[Xapian-discuss] Query with version-numbers won't allow pages with Omega

[Olly Betts]

On Wed, Jan 07, 2009 at 07:25:55PM +0100, Arjen van der Meijden wrote:
> A user discovered that searching for a query like 'counter strike 1.6' 
> it is impossible to navigate to any but the first page in the search 
> results.
> 
> We mimic the original 'query'-template behaviour with omega, by 
> supplying the 'xDB', 'xP' and 'xFILTERS'-parameters back to omega once a 
> search has been done. And my guess is that omega somehow misses the fact 
> that the '1.6' in this case is actually already in the xP (which is 
> something like Zcounter.Zstrik.1.6.), with strings with a '.' in between 
> it does work (paging with "counter.strike 1 6" works just fine).

Here's the relevant bit of the code from Omega's query.cc:

    // Long, long ago we used "word1#word2#" (with trailing #) but some broken
    // old browsers (versions of MSIE) don't quote # in form GET submissions
    // and everything after the # gets interpreted as an anchor.  We now allow
    // terms like `c#' so we want to avoid '#' anyway.
    //
    // So we switched to using "word1.word2." but that doesn't work if
    // the terms contain "." themselves (e.g. Tapplication/vnd.ms-excel)
    // so now we use "word1\tword2" instead (with no trailing separator).
    //
    // However for compatibility with templates which haven't been updated and
    // bookmarked queries from Omega 0.9.6 and earlier we still support ".".
    char separator = '\t';
    unsigned int n_old_terms = count(oldp.begin(), oldp.end(), '\t') + 1;
    if (n_old_terms == 1 && oldp[oldp.size() - 1] == '.') {
        separator = '.';
        n_old_terms = count(oldp.begin(), oldp.end(), '.');
    }

So the issue is really in your code which currently mimics the behaviour of a
query template from Omega 0.9.6 or earlier and which needs updating to fix
your bug.

If you change that to use a tab instead of a . like the current templates/query
then this will work correctly (since for multiple terms we'll find at least one
tab and so (n_old_terms > 1) while for a single term xP isn't likely to end in
a '.' - it certainly never will for TermGenerator generated terms).

I'm not sure if/when we should drop this compatibility code - it seems a shame
to break bookmarked queries needlessly, especially as any non-updated templates
will still be generating the old style xP even though 0.9.7 changed the default
template back in October 2006.

Cheers,
    Olly