BTW random() is not really a good test for ONCE - the way random works is bizzarre enough to add rather than reduce confusion. The purpose for ONCE would be, for example, to load a dictionary from a file, that takes a bit of time upfront but saves time each time it is referenced.

Agree - @Gavin Halliday any comment?

@Richard Taylor I tested your code on the lastest 7.2.x branch and it fails with the error you described. I think that error is worth another Jira.

OK, I just tried running this code on 7.0.24-1 VM (7.2.0-1 IDE/compiler), and got the same runtime error ("assert(match) failed - file: hqlttcpp.cpp, line 6093") with targets Thor, hThor, and Roxie. Switch comments on the H definition and the code runs fine. I'd have tried the 7.2 VM but it's not available for download yet.

I := RANDOM() : INDEPENDENT; //calculated once, period
G := RANDOM() : GLOBAL; //calculated once in each graph
// H := 0;
H := RANDOM() : ONCE;

ds := DATASET([{1,0,0,0,0},{2,0,0,0,0}],{UNSIGNED1 rec,UNSIGNED Ival, UNSIGNED Gval , UNSIGNED Hval, UNSIGNED Aval });

RECORDOF(ds) XF(ds L) := TRANSFORM
SELF.Ival := I;
SELF.Gval := G;
SELF.Hval := H;
SELF.Aval := RANDOM(); //calculated each time used
SELF := L;
END;

P1 := PROJECT(ds,XF(left)) : PERSIST('~RTTEST::PERSIST::IndependentVsGlobal1');
P2 := PROJECT(ds,XF(left)) : PERSIST('~RTTEST::PERSIST::IndependentVsGlobal2');

OUTPUT(P1);
OUTPUT(P2); //this gets the same Ival values as P1, but the Gval value is different than P1

OUTPUT(P1);
OUTPUT(P2); //this gets the same Ival values as P1, but the Gval value is different than P1

ONCE is only really relevant for Roxie. For other platform's it's the same as GLOBAL I think.