Commit ec783b49 authored by Jérome Perrin's avatar Jérome Perrin

fix indentation


git-svn-id: https://svn.erp5.org/repos/public/erp5/trunk@3674 20353a03-c40f-0410-a6d1-a30d3c3de9de
parent 3679cdcf
......@@ -43,10 +43,10 @@ import types, popen2, os
from tempfile import mktemp
try:
from webdav.Lockable import ResourceLockedError
SUPPORTS_WEBDAV_LOCKS = 1
from webdav.Lockable import ResourceLockedError
SUPPORTS_WEBDAV_LOCKS = 1
except ImportError:
SUPPORTS_WEBDAV_LOCKS = 0
SUPPORTS_WEBDAV_LOCKS = 0
# FIXME: Programs linked against mandriva libgcj v 3.4.0 ave a strange
# issue that make them impossible to popen within zope.
......@@ -55,459 +55,460 @@ except ImportError:
PDFTK_EXECUTABLE = "pdftk-emulation"
class PDFTk :
"""
A class to wrapp calls to pdftk executable, found at
http://www.accesspdf.com/pdftk/
"""
def catPages(self, pdfFile, cat_option) :
""" limit to a specific range of pages, like pdftk's cat option"""
return self._getOutput(
PDFTK_EXECUTABLE+
" - cat %s output - "%cat_option, pdfFile)
def dumpDataFields(self, pdfFile) :
""" returns the output of pdftk dump_data_fields as dict """
return self._parseDumpDataFields(self.dumpDataFieldsTxt(pdfFile))
def fillFormWithDict(self, pdfFile, values) :
""" fill the form with values in """
return self.fillFormWithFDF(pdfFile, self._createFdf(values))
def fillFormWithFDF(self, pdfFile, fdfFile) :
""" fill the form of pdfFile with the FDF data fdfFile """
pdfFormFileName = mktemp(suffix=".pdf")
fdfFormFileName = mktemp(suffix=".fdf")
if hasattr(pdfFile, "read") :
pdfFile = pdfFile.read()
tmpPdfFile=open(pdfFormFileName, "w")
tmpPdfFile.write(pdfFile)
tmpPdfFile.close()
if hasattr(fdfFile, "read") :
fdfFile = fdfFile.read()
tmpFdfFile=open(fdfFormFileName, "w")
tmpFdfFile.write(fdfFile)
tmpFdfFile.close()
out = self._getOutput(
PDFTK_EXECUTABLE+
" %s fill_form %s output - "%(
pdfFormFileName, fdfFormFileName))
os.remove(fdfFormFileName)
os.remove(pdfFormFileName)
return out
def dumpDataFieldsTxt(self, pdfFile) :
""" returns the output of pdftk dump_data_fields as text,
pdf file is either the file object or its content"""
return self._getOutput(
PDFTK_EXECUTABLE+" - dump_data_fields", pdfFile,
assert_not_empty=0)
def _parseDumpDataFields(self, data_fields_dump) :
""" parses the output of pdftk X.pdf dump_data_fields and
returns a sequence of dicts [{key = value}] """
fields = []
for txtfield in data_fields_dump.split("---") :
field = {}
for line in txtfield.splitlines() :
if line.strip() == "" :
continue
splits = line.split(":", 1)
if len(splits) == 2 :
field[splits[0]] = splits[1].strip()
if field != {} :
fields += [field]
return fields
def _getOutput(self, command, input=None, assert_not_empty=1) :
""" returns the output of command with sending input through command's
input stream (if input parameter is given) """
stdout, stdin = popen2.popen2(command)
if input:
if hasattr(input, "read") :
input = input.read()
try :
stdin.write(input)
except IOError, e:
raise IOError, str(e) + " ( make sure "\
"%s exists and is in your $PATH )"%PDFTK_EXECUTABLE
stdin.close()
ret = stdout.read()
stdout.close()
if assert_not_empty and len(ret) == 0 :
raise IOError, "Got no output from external program, make sure"\
" %s exists and is in your $PATH"%PDFTK_EXECUTABLE
return ret
def _escapeString(self, value) :
if value is None :
return ""
string = str(value)
escaped = ''
for c in string :
if (ord(c) == 0x28 or # open paren
ord(c) == 0x29 or # close paren
ord(c) == 0x5c): # backslash
escaped += '\\' + c
