Bug 1164421 - autocorrect im to I'm not in

address review comments

apps/keyboard/js/imes/latin/latin.js

@@ -762,6 +762,52 @@
       return;
     }
 
+    //
+    // If there was an exact match for the user input in the suggestions
+    // it has already been removed. Now we need to loop through again
+    // looking for approximate matches and bump those to the start of the
+    // list. A word is an approximate match if all the letters match
+    // when we ignore case, apostrophes and hyphens. The assumption here
+    // is that the user is not expected to have to use the shift key or
+    // the alternate forms menu or the punctuation menus. If they type all
+    // the right letters, a matching word should have higher priority than
+    // a non-matching word. In particular this code ensures that "im" gets
+    // autocorrected to "I'm" instead of "in" and "id" gets autocorrected to
+    // "I'd" instead of "is". (Bug 1164421)
+    //
+    // XXX In English, apostrophes and hyphens are the only punctuation that
+    // commonly occurs within words. If other wordlists include other
+    // punctuation, we may need to add them to the regular expression below.
+    //
+    // XXX In the future, we might want to generalize this to consider
+    // accented forms when looking for approximate matches. If "jose"
+    // was autocorrecting to "hose" instead of "José", for example, then
+    // we might need to extend the definition of an approximate match here.
+    // For now, I don't have any examples of corrections that are coming
+    // out incorrectly for accented letters, though, so am keeping this
+    // simple.
+    //
+    // We loop backward through the suggestions, bumping any approximate
+    // match to the start of the list. This ensures that the highest ranked
+    // approximate match comes before lower ranked approximate matches,
+    // and all approximate matches come before non-matching suggestions.
+    //
+    var i = suggestions.length - 1;
+    var approximateMatches = 0;  // how many approximate matches we've bumped
+    var lcInput = input.toLowerCase();
+    while(i > approximateMatches) {
+      if (suggestions[i][0].toLowerCase().replace(/[-']/g, '') === lcInput) {
+        // If this suggestion was an approximate match for the users input..
+        var match = suggestions[i];  // get the match,
+        suggestions.splice(i, 1);    // remove it,
+        suggestions.unshift(match);  // and put it back at the start.
+        approximateMatches++;        // Increment this instead of i-- here.
+      }
+      else {
+        i--;       // if not an approximate match move to next suggestion
+      }
+    }
+
     // Make sure we have no more than three words
     if (suggestions.length > 3) {
       // We want to keep at least a user dictionary word here (if the heighest

apps/keyboard/test/unit/imes/latin/latin_test.js

@@ -508,6 +508,32 @@ suite('latin.js', function() {
         }).then(done, done);
       });
 
+      test('Input "im" should autocorrect to "I\'m", not "in"',
+      function(done) {
+        activateAndTestPrediction('im', 'im', [
+          ['in', 21],
+          ['I\'m', 16],
+          ['km', 9],
+          ['um', 9]
+        ]).then(function() {
+          sinon.assert.calledWith(
+            glue.sendCandidates, ['*I\'m', 'in', 'km']);
+        }).then(done, done);
+      });
+
+      test('Input "id" should autocorrect to "I\'d", not "is"',
+      function(done) {
+        activateAndTestPrediction('id', 'id', [
+          ['is', 20],
+          ['I\'d', 19],
+          ['if', 17],
+          ['ID', 16]
+        ]).then(function() {
+          sinon.assert.calledWith(
+            glue.sendCandidates, ['*I\'d', 'ID', 'is']);
+        }).then(done, done);
+      });
+
       test('Space to accept suggestion', function(done) {
         activateAndTestPrediction('jan', 'jan', [
           ['Jan'],