Thanks to Peter Eisentraut, Bruce Momjian and Tom Lane. With the second CVS commit by Bruce Momjian it works fine now.
Peter Eisentraut <[EMAIL PROTECTED]> wrote: > > On second look, the patch is completely wrong anyway, because it does > > not export the variables; it depends on the user having exported them > > beforehand. I stand corrected. Tom Lane <[EMAIL PROTECTED]> wrote: > I saw that, but it seemed a non-problem to me: if the variables have not > been exported then they won't affect the sort program anyway. [...] > What I suspect is that Nicolai's environment supplies an explicit value > for LC_COLLATE, overriding both LC_ALL and LANG. If we want to be > bulletproof against that, then none of the proposals in this thread are > correct, and the correct patch is > > + LC_COLLATE=C > + export LC_COLLATE In my environment everything but LANG is set to tr_TR: $ locale LANG=tr_TR LC_CTYPE="tr_TR" LC_NUMERIC="tr_TR" LC_TIME="tr_TR" LC_COLLATE="tr_TR" LC_MONETARY="tr_TR" LC_MESSAGES="tr_TR" LC_PAPER="tr_TR" LC_NAME="tr_TR" LC_ADDRESS="tr_TR" LC_TELEPHONE="tr_TR" LC_MEASUREMENT="tr_TR" LC_IDENTIFICATION="tr_TR" LC_ALL= $ Method you suggested: LC_COLLATE=C export LC_COLLATE works fine too. Bruce Momjian <[EMAIL PROTECTED]> wrote: > > OK, new code is: > > LC_ALL=C > export LC_ALL > sort tags >/tmp/$$ && mv /tmp/$$ tags It works as expected. Thanks a lot. Best regards, Nicolai Tufar ---------------------------(end of broadcast)--------------------------- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match