elif ord(c) < 32 or 126 < ord(c):
escaped += "\\%03o" % ord(c)
else:
escaped += c
return escaped
def _createFdf(self, values, pdfFormUrl=None) :
""" create an fdf document with the dict values """
fdf = "%FDF-1.2\x0d%\xe2\xe3\xcf\xd3\x0d\x0a"
fdf += "1 0 obj\x0d<< \x0d/FDF << /Fields [ "
for key, value in values.items():
fdf += "<< /T (%s) /V (%s)>> \x0d"%(
self._escapeString(key),
self._escapeString(value))
fdf += "] \x0d"
# the PDF form filename or URL, if any
if pdfFormUrl not in ("", None) :
fdf += "/F ("+self._escapeString(pdfFormUrl)+") \x0d"
fdf += ">> \x0d>> \x0dendobj\x0d";
fdf += "trailer\x0d<<\x0d/Root 1 0 R \x0d\x0d>>\x0d%%EOF\x0d\x0a"
return fdf
"""
A class to wrapp calls to pdftk executable, found at
http://www.accesspdf.com/pdftk/
"""
def catPages(self, pdfFile, cat_option) :
""" limit to a specific range of pages, like pdftk's cat option"""
return self._getOutput(
PDFTK_EXECUTABLE+
" - cat %s output - "%cat_option, pdfFile)
def dumpDataFields(self, pdfFile) :
""" returns the output of pdftk dump_data_fields as dict """
return self._parseDumpDataFields(self.dumpDataFieldsTxt(pdfFile))
def fillFormWithDict(self, pdfFile, values) :
""" fill the form with values in """
return self.fillFormWithFDF(pdfFile, self._createFdf(values))
def fillFormWithFDF(self, pdfFile, fdfFile) :
""" fill the form of pdfFile with the FDF data fdfFile """
pdfFormFileName = mktemp(suffix=".pdf")
fdfFormFileName = mktemp(suffix=".fdf")
if hasattr(pdfFile, "read") :
pdfFile = pdfFile.read()
tmpPdfFile = open(pdfFormFileName, "w")
tmpPdfFile.write(pdfFile)
tmpPdfFile.close()
if hasattr(fdfFile, "read") :
fdfFile = fdfFile.read()
tmpFdfFile = open(fdfFormFileName, "w")
tmpFdfFile.write(fdfFile)
tmpFdfFile.close()
out = self._getOutput(
PDFTK_EXECUTABLE+
" %s fill_form %s output - "%(
pdfFormFileName, fdfFormFileName))
os.remove(fdfFormFileName)
os.remove(pdfFormFileName)
return out
def dumpDataFieldsTxt(self, pdfFile) :
""" returns the output of pdftk dump_data_fields as text,
pdf file is either the file object or its content"""
return self._getOutput(
PDFTK_EXECUTABLE+" - dump_data_fields", pdfFile,
assert_not_empty=0)
def _parseDumpDataFields(self, data_fields_dump) :
""" parses the output of pdftk X.pdf dump_data_fields and
returns a sequence of dicts [{key = value}] """
fields = []
for txtfield in data_fields_dump.split("---") :
field = {}
for line in txtfield.splitlines() :
if line.strip() == "" :
continue
splits = line.split(":", 1)
if len(splits) == 2 :
field[splits[0]] = splits[1].strip()
if field != {} :
fields += [field]
return fields
def _getOutput(self, command, input=None, assert_not_empty=1) :
""" returns the output of command with sending input through command's
input stream (if input parameter is given) """
stdout, stdin = popen2.popen2(command)
if input:
if hasattr(input, "read") :
input = input.read()
try :
stdin.write(input)
except IOError, e:
raise IOError, str(e) + " ( make sure "\
"%s exists and is in your $PATH )"%PDFTK_EXECUTABLE
stdin.close()
ret = stdout.read()
stdout.close()
if assert_not_empty and len(ret) == 0 :
raise IOError, "Got no output from external program, make sure"\
" %s exists and is in your $PATH"%PDFTK_EXECUTABLE
return ret
def _escapeString(self, value) :
if value is None :
return ""
string = str(value)
escaped = ''
for c in string :
if (ord(c) == 0x28 or # open paren
ord(c) == 0x29 or # close paren
ord(c) == 0x5c): # backslash
escaped += '\\' + c
elif ord(c) < 32 or 126 < ord(c):
escaped += "\\%03o" % ord(c)
