OCA reporting engine fork for dev and update.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

156 lines
5.3 KiB

from base64 import b64decode
import requests
from tempfile import NamedTemporaryFile
from openerp.report.report_sxw import report_sxw, rml_parse
from openerp import registry
from py3o.template import Template
_extender_functions = {}
def py3o_report_extender(report_name):
"""
A decorator to define function to extend the context sent to a template.
This will be called at the creation of the report.
The following arguments will be passed to it:
- pool: the model pool
- cr: the database cursor
- uid: the id of the user that call the renderer
- localcontext: The context that will be passed to the report engine
- context: the Odoo context
Method copied from CampToCamp report_webkit module.
:param report_name: xml id of the report
:return:
"""
def fct1(fct):
lst = _extender_functions.get(report_name)
if not lst:
lst = []
_extender_functions[report_name] = lst
lst.append(fct)
return fct
return fct1
class Py3oParser(report_sxw):
"""Custom class that use Py3o to render libroffice reports.
Code partially taken from CampToCamp."""
def __init__(self, name, table, rml=False, parser=rml_parse,
header=False, store=False, register=True):
self.localcontext = {}
super(Py3oParser, self).__init__(
name, table, rml=rml, parser=parser,
header=header, store=store, register=register
)
def create_single_pdf(self, cr, uid, ids, data, report_xml, context=None):
""" Overide this function to generate our py3o report
"""
if report_xml.report_type != 'py3o':
return super(Py3oParser, self).create_single_pdf(
cr, uid, ids, data, report_xml, context=context
)
pool = registry(cr.dbname)
model_data_ids = pool['ir.model.data'].search(
cr, uid, [
('model', '=', 'ir.actions.report.xml'),
('res_id', '=', report_xml.id),
]
)
xml_id = None
if model_data_ids:
model_data = pool['ir.model.data'].browse(
cr, uid, model_data_ids[0], context=context
)
xml_id = '%s.%s' % (model_data.module, model_data.name)
parser_instance = self.parser(cr, uid, self.name2, context=context)
parser_instance.set_context(
self.getObjects(cr, uid, ids, context),
data, ids, report_xml.report_type
)
if xml_id in _extender_functions:
for fct in _extender_functions[xml_id]:
fct(pool, cr, uid, parser_instance.localcontext, context)
template = report_xml.py3o_template_id
filetype = report_xml.py3o_fusion_filetype
# py3o.template operates on filenames so create temporary files.
with NamedTemporaryFile(suffix='.odt', prefix='py3o-template-') as \
in_temp, \
NamedTemporaryFile(suffix='.odt', prefix='py3o-report-') as \
out_temp:
in_temp.write(b64decode(template.py3o_template_data))
in_temp.flush()
template = Template(in_temp.name, out_temp.name)
print parser_instance.localcontext
template.render(parser_instance.localcontext)
out_temp.seek(0)
if filetype.human_ext != 'odt':
# Now we ask fusion server to convert our template
fusion_server_obj = pool['py3o.server']
fusion_server_id = fusion_server_obj.search(
cr, uid, [], context=context
)[0]
fusion_server = fusion_server_obj.browse(
cr, uid, fusion_server_id, context=context
)
files = {
'tmpl_file': out_temp,
}
fields = {
"targetformat": filetype.fusion_ext,
"datadict": "{}",
"image_mapping": "{}",
"skipfusion": True,
}
r = requests.post(fusion_server.url, data=fields, files=files)
chunk_size = 1024
with NamedTemporaryFile(
suffix=filetype.human_ext,
prefix='py3o-template-'
) as fd:
for chunk in r.iter_content(chunk_size):
fd.write(chunk)
fd.seek(0)
return fd.read(), filetype.human_ext
return out_temp.read(), 'odt'
def create(self, cr, uid, ids, data, context=None):
""" Override this function to handle our py3o report
"""
pool = registry(cr.dbname)
ir_action_report_obj = pool['ir.actions.report.xml']
report_xml_ids = ir_action_report_obj.search(
cr, uid, [('report_name', '=', self.name[7:])], context=context
)
if not report_xml_ids:
return super(Py3oParser, self).create(
cr, uid, ids, data, context=context
)
report_xml = ir_action_report_obj.browse(
cr, uid, report_xml_ids[0], context=context
)
result = self.create_source_pdf(
cr, uid, ids, data, report_xml, context
)
if not result:
return False, False
return result