Re: [Gretl-devel] join filter and >=
Am 10.09.2013 17:35, schrieb Allin Cottrell:
On Tue, 10 Sep 2013, Hélio Guilherme wrote:
> This is what I have in OpenSUSE 12.2 64bit, freshly compiled gretl:
> 2001:01 25 146,997 20010216
> 2001:02 26 146,001 20010316
> 2001:03 27 146,466 20010417
> 2001:04 28 144,866 20010514
> 2001:05 29 143,120 20010717
> 2001:06 30 142,501 20010815
> 2001:07 31 142,849 20011016
> 2001:08 32 141,499 20011116
> 2001:09 33 140,300 20011016
> 2001:10 34 139,307 20011127
> 2001:11 35 137,139 20011214
> 2001:12 36 136,737 20020116
Hélio and Sven, I think I see what the trouble might be, and I just
committed what I hope is a fix to CVS.
(If I'm right, the problem was that the sorting by keys in join was not
guaranteed to be stable -- that is, not to touch the original ordering
of values that compare equal -- although it's stable on the glibc
version I'm running.)
ok good -- at this stage maybe it's worth repeating that if you look at
it from a (relational) database perspective, any ordering is
coincidental, except if you do an explicit ORDER BY (in SQL jargon). My
use of "-aggr=seq:1" basically was a workaround that relied on the
ordering in the file which in this case happened to be known. The proper
solution would have been a double filtering operation here.
thanks,
sven