else:
escaped += c
return escaped
def _createFdf(self, values, pdfFormUrl=None) :
""" create an fdf document with the dict values """
fdf = "%FDF-1.2\x0d%\xe2\xe3\xcf\xd3\x0d\x0a"
fdf += "1 0 obj\x0d<< \x0d/FDF << /Fields [ "
for key, value in values.items():
fdf += "<< /T (%s) /V (%s)>> \x0d" % (
self._escapeString(key),
self._escapeString(value))
fdf += "] \x0d"
# the PDF form filename or URL, if any
if pdfFormUrl not in ("", None) :
fdf += "/F ("+self._escapeString(pdfFormUrl)+") \x0d"
fdf += ">> \x0d>> \x0dendobj\x0d";
fdf += "trailer\x0d<<\x0d/Root 1 0 R \x0d\x0d>>\x0d%%EOF\x0d\x0a"
return fdf
# Constructors
manage_addPDFForm = DTMLFile("dtml/PDFForm_add", globals())
def addPDFForm(self, id, title="", pdf_file=None, REQUEST=None):
""" Add a pdf form to folder. """
# add actual object
id = self._setObject(id, PDFForm(id, title, pdf_file))
""" Add a pdf form to folder. """
# add actual object
id = self._setObject(id, PDFForm(id, title, pdf_file))
# upload content
if pdf_file:
self._getOb(id).manage_upload(pdf_file)
self._getOb(id).content_type="application/pdf"
# upload content
if pdf_file:
self._getOb(id).manage_upload(pdf_file)
self._getOb(id).content_type="application/pdf"
if REQUEST :
u = REQUEST['URL1']
if REQUEST['submit'] == " Add and Edit ":
u = "%s/%s" % (u, quote(id))
REQUEST.RESPONSE.redirect(u+'/manage_main')
if REQUEST :
u = REQUEST['URL1']
if REQUEST['submit'] == " Add and Edit ":
u = "%s/%s" % (u, quote(id))
REQUEST.RESPONSE.redirect(u+'/manage_main')
class CalculatedValues :
"""
This class holds a reference to calculated values, for use in TALES,
because in PDF Form filling, there is lots of references to others cell
values (sums ...). This class will be in TALES context under the key 'cell'
It will make possible the use of TALES expressions like :
cell/a95
python: cell['a1'] + cell['a2']
"""
security = ClassSecurityInfo()
def __init__(self, values, key, not_founds) :
""" 'values' are a dict of already calculated values
'key' is the key we are evaluating
'not_founds' is the list in which we will put not found values """
self.__values = values
self.__key = key
self.__not_founds = not_founds
def __getitem__(self, attr) :
if not self.__values.has_key(attr) :
self.__not_founds.append(attr)
return 0 # We do not return None, so that cell['a1'] + cell['a2']
# doesn't complain that NoneType doesn't support + when a1 not found
return self.__values[attr]
__getattr__ = __getitem__
"""
This class holds a reference to calculated values, for use in TALES,
because in PDF Form filling, there is lots of references to others cell
values (sums ...). This class will be in TALES context under the key 'cell'
It will make possible the use of TALES expressions like :
cell/a95
python: cell['a1'] + cell['a2']
"""
security = ClassSecurityInfo()
def __init__(self, values, key, not_founds) :
""" 'values' are a dict of already calculated values
'key' is the key we are evaluating
'not_founds' is the list in which we will put not found values """
self.__values = values
self.__key = key
self.__not_founds = not_founds
def __getitem__(self, attr) :
if not self.__values.has_key(attr) :
self.__not_founds.append(attr)
return 0 # We do not return None, so that cell['a1'] + cell['a2']
# doesn't complain that NoneType doesn't support + when a1 not found
return self.__values[attr]
__getattr__ = __getitem__
allow_class(CalculatedValues)
class PDFForm(File):
"""
This class allows to fill PDF Form with TALES expressions,
using a TALES expression for each cell.
