On Nov6, 2011, at 18:43 , J Smith wrote:
> I put some elog debugging lines into unaccent.c and found that sscanf
> sometimes reads the scanned line by finding only one byte for the for
> the source character rather than the two required for the complete
> UTF-8 code point. It appears that the following characters are causing
> the problem, along with the code points and such:
>
> 'Å' => 'A' | c3,85 => 41
> 'à' => 'a' | c3,a0 => 61
> 'ą' => 'a' | c4,85 => 61
> 'Ġ' => 'G' | c4,a0 => 47
> 'Ņ' => 'N' | c5,85 => 4e
> 'Š' => 'S' | c5,a0 => 53
>
> In each case, one byte was being read in the source string rather than
> two, leading to the "duplicate TO" warnings above. This later leads to
> the characters that produced the warning being ignored when unaccent
> is called and left in the output.
What's the locale of the database you're seeing this in, and which charset
does it use?
I think scanf() uses isspace() and friends, and last time I looked the
locale definitions where all pretty bogus on OSX. So maybe scanf() somehow
decides that 0xA0 is whitespace.
> I haven't been able to reproduce in a smaller example, and haven't
> been able to reproduce on a CentOS server, so at this point I'm at a
> loss as to the problem.
Have you tried to set the same locale as postgres (using setlocale()) in
your tests?
best regards,
Florian Pflug