Fujii Masao
masao****@gmail*****
2013年 10月 3日 (木) 20:08:55 JST
On Thu, Oct 3, 2013 at 7:55 PM, Fujii Masao <masao****@gmail*****> wrote: > On Thu, Oct 3, 2013 at 7:34 PM, Amit Langote <amitl****@gmail*****> wrote: >> On Thu, Oct 3, 2013 at 7:28 PM, Fujii Masao <masao****@gmail*****> wrote: >>> On Thu, Oct 3, 2013 at 5:52 PM, Amit Langote <amitl****@gmail*****> wrote: >>>> On Thu, Oct 3, 2013 at 5:41 PM, Fujii Masao <masao****@gmail*****> wrote: >>>> >>>>>> So the way to make similarity function case-insensitive would be to change >>>>>> generate_bigm and not the similarity code itself. Also, the change will make >>>>>> the show_bigm function behave differently. >>>>> >>>>> Yes, generate_bigm would need to be updated to make bigm_similarity >>>>> case-sensitive. >>>>> >>>>> *From a user point of view*, bigm_similarity() and upcoming similarity >>>>> search should be case-sensitive? If yes, we should change generate_bigm, >>>>> but its change must not affect the behavior of the full-text search at all. >>>>> >>>>> Or we should just implement both case-sensitive and -insensitive >>>>> bigm_similarity() and similarity search? >>>>> >>>>> Thought? >>>> >>>> Could we say that case-sensitivity applies more to the comparison >>>> functions as in strcmp() vs strcmpi() than similarity() function? What >>>> do you think? >>> >>> It depends the whole design of the full-text (similarity) search, I think. >>> For example, pg_trgm's comparison function is case-*sensitive*. >>> Its generate_trgm function is case-*insensitive*. >>> >> >> How about make generate_bigm() accept a case sensitivity parameter? > > Yep, I had the same idea. > >> We could make some parts case-sensitive (text matching) while other >> parts case-insensitive (similarity())? > > On the second thought, I'm afraid that this means that we need to build > two kinds (case-sensitive and -insensitive) of GIN indexes when we'd > like to use both text-matching and similarity search. Anyway, my opinion is just to implement case-sensitive similarity search first (i.e., generate_bigm doesn't need to be changed). It's simple. Then we can improve the feature later if needed. Regards, -- Fujii Masao
Fork