ends with a space", async () => {
+        await expectSearch({
+          query: {
+            string: {
+              "name ": "foo",
+            },
+          },
+        }).toContainExactly([{ ["name "]: "foo" }])
+      })
+
+      it("should be able to query a column that ends with a space using numeric notation", async () => {
+        await expectSearch({
+          query: {
+            string: {
+              "1:name ": "foo",
+            },
+          },
+        }).toContainExactly([{ ["name "]: "foo" }])
+      })
+    })
+
+  // This was never actually supported in Lucene but SQS does support it, so may
+  // as well have a test for it.
+  ;(isSqs || isInMemory) &&
+    describe("space at start of column name", () => {
+      beforeAll(async () => {
+        table = await createTable({
+          " name": {
+            name: " name",
+            type: FieldType.STRING,
+          },
+        })
+        await createRows([{ [" name"]: "foo" }, { [" name"]: "bar" }])
+      })
+
+      it("should be able to query a column that starts with a space", async () => {
+        await expectSearch({
+          query: {
+            string: {
+              " name": "foo",
+            },
+          },
+        }).toContainExactly([{ [" name"]: "foo" }])
+      })
+
+      it("should be able to query a column that starts with a space using numeric notation", async () => {
+        await expectSearch({
+          query: {
+            string: {
+              "1: name": "foo",
+            },
+          },
+        }).toContainExactly([{ [" name"]: "foo" }])
+      })
+    })
 })

packages/server/src/sdk/app/rows/search/sqs.ts

@@ -210,7 +210,6 @@ async function runSqlQuery(
     let bindings = query.bindings
 
     // quick hack for docIds
-
     const fixJunctionDocs = (field: string) =>
       ["doc1", "doc2"].forEach(doc => {
         sql = sql.replaceAll(`\`${doc}\`.\`${field}\``, `\`${doc}.${field}\``)