Merge lp:~jpakkane/unity-lens-applications/lessfuzz-test into lp:unity-lens-applications

[Jussi Pakkanen]

Tighten fuzzy matching rules so less nonrelevant matches are returned.

Fuzzy matching is currently a bit loose, which turned out to be confusing for lots of users. This patch makes the search stricter by not indexing text other than the visible application name. This reduces the amount of matches users might consider non-relevant.

This patch is intended both for trunk and raring. Does it get to both from this or do I need to file this for a separate raring branch as well?

[PS Jenkins bot (ps-jenkins)]

PASSED: Continuous integration, rev:332
http://jenkins.qa.ubuntu.com/job/unity-lens-applications-ci/1/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/unity-lens-applications-raring-amd64-ci/1
    SUCCESS: http://jenkins.qa.ubuntu.com/job/unity-lens-applications-raring-armhf-ci/1
    SUCCESS: http://jenkins.qa.ubuntu.com/job/unity-lens-applications-raring-i386-ci/1

Click here to trigger a rebuild:
http://s-jenkins:8080/job/unity-lens-applications-ci/1/rebuild

review: Approve (continuous-integration)

[Mathieu Trudel-Lapierre (cyphermox)]

We can think at least EXENAME might be something the users expect to be matched against, while it does make some sense that SUMMARY could be confusing.

What did hud match against in Quantal? Can we compare and attempt to keep the same behavior, though with the somewhat stricter fuzzy-matching than it is now?

review: Needs Fixing

=== modified file 'src/unity-package-search.cc'
--- src/unity-package-search.cc	2013-03-05 10:34:21 +0000
+++ src/unity-package-search.cc	2013-04-11 09:11:22 +0000
@@ -96,22 +96,33 @@
         std::string val;
 
         val = xdoc.get_value(XAPIAN_VALUE_APPNAME);
-        if(!val.empty())
+        if(!val.empty()) {
             cdoc.addText(appnameField, val.c_str());
+        }
+        /* It turned out that fuzzy matching non-visible content
+         * is confusing to users, so let's not index these at all.
+         * In the future we might want to explore better ways
+         * of exposing this info.
         val = xdoc.get_value(XAPIAN_VALUE_SUMMARY);
-        if(!val.empty())
+        if(!val.empty()) {
             cdoc.addText(summaryField, val.c_str());
+        }
         val = xdoc.get_value(XAPIAN_VALUE_PKGNAME);
-        if(!val.empty())
+        if(!val.empty()) {
             cdoc.addText(pkgnameField, val.c_str());
+        }
         val = xdoc.get_value(XAPIAN_VALUE_EXENAME);
-        if(!val.empty())
+        if(!val.empty()) {
             cdoc.addText(exenameField, val.c_str());
+        }
+        */
         c.addDocument(cdoc);
     }
     m->index(c);
     m->getErrorValues().addStandardErrors();
-    m->getErrorValues().setSubstringMode();
+    // Substring mode resulted in too many false positives,
+    // so disable it for the time being.
+    //m->getErrorValues().setSubstringMode();
     m->getIndexWeights().setWeight(summaryField, 0.5);
 }
 
@@ -633,7 +644,8 @@
   if(g_slist_length(xap_result->results) == 0 && (!has_category && col_query_str)) {
       g_slice_free(UnityPackageSearchResult, xap_result);
       col_query_str += 3;
-      return libcolumbus_search(searcher, col_query_str);
+      UnityPackageSearchResult *res = libcolumbus_search(searcher, col_query_str);
+      return res;
   }
   return xap_result;
 }