[
https://issues.apache.org/jira/browse/TRAFODION-3042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16451392#comment-16451392
]
ASF GitHub Bot commented on TRAFODION-3042:
-------------------------------------------
Github user zellerh commented on a diff in the pull request:
https://github.com/apache/trafodion/pull/1539#discussion_r183907638
--- Diff: core/sql/exp/exp_function.cpp ---
@@ -5937,45 +5938,22 @@ void ExFunctionRandomNum::initSeed(char *op_data[])
// Pick an initial seed. According to the reference given below
// (in the eval function), all initial seeds between 1 and
- // 2147483646 are equally valid. So, we just need to pick one
+ // 2147483647 are equally valid. So, we just need to pick one
// in this range. Do this based on a timestamp.
+ struct timespec seedTime;
- // Use ex_function_current to get timestamp.
- //
- char currBuff[32];
- char *opData[1];
- opData[0] = currBuff;
- ex_function_current currentFun;
- currentFun.eval(&opData[0], 0, 0);
+ clock_gettime(CLOCK_REALTIME, &seedTime);
- // Extract year, month, etc.
- //
- char *p = currBuff;
- short year = *((short*) p);
- p += sizeof(short);
- char month = *p++;
- char day = *p++;
- char hour = *p++;
- char minute = *p++;
- char second = *p++;
- Lng32 fraction = *((Lng32*) p);
-
- // Local variables year, ..., fraction are now initialized.
- // From the values of these variables, generate a seed in the
- // desired range.
-
- Lng32 x = year * month * day;
- if (hour) x *= hour;
- p = (char*) &x;
-
- assert(sizeof(Lng32)==4);
-
- p[0] |= (second<<1);
- p[1] |= (minute<<1);
- p[2] |= (minute<<2);
- p[3] |= second;
-
- seed_ = x + fraction;
+ seed_ = (Int32) (seedTime.tv_sec % 2147483648);
+ seed_ ^= (Int32) (seedTime.tv_nsec % 2147483648L);
+
+ // Go through one step of a linear congruential random generator.
+ // (https://en.wikipedia.org/wiki/Linear_congruential_generator).
+ // This is to avoid seed values that are close to each other when
+ // we call this method again within a short time. The eval() method
+ // below doesn't handle seed values that are close to each other
+ // very well.
+ seed_ = (((Int64) seed_) * 1664525L + 1664525L) % 2147483648;
--- End diff --
Thanks, yes, good catch! Looks like a copy & paste error.
> RAND() function is not always random
> ------------------------------------
>
> Key: TRAFODION-3042
> URL: https://issues.apache.org/jira/browse/TRAFODION-3042
> Project: Apache Trafodion
> Issue Type: Bug
> Components: sql-exe
> Affects Versions: 2.2.0
> Reporter: Hans Zeller
> Assignee: Hans Zeller
> Priority: Major
>
> When trying the RAND() function, I found that it does not always behave as I
> would expect. I tried this example:
> {noformat}
> >>select rand(), rand(), rand() from dual;
> (EXPR) (EXPR) (EXPR)
> ---------- ---------- ----------
> 78242471 78511383 78612225
> --- 1 row(s) selected.
> >>select rand(), rand(), rand() from dual;
> (EXPR) (EXPR) (EXPR)
> ---------- ---------- ----------
> 1589208276 1589460381 1589561223
> --- 1 row(s) selected.
> >>select rand(), rand(), rand() from dual;
> (EXPR) (EXPR) (EXPR)
> ---------- ---------- ----------
> 134371215 134724162 134858618
> --- 1 row(s) selected.
> >>
> {noformat}
> So, it seems that if we call rand() within a very short time, the value will
> be non-random.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
