NewsBlur/vendor/haystack/utils/highlighting.py

from django.utils.html import strip_tags


class Highlighter(object):
    css_class = 'highlighted'
    html_tag = 'span'
    max_length = 200
    text_block = ''
    
    def __init__(self, query, **kwargs):
        self.query = query
        
        if 'max_length' in kwargs:
            self.max_length = int(kwargs['max_length'])
        
        if 'html_tag' in kwargs:
            self.html_tag = kwargs['html_tag']
        
        if 'css_class' in kwargs:
            self.css_class = kwargs['css_class']
        
        self.query_words = set([word.lower() for word in self.query.split() if not word.startswith('-')])
    
    def highlight(self, text_block):
        self.text_block = strip_tags(text_block)
        highlight_locations = self.find_highlightable_words()
        start_offset, end_offset = self.find_window(highlight_locations)
        return self.render_html(highlight_locations, start_offset, end_offset)
    
    def find_highlightable_words(self):
        # Use a set so we only do this once per unique word.
        word_positions = {}
        
        # Pre-compute the length.
        end_offset = len(self.text_block)
        lower_text_block = self.text_block.lower()
        
        for word in self.query_words:
            if not word in word_positions:
                word_positions[word] = []
            
            start_offset = 0
            
            while start_offset < end_offset:
                next_offset = lower_text_block.find(word, start_offset, end_offset)
                
                # If we get a -1 out of find, it wasn't found. Bomb out and
                # start the next word.
                if next_offset == -1:
                    break
                
                word_positions[word].append(next_offset)
                start_offset = next_offset + len(word)
        
        return word_positions
    
    def find_window(self, highlight_locations):
        best_start = 0
        best_end = self.max_length
        
        # First, make sure we have words.
        if not len(highlight_locations):
            return (best_start, best_end)
        
        words_found = []
        
        # Next, make sure we found any words at all.
        for word, offset_list in highlight_locations.items():
            if len(offset_list):
                # Add all of the locations to the list.
                words_found.extend(offset_list)
        
        if not len(words_found):
            return (best_start, best_end)
        
        if len(words_found) == 1:
            return (words_found[0], words_found[0] + self.max_length)
        
        # Sort the list so it's in ascending order.
        words_found = sorted(words_found)
        
        # We now have a denormalized list of all positions were a word was
        # found. We'll iterate through and find the densest window we can by
        # counting the number of found offsets (-1 to fit in the window).
        highest_density = 0
        
        if words_found[:-1][0] > self.max_length:
            best_start = words_found[:-1][0]
            best_end = best_start + self.max_length
        
        for count, start in enumerate(words_found[:-1]):
            current_density = 1
            
            for end in words_found[count + 1:]:
                if end - start < self.max_length:
                    current_density += 1
                else:
                    current_density = 0
                
                # Only replace if we have a bigger (not equal density) so we
                # give deference to windows earlier in the document.
                if current_density > highest_density:
                    best_start = start
                    best_end = start + self.max_length
                    highest_density = current_density
        
        return (best_start, best_end)
    
    def render_html(self, highlight_locations=None, start_offset=None, end_offset=None):
        # Start by chopping the block down to the proper window.
        text = self.text_block[start_offset:end_offset]
        
        # Invert highlight_locations to a location -> term list
        term_list = []
        
        for term, locations in highlight_locations.items():
            term_list += [(loc - start_offset, term) for loc in locations]
            
        loc_to_term = sorted(term_list)
        
        # Prepare the highlight template
        if self.css_class:
            hl_start = '<%s class="%s">' % (self.html_tag, self.css_class)
        else:
            hl_start = '<%s>' % (self.html_tag)
        
        hl_end = '</%s>' % self.html_tag
        
        # Copy the part from the start of the string to the first match,
        # and there replace the match with a highlighted version.
        highlighted_chunk = ""
        matched_so_far = 0
        prev = 0
        prev_str = ""
        
        for cur, cur_str in loc_to_term:
            # This can be in a different case than cur_str
            actual_term = text[cur:cur + len(cur_str)]
            
            # Handle incorrect highlight_locations by first checking for the term
            if actual_term.lower() == cur_str:
                if cur < prev + len(prev_str):
                    continue

                highlighted_chunk += text[prev + len(prev_str):cur] + hl_start + actual_term + hl_end
                prev = cur
                prev_str = cur_str
                
                # Keep track of how far we've copied so far, for the last step
                matched_so_far = cur + len(actual_term)
        
        # Don't forget the chunk after the last term
        highlighted_chunk += text[matched_so_far:]
        
        if start_offset > 0:
            highlighted_chunk = '...%s' % highlighted_chunk
        
        if end_offset < len(self.text_block):
            highlighted_chunk = '%s...' % highlighted_chunk
        
        return highlighted_chunk