var search_suggestion = (function () { var exports = {}; function phrase_match(phrase, q) { // match "tes" to "test" and "stream test" but not "hostess" var i; q = q.toLowerCase(); phrase = phrase.toLowerCase(); if (phrase.indexOf(q) === 0) { return true; } var parts = phrase.split(' '); for (i = 0; i < parts.length; i += 1) { if (parts[i].indexOf(q) === 0) { return true; } } return false; } function stream_matches_query(stream_name, q) { return phrase_match(stream_name, q); } function highlight_person(query, person) { var hilite = typeahead_helper.highlight_query_in_phrase; return hilite(query, person.full_name) + " <" + hilite(query, person.email) + ">"; } function match_criteria(operators, criteria) { var filter = new Filter(operators); return _.any(criteria, function (cr) { if (_.has(cr, 'operand')) { return filter.has_operand(cr.operator, cr.operand); } return filter.has_operator(cr.operator); }); } function compare_by_huddle(huddle) { huddle = _.map(huddle.slice(0, -1), function (person) { person = people.get_by_email(person); if (person) { return person.user_id; } }); // Construct dict for all huddles, so we can lookup each's recency var huddles = activity.get_huddles(); var huddle_dict = {}; for (var i = 0; i < huddles.length; i += 1) { huddle_dict[huddles[i]] = i + 1; } return function (person1, person2) { var huddle1 = huddle.concat(person1.user_id).sort().join(','); var huddle2 = huddle.concat(person2.user_id).sort().join(','); // If not in the dict, assign an arbitrarily high index var score1 = huddle_dict[huddle1] || 100; var score2 = huddle_dict[huddle2] || 100; var diff = score1 - score2; if (diff !== 0) { return diff; } return typeahead_helper.compare_by_pms(person1, person2); }; } function get_stream_suggestions(last, operators) { if (!(last.operator === 'stream' || last.operator === 'search' || last.operator === '')) { return []; } var invalid = [ {operator: 'stream'}, {operator: 'is', operand: 'private'}, {operator: 'pm-with'}, ]; if (match_criteria(operators, invalid)) { return []; } var query = last.operand; var streams = stream_data.subscribed_streams(); streams = _.filter(streams, function (stream) { return stream_matches_query(stream, query); }); streams = typeahead_helper.sorter(query, streams); var objs = _.map(streams, function (stream) { var prefix = 'stream'; var highlighted_stream = typeahead_helper.highlight_with_escaping(query, stream); var description = prefix + ' ' + highlighted_stream; var term = { operator: 'stream', operand: stream, }; var search_string = Filter.unparse([term]); return {description: description, search_string: search_string}; }); return objs; } function get_group_suggestions(all_persons, last, operators) { if (last.operator !== 'pm-with' ) { return []; } var invalid = [ {operator: 'stream'}, ]; if (match_criteria(operators, invalid)) { return []; } var operand = last.operand; var negated = last.negated; // The operand has the form "part1,part2,pa", where all but the last part // are emails, and the last part is an arbitrary query. // // We only generate group suggestions when there's more than one part, and // we only use the last part to generate suggestions. var all_but_last_part; var last_part; var last_comma_index = operand.lastIndexOf(','); if (last_comma_index < 0) { return []; } // Neither all_but_last_part nor last_part include the final comma. all_but_last_part = operand.slice(0, last_comma_index); last_part = operand.slice(last_comma_index + 1); // We don't suggest a person if their email is already present in the // operand (not including the last part). var parts = all_but_last_part.split(',').concat(people.my_current_email()); var persons = _.filter(all_persons, function (person) { if (_.contains(parts, person.email)) { return false; } return (last_part === '') || people.person_matches_query(person, last_part); }); persons.sort(compare_by_huddle(parts)); // Take top 15 persons, since they're ordered by pm_recipient_count. persons = persons.slice(0, 15); var prefix = Filter.operator_to_prefix('pm-with', negated); var suggestions = _.map(persons, function (person) { var term = { operator: 'pm-with', operand: all_but_last_part + ',' + person.email, negated: negated, }; var name = highlight_person(last_part, person); var description = prefix + ' ' + all_but_last_part + ',' + name; var terms = [term]; if (negated) { terms = [{operator: 'is', operand: 'private'}, term]; } var search_string = Filter.unparse(terms); return {description: description, search_string: search_string}; }); return suggestions; } // Possible args for autocomplete_operator: pm-with, sender, from function get_person_suggestions(all_persons, last, operators, autocomplete_operator) { if (last.operator === "is" && last.operand === "private") { // Interpret 'is:private' as equivalent to 'pm-with:' last = {operator: "pm-with", operand: "", negated: false}; } var query = last.operand; // Only accept queries who match the specified operator, or no operator (search) if (!(last.operator === 'search' || last.operator === autocomplete_operator)) { return []; } // Be especially strict about the less common "from" operator. if (autocomplete_operator === 'from' && last.operator !