With the fixed integers test including the case statement, I ran 1e8 loops in Oracle in 16.4 seconds against 1e7 loops in SQL Server in 7.5 seconds.

You’re right, that’s the way we should always do it… and of course Uncle Tom would agree with you !
“You assume : i don’t care
Test, test, test, show me results and evidence !”
(Right, it’s probably not an exact quote :))

But i’m a bit sad that once again, your results show that SQL Server performs better :) :) :)

Two tricks :
- As i generated the months between two bounds, the resulting interval was : [0, 11] instead of [1, 12]
… where the value 0 should have been 12.
How can you fix this using only "arithmetic" operators ?
After trying various combinations of interval translations, i was just looking for the "discontinous" expression that would return 1 when 0 and 0 when between 1 and 11.
My first thought was that "floor" is the simplest "discontinous" function.
1 <= x 0 < (12 – x) / 12 floor((12 – x) / 12) = 0
x = 0 => (12 – x) / 12 = 1

Then, we just have to use this boolean “belonging” function to add 12 to the base expression only when x = 0 :
x + 12 * floor((12 – x) / 12)
(instead of x)

- In the real world, we use date functions to perform operations between two dates, because the year and the months are not in the same radix.
But of course, when you substract yyyymm1 to yyyymm, there should be a simple arithmetic expression that transforms the result into a year / month expression. That’s what all the mod(, 88) stuff is :
mm – mm1 belongs to [1, 12] when mm > mm1, [99, 88] when mm1 < mm
(mod 88 lets the first interval unchanged, translates the second one into [1, 11])

Thanks for linking my Pacblog !
Of course i’ve linked yours too (first of all i had to find out how the administration interface works :)). There are now too sites : your’s and the Master’s !

You did a small cheat in your solution adding the current month into the initial rowset !

It’s always nice to see other ways on how to solve the same problem !

It's always a pleasure to visit your blog.
Until now, i just enjoyed your posts, and was each time really impressed to see you were always learning / finding new tricks :)

But today, i was really motivated by this #12 challenge and tried to write a little variation of your query.

I tried it in the opposite way : prepare the data with lead function, then theta-join it with the number CTE...
The rest are just practical jokes (maybe because i was bored at the office :)) :
- using arithmetic difference and mod manipulation to avoid the date functions
- arithmetic tricks to change a 0 into a 12 without CASE
- row generation with dba_objects...

Hum, i'm just looking forward to see your T-SQL implementation !

Good luck ;)