Leading with commas – ugly or efficient? An investigation over 320GB of SQL | Hacker News Reader

Leading with commas – ugly or efficient? An investigation over 320GB of SQL

towardsdatascience.com

30 pointsdanso7 comments

That this has three lines could not be easier to see:

  [  la la la la la la la la la la la
  ,  la la la la la la la la la la la 
     la la la la la
  ,  la la la la la la la la la la la 
     la la la la la la la la la la la ]

Seeing that in this requires a little more work:

  [  la la la la la la la la la la la,
     la la la la la la la la la la la 
     la la la la la,
     la la la la la la la la la la la 
     la la la la la la la la la la la ]

In real-world code, the lists might be of expressions that can themselves contain commas, and the lines can be longer and more heterogeneous. All of those factors make the leading comma idiom even more readable relative to the trailing one.

It is easier for normal humans to read trailing commas, but difficult to change during development. If you are actively developing a query, perhaps you could just add a dummy column:

SELECT a, B, c, 0 ...

I like Go’s solution: mandatory trailing commas. Doesn’t work for set-in-stone sql, sadly.

Java, in a few places, has optional trailing commas. That is, if your last item also has a trailing comma, that is okay.

To put this into SQL notation:

SELECT col1, col2, col3, FROM Table

Now the columns were on separate lines, it is easy to comment/uncomment, or copy/paste columns around. just write every column with a comma at the end. But not in SQL -- just in various places in Java code. But I wish that were in SQL.

In Java enums are the first that immediately comes to mind. Every item can have a trailing comma. After the final item (and its optional comma) use a semicolon to begin class level declarations for the enum (if any) such as a custom constructor.

In Java, another example is array initializers:

int[] values = { 1, 2, 3, 4, 5, };

If those initializer items were on separate lines, it is easy to comment/remove/interchange rows because they ALL have a comma.

Do the results stay the same if you exclude files longer than say 1000 lines?

I think a lot of big .sql files in repos might be autogenerated by tools, and being long would be a good way to remove some of them.

I prefer leading with commas, and one column per line.

Another argument for designing languages and formats so that there is no need for the comma.

Reply on news.ycombinator.com