##// END OF EJS Templates
fixed @repo into :repo for docs...
fixed @repo into :repo for docs fixed routes descriptions

File last commit:

r604:5cc96df7 default
r604:5cc96df7 default
Show More
__init__.py
142 lines | 4.9 KiB | text/x-python | PythonLexer
from os.path import dirname as dn, join as jn
from rhodecode.config.environment import load_environment
from rhodecode.model.hg_model import HgModel
from shutil import rmtree
from webhelpers.html.builder import escape
from vcs.utils.lazy import LazyProperty
from whoosh.analysis import RegexTokenizer, LowercaseFilter, StopFilter
from whoosh.fields import TEXT, ID, STORED, Schema, FieldType
from whoosh.index import create_in, open_dir
from whoosh.formats import Characters
from whoosh.highlight import highlight, SimpleFragmenter, HtmlFormatter
import os
import sys
import traceback
#to get the rhodecode import
sys.path.append(dn(dn(dn(os.path.realpath(__file__)))))
#LOCATION WE KEEP THE INDEX
IDX_LOCATION = jn(dn(dn(dn(dn(os.path.abspath(__file__))))), 'data', 'index')
#EXTENSIONS WE WANT TO INDEX CONTENT OFF
INDEX_EXTENSIONS = ['action', 'adp', 'ashx', 'asmx', 'aspx', 'asx', 'axd', 'c',
'cfg', 'cfm', 'cpp', 'cs', 'css', 'diff', 'do', 'el', 'erl',
'h', 'htm', 'html', 'ini', 'java', 'js', 'jsp', 'jspx', 'lisp',
'lua', 'm', 'mako', 'ml', 'pas', 'patch', 'php', 'php3',
'php4', 'phtml', 'pm', 'py', 'rb', 'rst', 's', 'sh', 'sql',
'tpl', 'txt', 'vim', 'wss', 'xhtml', 'xml', 'xsl', 'xslt',
'yaws']
#CUSTOM ANALYZER wordsplit + lowercase filter
ANALYZER = RegexTokenizer(expression=r"\w+") | LowercaseFilter()
#INDEX SCHEMA DEFINITION
SCHEMA = Schema(owner=TEXT(),
repository=TEXT(stored=True),
path=TEXT(stored=True),
content=FieldType(format=Characters(ANALYZER),
scorable=True, stored=True),
modtime=STORED(), extension=TEXT(stored=True))
IDX_NAME = 'HG_INDEX'
FORMATTER = HtmlFormatter('span', between='\n<span class="break">...</span>\n')
FRAGMENTER = SimpleFragmenter(200)
class ResultWrapper(object):
def __init__(self, search_type, searcher, matcher, highlight_items):
self.search_type = search_type
self.searcher = searcher
self.matcher = matcher
self.highlight_items = highlight_items
self.fragment_size = 200 / 2
@LazyProperty
def doc_ids(self):
docs_id = []
while self.matcher.is_active():
docnum = self.matcher.id()
chunks = [offsets for offsets in self.get_chunks()]
docs_id.append([docnum, chunks])
self.matcher.next()
return docs_id
def __str__(self):
return '<%s at %s>' % (self.__class__.__name__, len(self.doc_ids))
def __repr__(self):
return self.__str__()
def __len__(self):
return len(self.doc_ids)
def __iter__(self):
"""
Allows Iteration over results,and lazy generate content
*Requires* implementation of ``__getitem__`` method.
"""
for docid in self.doc_ids:
yield self.get_full_content(docid)
def __getslice__(self, i, j):
"""
Slicing of resultWrapper
"""
slice = []
for docid in self.doc_ids[i:j]:
slice.append(self.get_full_content(docid))
return slice
def get_full_content(self, docid):
res = self.searcher.stored_fields(docid[0])
f_path = res['path'][res['path'].find(res['repository']) \
+ len(res['repository']):].lstrip('/')
content_short = self.get_short_content(res, docid[1])
res.update({'content_short':content_short,
'content_short_hl':self.highlight(content_short),
'f_path':f_path})
return res
def get_short_content(self, res, chunks):
return ''.join([res['content'][chunk[0]:chunk[1]] for chunk in chunks])
def get_chunks(self):
"""
Smart function that implements chunking the content
but not overlap chunks so it doesn't highlight the same
close occurrences twice.
:param matcher:
:param size:
"""
memory = [(0, 0)]
for span in self.matcher.spans():
start = span.startchar or 0
end = span.endchar or 0
start_offseted = max(0, start - self.fragment_size)
end_offseted = end + self.fragment_size
if start_offseted < memory[-1][1]:
start_offseted = memory[-1][1]
memory.append((start_offseted, end_offseted,))
yield (start_offseted, end_offseted,)
def highlight(self, content, top=5):
if self.search_type != 'content':
return ''
hl = highlight(escape(content),
self.highlight_items,
analyzer=ANALYZER,
fragmenter=FRAGMENTER,
formatter=FORMATTER,
top=top)
return hl