== 'from') { return []; } var invalid; if (autocomplete_operator === 'pm-with') { invalid = [{operator: 'pm-with'}, {operator: 'stream'}]; } else { // If not pm-with, then this must either be 'sender' or 'from' invalid = [{operator: 'sender'}, {operator: 'from'}]; } if (match_criteria(operators, invalid)) { return []; } var persons = _.filter(all_persons, function (person) { return people.person_matches_query(person, query); }); persons.sort(typeahead_helper.compare_by_pms); var prefix = Filter.operator_to_prefix(autocomplete_operator, last.negated); var objs = _.map(persons, function (person) { var name = highlight_person(query, person); var description = prefix + ' ' + name; var terms = [{ operator: autocomplete_operator, operand: person.email, negated: last.negated, }]; if (autocomplete_operator === 'pm-with' && last.negated) { // In the special case of '-pm-with', add 'is:private' before it // because we assume the user still wants to narrow to PMs terms.unshift({operator: 'is', operand: 'private'}); } var search_string = Filter.unparse(terms); return {description: description, search_string: search_string}; }); return objs; } function get_default_suggestion(operators) { // Here we return the canonical suggestion for the full query that the // user typed. (The caller passes us the parsed query as "operators".) if (operators.length === 0) { return {description: '', search_string: ''}; } var search_string = Filter.unparse(operators); var description = Filter.describe(operators); description = Handlebars.Utils.escapeExpression(description); return {description: description, search_string: search_string}; } function get_topic_suggestions(last, operators) { if (last.operator === '') { return []; } var invalid = [ {operator: 'pm-with'}, {operator: 'is', operand: 'private'}, {operator: 'topic'}, ]; if (match_criteria(operators, invalid)) { return []; } var operator = Filter.canonicalize_operator(last.operator); var operand = last.operand; var negated = (operator === 'topic') && (last.negated); var stream; var guess; var filter = new Filter(operators); var suggest_operators = []; // stream:Rome -> show all Rome topics // stream:Rome topic: -> show all Rome topics // stream:Rome f -> show all Rome topics with a word starting in f // stream:Rome topic:f -> show all Rome topics with a word starting in f // stream:Rome topic:f -> show all Rome topics with a word starting in f // When narrowed to a stream: // topic: -> show all topics in current stream // foo -> show all topics in current stream with words starting with foo // If somebody explicitly types search:, then we might // not want to suggest topics, but I feel this is a very // minor issue, and Filter.parse() is currently lossy // in terms of telling us whether they provided the operator, // i.e. "foo" and "search:foo" both become [{operator: 'search', operand: 'foo'}]. switch (operator) { case 'stream': guess = ''; stream = operand; suggest_operators.push(last); break; case 'topic': case 'search': guess = operand; if (filter.has_operator('stream')) { stream = filter.operands('stream')[0]; } else { stream = narrow_state.stream(); suggest_operators.push({operator: 'stream', operand: stream}); } break; default: return []; } if (!stream) { return []; } var topics = stream_data.get_recent_topics(stream); stream = stream_data.get_name(stream); if (!topics) { return []; } // Be defensive here in case stream_data.get_recent_topics gets // super huge, but still slice off enough topics to find matches. topics = topics.slice(0, 300); topics = _.map(topics, function (topic) { return topic.subject; // "subject" is just the name of the topic }); if (guess !== '') { topics = _.filter(topics, function (topic) { return phrase_match(topic, guess); }); } topics = topics.slice(0, 10); // Just use alphabetical order. While recency and read/unreadness of // topics do matter in some contexts, you can get that from the left sidebar, // and I'm leaning toward high scannability for autocompletion. I also don't // care about case. topics.sort(); return _.map(topics, function (topic) { var topic_term = {operator: 'topic', operand: topic, negated: negated}; var operators = suggest_operators.concat([topic_term]); var search_string = Filter.unparse(operators); var description = Filter.describe(operators); return {description: description, search_string: search_string}; }); } function get_operator_subset_suggestions(operators) { // For stream:a topic:b search:c, suggest: // stream:a topic:b // stream:a if (operators.length < 1) { return []; } var i; var suggestions = []; for (i = operators.length - 1; i >= 1; i -= 1) { var subset = operators.slice(0, i); var search_string = Filter.unparse(subset); var description = Filter.describe(subset); var suggestion = {description: description, search_string: search_string}; suggestions.push(suggestion); } return suggestions; } function get_special_filter_suggestions(last, operators) { var suggestions = [ { search_string: 'is:private', description: 'private messages', invalid: [ {operator: 'is', operand: 'private'}, {operator: 'stream'}, {operator: 'pm-with'}, {operator: 'in'}, ], }, { search_string: 'is:starred', description: 'starred messages', invalid: [ {operator: 'is', operand: 'starred'}, ], }, { search_string: 'is:mentioned', description: '@-mentions', invalid: [ {operator: 'is', operand: 'mentioned'}, ], }, { search_string: 'is:alerted', description: 'alerted messages', invalid: [ {operator: 'is', operand: 'alerted'}, ], }, { search_string: 'is:unread', description: 'unread messages', invalid: [ {operator: 'is', operand: 'unread'}, ], }, ]; var last_string = Filter.unparse([last]).toLowerCase(); suggestions = _.filter(suggestions, function (s) { if (match_criteria(operators, s.invalid)) { return false; } if (last_string === '') { return true; } return (s.search_string.toLowerCase().indexOf(last_string) === 