Question

I enabled fts in sqlite for iphone and tried this and works, although very slow:

SELECT field FROM table_fts WHERE replace(replace(replace(replace(replace(lower(field), 'á','a'),'é','e'),'í','i'),'ó','o'),'ú','u') LIKE replace(replace(replace(replace(replace(lower('%string%'), 'á','a'),'é','e'),'í','i'),'ó','o'),'ú','u')

But it does not work when I want to use MATCH, it does not bring me results and there is no error

SELECT field FROM table_fts WHERE replace(replace(replace(replace(replace(lower(field), 'á','a'),'é','e'),'í','i'),'ó','o'),'ú','u') MATCH replace(replace(replace(replace(replace(lower('string'), 'á','a'),'é','e'),'í','i'),'ó','o'),'ú','u')

Is there any error or is there any other approach where I can make a tilde insensitive search?. I looked answers in the web with no success.

Answer 1

Two approaches:

First, you can violate normal-form and add columns to your table containing ASCII-only representation of your searchable fields. Furthermore, before doing a search against this secondary search column, you also remove international characters from the string that being searched for, too (that way you’re looking for ASCII-only string in a field with the ASCII-only representation).

By the way, if you want a more general purpose conversion of international characters with ASCII, you can try something like:

- (NSString *)replaceInternationalCharactersIn:(NSString *)text
{    
    NSData *stringData = [text dataUsingEncoding:NSASCIIStringEncoding allowLossyConversion:YES];

    return [[[NSString alloc] initWithData:stringData encoding:NSASCIIStringEncoding] autorelease];
}

Second, you could presumably use sqlite3_create_function() to write your own function (that presumably invokes a permutation of the above) that you can use right in your SQL statements themselves. See the SQLite documentation.

Update:

By the way, given that you’re doing FTS, the sqlite3_create_function() approach is probably not possible, but it strikes me that you could either do FTS on the field containing the ASCII-only string, or write your own tokenizer that does something along those lines.