"""
meta_type = "ERP5 PDF Form"
icon = "www/PDFForm.png"
# Declarative Security
security = ClassSecurityInfo()
# Declarative properties
property_sheets = ( PropertySheet.Base
, PropertySheet.SimpleItem
)
# Constructors
constructors = (manage_addPDFForm, addPDFForm)
manage_options = ( (
{'label':'Edit Cell TALES', 'action':'manage_cells'},
{'label':'Display Cell Names', 'action':'showCellNames'},
{'label':'Test PDF generation', 'action':'generatePDF'},
{'label':'View original', 'action':'viewOriginal'},
) +
filter(lambda option:option['label'] != "View", File.manage_options)
)
pdftk = PDFTk()
def __init__ (self, id, title, pdf_file) :
# holds all the cell informations, even those not related to this form
self.all_cells = PersistentMapping()
# holds the cells related to this pdf form
self.cells = PersistentMapping()
# the page range we want to print
self.__page_range__ = ""
# the method to format values
self.__format_method__ = ""
if not pdf_file :
raise ValueError ("The pdf form file should not be empty")
# File constructor will call manage_upload, so we don't need to call it
File.__init__(self, id, title, pdf_file)
security.declareProtected('Change Images and Files', 'manage_upload')
def manage_upload(self, file=None, REQUEST=None) :
""" Zope calls this when the content of the enclosed file changes.
The 'cells' attribute is updated, but already defined cells are not
erased, they are saved in the 'all_cells' attribute so if the pdf
file is reverted, you do not loose the cells definitions.
"""
This class allows to fill PDF Form with TALES expressions,
using a TALES expression for each cell.
if not file or not hasattr(file, "read") :
raise ValueError ("The pdf form file should not be empty")
file.seek(0) # file is always valid here
values = self.pdftk.dumpDataFields(file)
for v in values :
if v["FieldType"] != "Button" :
k = v["FieldName"]
if not self.all_cells.has_key(k) :
self.cells[k] = ""
else :
self.cells[k] = self.all_cells[k]
self.all_cells.update(self.cells)
file.seek(0)
File.manage_upload(self, file, REQUEST)
if REQUEST:
message = "Saved changes."
return self.manage_main(self, REQUEST, manage_tabs_message=message)
security.declareProtected('View management screens', 'manage_cells')
manage_cells = PageTemplateFile('www/PDFForm_manageCells',
globals(), __name__='manage_cells')
security.declareProtected('View', 'manage_FTPget')
def manage_FTPget(self, REQUEST, RESPONSE) :
""" get this pdf form via webDAV/FTP, it returns an XML
representation of all the fields, then the pdf itself."""
from xml.dom.minidom import getDOMImplementation
impl = getDOMImplementation()
newdoc = impl.createDocument(None, "pdfform", None)
top_element = newdoc.documentElement
cells = newdoc.createElement('cells')
for cell in self.cells.keys() :
cell_node = newdoc.createElement('cell')
cell_node.setAttribute('name', cell)
tales = newdoc.createTextNode(self.cells[cell])
cell_node.appendChild(tales)
cells.appendChild(cell_node)
top_element.appendChild(cells)
pdf = newdoc.createTextNode(str(self.data))
top_element.appendChild(pdf)
content = newdoc.toprettyxml()
RESPONSE.setHeader('Content-Type', 'application/x-erp5-pdfform')
RESPONSE.setHeader('Content-Length', len(content))
RESPONSE.write(content)
manage_DAVget = manage_FTPget
def PUT(self, REQUEST, RESPONSE):
"""(does not) Handle HTTP PUT requests."""
