I think this is called the "Halloween problem", which is in some respect
similar:
update emp
set salary = salary * 1.05
where salary < 20000.0;
in some of the first test releases of System R,
when the index on column salary was used in this update,
the salary of - say - 10000 was first updated to 10500,
thus the record was inserted into the index at another place,
than later the record was found again, updated again and so on.
In the end, all salaries were >= 20000.0
The columns to be updated must not be used in retrieving
the records in this case !!
AFAIK, the problem is called the Halloween problem, because it was
discovered at Halloween (probably 1975 or so ...).
Kind regards
Bernd
Am 09.08.2013 16:12, schrieb Paul Gilmartin:
And this may have have been reported here, or it may be urban legend:
A shop set a task scheduled, at 0159 on a certain Sunday in Fall, to
reset the localtime to 0059. Employees were astonished to come to work
at 0730 on Monday and find the system reporting the time as 0130 on
Sunday. (If the coding is correctly designed, there is no such thing
as a semiannual "time change".)
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
