On 6/10/07, Andrey Semashev <andysem@mail.ru> wrote:
Hello Matias,
Sunday, June 10, 2007, 1:04:47 AM, you wrote:
I failed to make use of B.MI ordered or hashed indices because of the constraints they apply to the CompatibleKey, CompatibleCompare, CompatibleHash and CompatiblePred objects I would have to provide.
If given two strings s1 and s2 with or without wildcards you can define an order between them, something like:
? * ?? ?* .. .. a a? a* .. ..
You can use ordered_index if your application can work with O(log(n)) lookup.
I don't see how could I do that since for ordering the elements in the container I need strict weak ordering, and for element lookup by a full string I need a more relaxed ordering. For example:
ax a? a*
It was a long shot, I was more interested in you trying out bimap :) Anyway as you have put it, it seem that the lexicographical approach is not bothering you to much, you may want to try orered_non_unique that allows you to insert multiple keys.
With respect to the structure layout you are using, a bimap can be used:
bimap< set_of<string, WildcardOrder>, unordered_set_of<int> >
No, bimap doesn't seem to suit me since the container holds not only the wildcard and int.
You can use it anyway :) The last version of the lib includes a new feature called: relations information. If you do not need indexes over that information you can include it in bimap: bimap< set_of<string, WildcardOrder>, unordered_set_of<int>, with_info< string > > And then: bm.left.find("ab*c")->info = "info"; // or... bm.left.info_at("ab*c") = "info"; bm.right.find(3)->info = "info"; // or... bm.right.info_at(3) = "info"; Best regards Matias