0) || (s.description.toLowerCase().indexOf(last_string) === 0); }); // Only show home if there's an empty bar if (operators.length === 0 && last_string === '') { suggestions.unshift({search_string: '', description: 'Home'}); } return suggestions; } function get_sent_by_me_suggestions(last, operators) { var last_string = Filter.unparse([last]).toLowerCase(); var sender_query = 'sender:' + people.my_current_email(); var from_query = 'from:' + people.my_current_email(); var description = 'sent by me'; var invalid = [ {operator: 'sender'}, {operator: 'from'}, ]; if (match_criteria(operators, invalid)) { return []; } if (last.operator === '' || sender_query.indexOf(last_string) === 0 || 'sender:me'.indexOf(last_string) === 0 || last_string === 'sent') { return [ { search_string: sender_query, description: description, }, ]; } else if (from_query.indexOf(last_string) === 0 || 'sender:me'.indexOf(last_string) === 0) { return [ { search_string: from_query, description: description, }, ]; } return []; } function get_containing_suggestions(last) { if (!(last.operator === 'search' || last.operator === 'has')) { return []; } var choices = ['link', 'image', 'attachment']; choices = _.filter(choices, function (choice) { return phrase_match(choice, last.operand); }); return _.map(choices, function (choice) { var op = [{operator: 'has', operand: choice, negated: last.negated}]; var search_string = Filter.unparse(op); var description = Filter.describe(op); return {description: description, search_string: search_string}; }); } function get_operator_suggestions(last) { if (!(last.operator === 'search')) { return []; } var negated = false; if (last.operand.indexOf("-") === 0) { negated = true; last.operand = last.operand.slice(1); } var choices = ['stream', 'topic', 'pm-with', 'sender', 'near', 'has', 'from']; choices = _.filter(choices, function (choice) { return phrase_match(choice, last.operand); }); return _.map(choices, function (choice) { var op = [{operator: choice, operand: '', negated: negated}]; var search_string = Filter.unparse(op); var description = Filter.describe(op); return {description: description, search_string: search_string}; }); } function attach_suggestions(result, base, suggestions) { _.each(suggestions, function (suggestion) { if (base.description.length > 0) { suggestion.search_string = base.search_string + " " + suggestion.search_string; suggestion.description = base.description + ", " + suggestion.description; } result.push(suggestion); }); } exports.get_suggestions = function (query) { // This method works in tandem with the typeahead library to generate // search suggestions. If you want to change its behavior, be sure to update // the tests. Its API is partly shaped by the typeahead library, which wants // us to give it strings only, but we also need to return our caller a hash // with information for subsequent callbacks. var result = []; var suggestion; var base; //base, default suggestion var suggestions; // Add an entry for narrow by operators. var operators = Filter.parse(query); var last = {operator: '', operand: '', negated: false}; if (operators.length > 0) { last = operators.slice(-1)[0]; } // Display the default first if (last.operator !== '') { suggestion = get_default_suggestion(operators); result = [suggestion]; } var base_operators = []; if (operators.length > 1) { base_operators = operators.slice(0, -1); } base = get_default_suggestion(base_operators); // Get all individual suggestions, and then attach_suggestions // mutates the list 'result' to add a properly-formatted suggestion suggestions = get_special_filter_suggestions(last, base_operators); attach_suggestions(result, base, suggestions); suggestions = get_sent_by_me_suggestions(last, base_operators); attach_suggestions(result, base, suggestions); suggestions = get_stream_suggestions(last, base_operators); attach_suggestions(result, base, suggestions); var persons = people.get_all_persons(); suggestions = get_person_suggestions(persons, last, base_operators, 'pm-with'); attach_suggestions(result, base, suggestions); suggestions = get_person_suggestions(persons, last, base_operators, 'sender'); attach_suggestions(result, base, suggestions); suggestions = get_person_suggestions(persons, last, base_operators, 'from'); attach_suggestions(result, base, suggestions); suggestions = get_group_suggestions(persons, last, base_operators); attach_suggestions(result, base, suggestions); suggestions = get_topic_suggestions(last, base_operators); attach_suggestions(result, base, suggestions); suggestions = get_operator_suggestions(last); attach_suggestions(result, base, suggestions); suggestions = get_containing_suggestions(last); attach_suggestions(result, base, suggestions); suggestions = get_operator_subset_suggestions(operators); result = result.concat(suggestions); _.each(result, function (sug) { var first = sug.description.charAt(0).toUpperCase(); sug.description = first + sug.description.slice(1); }); // Typeahead expects us to give it strings, not objects, so we maintain our own hash // back to our objects, and we also filter duplicates here. var lookup_table = {}; var unique_suggestions = []; _.each(result, function (obj) { if (!lookup_table[obj.search_string]) { lookup_table[obj.search_string] = obj; unique_suggestions.push(obj); } }); var strings = _.map(unique_suggestions, function (obj) { return obj.search_string; }); return { strings: strings, lookup_table: lookup_table, }; }; return exports; }()); if (typeof module !== 'undefined') { module.exports = search_suggestion; }