Excel Date Cells Are Inconsistently Inferred As Strings or Serial Numbers in ST_READ

[manickmanick]

When reading an .xlsx file using ST_READ() from the duckdb npm package , the date column is inconsistently interpreted.

• Some dates are returned as strings (e.g., '22/05/2000')
• Others are returned as Excel serial numbers (e.g., '37349' for '04/03/2002')

Environment :
npm package :
duckdb - 1.3.1
duckdb-async - 1.3.1

Platform - macos
node version - v18.20.8
Excel Editor: Microsoft Excel on macOS

My node js code below :

duckDbConnection.executeQuery( CREATE TABLE ${excelTempTableName} AS SELECT * FROM ST_READ('${filePath}', layer='${sheet}',
open_options=['HEADERS=FORCE','CELL_TEXT=YES'], allowed_drivers=["XLSX"]), )

steps to reproduce :

1. create an xlsx file

2. Upload the excel and try to read the excel using duckdb npm package like the code i shared above.

3. Observed output :

so when i write below query to get the type of data for that column, i receive a bigint type.

const res = await duckDbConnection.executeQuery(SELECT * FROM ${excelTempTableName}); console.log('table ============> ', res);

4. Expected output :

All date1 values should be consistently interpreted as proper date type.

I tried below things to fix :

1. I formatted the date1 column as date in excel file but when i upload the file then that date1 column is interpreted as varchar type.
   [
   {
   column_name: 'id',
   column_type: 'INTEGER',
   min: '1',
   max: '7',
   approx_unique: 5n,
   avg: '3.6',
   std: '2.4083189157584592',
   q25: '2',
   q50: '3',
   q75: '6',
   count: 5n,
   null_percentage: 0
   },
   {
   column_name: 'date1',
   column_type: 'VARCHAR',
   min: '22/05/2000',
   max: '38689',
   approx_unique: 5n,
   avg: null,
   std: null,
   q25: null,
   q50: null,
   q75: null,
   count: 5n,
   null_percentage: 0
   }
   ]

Request :

1. Consistently infer date columns (preferably to DATE or ISO string format)
2. Provide an option like PARSE_DATES=TRUE to always convert serials to readable date strings
3. Respect CELL_TEXT=YES to return visible Excel values, not underlying serials