Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-26979

[PySpark] Some SQL functions do not take column names

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Resolved
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: 2.4.0
    • Fix Version/s: 3.0.0
    • Component/s: PySpark
    • Labels:
    • Docs Text:
      Hide
      In Spark 3.0, certain Pyspark SQL functions that previously only accepted a Column object to specify a column can now take a string specifying the column by name, for consistency with the rest of the API:

      lower()
      upper()
      abs()
      bitwiseNOT()
      ltrim()
      rtrim()
      trim()
      ascii()
      base64()
      unbase64()
      Show
      In Spark 3.0, certain Pyspark SQL functions that previously only accepted a Column object to specify a column can now take a string specifying the column by name, for consistency with the rest of the API: lower() upper() abs() bitwiseNOT() ltrim() rtrim() trim() ascii() base64() unbase64()

      Description

      Most SQL functions defined in org.apache.spark.sql.functions have two variations, one taking a Column object as input, and another taking a string representing a column name, which is then converted into a Column object internally.

      There are, however, a few notable exceptions:

      • lower()
      • upper()
      • abs()
      • bitwiseNOT()

      While this doesn't break anything, as you can easily create a Column object yourself prior to passing it to one of these functions, it has two undesirable consequences:

      1. It is surprising - it breaks coder's expectations when they are first starting with Spark. Every API should be as consistent as possible, so as to make the learning curve smoother and to reduce causes for human error;
      2. It gets in the way of stylistic conventions. Most of the time it makes Python/Scala/Java code more readable to use literal names, and the API provides ample support for that, but these few exceptions prevent this pattern from being universally applicable.

      This is a very easy fix, and I see no reason not to apply it. I have a PR ready.

      UPDATE: Turns out there are many exceptions over this pattern that I wasn't aware of. The reason I missed them is because I had been looking at things from PySpark's point of view, and the API there does support column name literals for almost all SQL functions.

      Exceptions for the PySpark API include all the above plus:

      • ltrim()
      • rtrim()
      • trim()
      • ascii()
      • base64()
      • unbase64()

      The argument for making the API consistent still stands, however. I have been working on a PR to fix this on PySpark's side, and it should still be a painless change. 

        Attachments

          Activity

            People

            • Assignee:
              amello Andre Sa de Mello
              Reporter:
              amello Andre Sa de Mello
            • Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: