There was an interesting (aka nerdy) question in the SAP Community Q&A the other day:
“Are SQLScript and HANA SQL Turing complete?”
I took a swing at it and the following is my result (a slightly edited version of my original answer).
Alright, let’s do this! (even though I don’t have an idea why it would matter at all…)
So, I’m not a computer scientist, and only a third through “The Annotated Turing” (which seems to be a great book, btw). This means what follows is the layman’s approach on this.
First off, I had to understand, what is meant by “Turing complete” and how to show that a language has that characteristic.
For that, there are a couple of Wikipedia entries and SO discussions available and I leave that for everyone to read (e.g. here and here).
One of those discussions linked to a presentation that claimed to prove that PostgreSQL-SQL with recursive common table expressions (CTS) is Turing complete. In order to prove this, the author of the presentation (here) said, that it’s enough to prove that a language can emulate another language that has already been shown to be Turing complete.
The author’s choice was a cyclic tag system with a specific rule set (rule 110) which apparently has been shown to be Turing complete.
Then the author goes on and implements this cyclic tag system with a recursive common table expression and thereby proves the claim.
So, what does that mean for SAP HANA SQL?
SAP HANA SQL/SQLScript does not support recursive common table expressions (much to the distaste of everyone who tries to handle hierarchies and does not know about SAP HANA’s special hierarchy views and functions (look there) and it also does not support recursive procedure calls.
Bummer, one might think.
Fortunately, every recursion can be expressed as an iteration (compare here), so I thought, let’s try this cyclic tag system in SQLScript.
This is the result (HANA 1, rev.122.15, on my NUC system). The SQL code is also available in my GitHub repo.
do begin declare prods VARCHAR(4) ARRAY; declare currProd, initWord, currWord VARC<span data-mce-type="bookmark" style="display: inline-block; width: 0px; overflow: hidden; line-height: 0;" class="mce_SELRES_start"></span>HAR(300); -- 300 is arbitrary and would be exceeded for more runs declare currProdNo integer = 0; declare runs, maxruns bigint = 0; initWord :='11001'; -- the starting/initial 'word' maxruns := 100; -- a limit to the number of iterations -- rule 110 is suspected to run indefinitively prods = ARRAY ('010', '000', '1111'); -- the three 'producer rules' stored in a string array currWord := :initWord; runs := 0; -- dummy table var to monitor output tmp = select :runs as RUNS, :currProd as CURRPROD, :currWord as CURRWORD from dummy; while (:runs < :maxruns) DO runs := :runs+1; currProdNo := mod(:runs,3)+1; -- pick rule no. 1,2 or 3 but never 0 -- as SQLScript arrays are 1 based currProd := :prods[:currProdNo]; if (left (:currWord, 1)='1') then -- add current producer to the 'word' currWord := :currWord || :currProd; end if; currWord := substring (:currWord, 2); -- remove leftmost character -- save current state into temp table var tmp = select RUNS, CURRPROD, CURRWORD from :tmp union all select :runs as RUNS, :currProd as CURRPROD, :currWord as CURRWORD from dummy; end while; select * from :tmp; -- output the table var end;
Running this gives the following output:
/* Statement 'do begin declare prods VARCHAR(4) ARRAY; declare currProd, initWord, currWord VARCHAR(300); declare ...' successfully executed in 7<span data-mce-type="bookmark" style="display: inline-block; width: 0px; overflow: hidden; line-height: 0;" class="mce_SELRES_start"></span>17 ms 39 µs (server processing time: 715 ms 590 µs) Fetched 101 row(s) in 2 ms 517 µs (server processing time: 0 ms 424 µs) RUNS CURRPROD CURRWORD 0 ? 11001 1 000 1001000 2 1111 0010001111 3 010 010001111 4 000 10001111 5 1111 00011111111 6 010 0011111111 7 000 011111111 8 1111 11111111 9 010 1111111010 10 000 111111010000 11 1111 111110100001111 12 010 11110100001111010 13 000 1110100001111010000 14 1111 1101000011110100001111 15 010 101000011110100001111010 16 000 01000011110100001111010000 17 1111 1000011110100001111010000 18 010 000011110100001111010000010 19 000 00011110100001111010000010 20 1111 0011110100001111010000010 21 010 011110100001111010000010 22 000 11110100001111010000010 23 1111 11101000011110100000101111 24 010 1101000011110100000101111010 25 000 101000011110100000101111010000 [...] */
That looks suspiciously like the output from the Wikipedia link (above) and does not seem to stop (except for the super-clever maxruns variable ).
With that, I’d say, it’s possible to create a cyclic tag system rule 110 with SQLScript.
The tag system is Turing complete, therefore SQLScript must be Turing complete.
Still, I have no idea why this would matter at all and that’s really all I can say/write about this.
It matters because it’s means anything you can do in any other turing complete language can be done in Hana sql script.
Thanks for the comment Matthew.
I see the general point of turing-completeness as a classification mechanism for programming languages.
It’s just that in my experience, certain problems – while possible to be solved in, say, SQLScript – just tend to be better dealt with in a different language and that the classes of problems for which SQL/SQLScript are good at are widely known.
Or the other way around: while turing-complete tells me the about whether a language is generally capable of doing something, it does not give me the slightest hint towards whether or not it would be good at doing it. That makes the whole question rather academic – as in “not relevant to the problem at hand”.
I was typing on my phone so I didn’t elaborate, but what I would have added is that while anything your can write in one Turing complete language can be written in a another (and as a corollary, any computer can emulate any other computer), it ain’t necessarily going to be pretty, efficient or easy!
In the answer I gave here: https://answers.sap.com/questions/545145/migrating-complex-abap-calculation-logic-to-cds.html?childToView=545157#answer-545157
it was kind of meta-relevant to the problem at hand, in that I used to try to elicit more information from the questioner about what he was wanting to achieve. But I think I just scared him away… 😉