Craig Ringer <craig@postnewspapers.com.au> writes:
> Tom Lane wrote:
>> I don't really see the problem. I assume from your reference to pg_trgm
>> that you're using trigram similarity as the prefilter for potential
>> matches
> It turns out that's no good anyway, as it appears to ignore characters
> outside the ASCII range. Rather less than useful for searching a
> database of translated strings ;-)
A quick look at the pg_trgm code suggests that it is only prepared to
deal with single-byte encodings; if you're working in UTF8, which I
suppose you'd have to be, it's dead in the water :-(. Perhaps fixing
that should be on the TODO list.
But in any case maybe the full-text-search stuff would be more useful
as a prefilter? Although honestly, for the speed we need here, I'm
not sure a prefilter is needed at all. Full text might be useful
if a LIKE-based match fails, though.
>> (And besides, speed doesn't seem like the be-all and end-all here.)
> True. It's not so much the speed as the fragility when faced with small
> changes to formatting. In addition to whitespace, some clients mangle
> punctuation with features like automatic "curly"-quoting.
Yeah. I was wondering whether encoding differences wouldn't be a huge
problem in practice, as well.
regards, tom lane