Switched from TwistedWeb2 to TwistedWeb
- added HTTPResponse class in etwist/http.py (could be then abstracted in cubicweb/web)
- added twisted.web2 http_headers.py file in cubicweb/web to handle HTTP headers conversion between raw headers and python object
- deleted caching for base views (except for startup views). A better solution would be using weak entity tags (but they don't seem
to be implemented in twisted.web).
- added forbidden access message when browsing static local directories
- tested with TwistedWeb 8, 9 and 10
TODO:
=====
- Handle file uploading in forms.
twisted.web seems to keep very little information (only file content) about uploaded files in twisted_request.args['input_field_name']. But it doesn't seem to keep track of filenames.
Possible solutions :
- use web2 code to parse raw request content still stored and available in twisted_request.content
- find a magic function in twisted.web API to get the filenames
- More tests.
"""Views, forms, actions... for the CubicWeb web client
:organization: Logilab
:copyright: 2001-2010 LOGILAB S.A. (Paris, FRANCE), license is LGPL v2.
:contact: http://www.logilab.fr/ -- mailto:contact@logilab.fr
:license: GNU Lesser General Public License, v2.1 - http://www.gnu.org/licenses
"""
__docformat__ = "restructuredtext en"
import os
import sys
import tempfile
from rql import nodes
def need_table_view(rset, schema):
"""return True if we think that a table view is more appropriate than a
list or primary view to display the given result set
"""
rqlst = rset.syntax_tree()
if len(rqlst.children) > 1:
# UNION query, use a table
return True
selected = rqlst.children[0].selection
try:
mainvar = selected[0]
except AttributeError:
# not a variable ref, using table view is probably a good option
return True
if not (isinstance(mainvar, nodes.VariableRef) or
(isinstance(mainvar, nodes.Constant) and mainvar.uid)):
return True
for i, etype in enumerate(rset.description[0][1:]):
# etype may be None on outer join
if etype is None:
return True
# check the selected index node is a VariableRef (else we
# won't detect aggregate function
if not isinstance(selected[i+1], nodes.VariableRef):
return True
# if this is not a final entity
if not schema.eschema(etype).final:
return True
# if this is a final entity not linked to the main variable
var = selected[i+1].variable
for vref in var.references():
rel = vref.relation()
if rel is None:
continue
if mainvar.is_equivalent(rel.children[0]):
break
else:
return True
return False
# FIXME: VID_BY_MIMETYPE is unfortunately a bit too naive since
# some browsers (e.g. FF2) send a bunch of mimetypes in
# the Accept header, for instance:
# text/xml,application/xml,application/xhtml+xml,text/html;q=0.9,
# text/plain;q=0.8,image/png,*/*;q=0.5
VID_BY_MIMETYPE = {
#'text/xml': 'xml',
# XXX rss, owl...
}
def vid_from_rset(req, rset, schema):
"""given a result set, return a view id"""
if rset is None:
return 'index'
for mimetype in req.parse_accept_header('Accept'):
if mimetype in VID_BY_MIMETYPE:
return VID_BY_MIMETYPE[mimetype]
nb_rows = len(rset)
# empty resultset
if nb_rows == 0 :
return 'noresult'
# entity result set
if not schema.eschema(rset.description[0][0]).final:
if need_table_view(rset, schema):
return 'table'
if nb_rows == 1:
if req.search_state[0] == 'normal':
return 'primary'
return 'outofcontext-search'
if len(rset.column_types(0)) == 1:
return 'sameetypelist'
return 'list'
return 'table'
def linksearch_select_url(req, rset):
"""when searching an entity to create a relation, return an url to select
entities in the given rset
"""
req.add_js( ('cubicweb.ajax.js', 'cubicweb.edition.js') )
target, eid, r_type, searchedtype = req.search_state[1]
if target == 'subject':
id_fmt = '%s:%s:%%s' % (eid, r_type)
else:
id_fmt = '%%s:%s:%s' % (r_type, eid)
triplets = '-'.join(id_fmt % row[0] for row in rset.rows)
return "javascript: selectForAssociation('%s', '%s');" % (triplets, eid)
class TmpFileViewMixin(object):
binary = True
content_type = 'application/octet-stream'
cache_max_age = 60*60*2 # stay in http cache for 2 hours by default
def call(self):
self.cell_call()
def cell_call(self, row=0, col=0):
self.cw_row, self.cw_col = row, col # in case one needs it
fd, tmpfile = tempfile.mkstemp('.png')
os.close(fd)
self._generate(tmpfile)
self.w(open(tmpfile, 'rb').read())
os.unlink(tmpfile)