RESPONSE.setStatus(501)
return RESPONSE
manage_FTPput = PUT
security.declareProtected('View', 'viewOriginal')
def viewOriginal(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" publish original pdf """
pdf = File.index_html(self, REQUEST, RESPONSE, *args, **kwargs)
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Disposition', 'inline;filename="%s.pdf"'
% (self.title_or_id()))
return pdf
security.declareProtected('View', 'showCellNames')
def showCellNames(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" generates a pdf with fields filled-in by their names,
usefull to fill in settings.
"""
meta_type = "ERP5 PDF Form"
icon = "www/PDFForm.png"
# Declarative Security
security = ClassSecurityInfo()
# Declarative properties
property_sheets = ( PropertySheet.Base
, PropertySheet.SimpleItem
)
# Constructors
constructors = (manage_addPDFForm, addPDFForm)
manage_options = (
(
{'label':'Edit Cell TALES', 'action':'manage_cells'},
{'label':'Display Cell Names', 'action':'showCellNames'},
{'label':'Test PDF generation', 'action':'generatePDF'},
{'label':'View original', 'action':'viewOriginal'},
) +
filter(lambda option:option['label'] != "View", File.manage_options)
)
pdftk = PDFTk()
def __init__ (self, id, title, pdf_file) :
# holds all the cell informations, even those not related to this form
self.all_cells = PersistentMapping()
# holds the cells related to this pdf form
self.cells = PersistentMapping()
# the page range we want to print
self.__page_range__ = ""
# the method to format values
self.__format_method__ = ""
if not pdf_file :
raise ValueError ("The pdf form file should not be empty")
# File constructor will call manage_upload, so we don't need to call it
File.__init__(self, id, title, pdf_file)
security.declareProtected('Change Images and Files', 'manage_upload')
def manage_upload(self, file=None, REQUEST=None) :
""" Zope calls this when the content of the enclosed file changes.
The 'cells' attribute is updated, but already defined cells are not
erased, they are saved in the 'all_cells' attribute so if the pdf
file is reverted, you do not loose the cells definitions.
"""
if not file or not hasattr(file, "read") :
raise ValueError ("The pdf form file should not be empty")
file.seek(0) # file is always valid here
values = self.pdftk.dumpDataFields(file)
for v in values :
if v["FieldType"] != "Button" :
k = v["FieldName"]
if not self.all_cells.has_key(k) :
self.cells[k] = ""
else :
self.cells[k] = self.all_cells[k]
self.all_cells.update(self.cells)
file.seek(0)
File.manage_upload(self, file, REQUEST)
if REQUEST:
message="Saved changes."
return self.manage_main(self, REQUEST, manage_tabs_message=message)
security.declareProtected('View management screens', 'manage_cells')
manage_cells = PageTemplateFile('www/PDFForm_manageCells',
globals(), __name__='manage_cells')
security.declareProtected('View', 'manage_FTPget')
def manage_FTPget(self, REQUEST, RESPONSE) :
""" get this pdf form via webDAV/FTP, it returns an XML
representation of all the fields, then the pdf itself."""
from xml.dom.minidom import getDOMImplementation
impl = getDOMImplementation()
newdoc = impl.createDocument(None, "pdfform", None)
top_element = newdoc.documentElement
cells = newdoc.createElement('cells')
for cell in self.cells.keys() :
cell_node = newdoc.createElement('cell')
cell_node.setAttribute('name', cell)
tales = newdoc.createTextNode(self.cells[cell])
cell_node.appendChild(tales)
cells.appendChild(cell_node)
top_element.appendChild(cells)
pdf = newdoc.createTextNode(str(self.data))
top_element.appendChild(pdf)
content = newdoc.toprettyxml()
RESPONSE.setHeader('Content-Type', 'application/x-erp5-pdfform')
RESPONSE.setHeader('Content-Length', len(content))
RESPONSE.write(content)
manage_DAVget = manage_FTPget
def PUT(self, REQUEST, RESPONSE):
"""(does not) Handle HTTP PUT requests."""
RESPONSE.setStatus(501)
return RESPONSE
manage_FTPput = PUT
security.declareProtected('View', 'viewOriginal')
def viewOriginal(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" publish original pdf """
pdf = File.index_html(self, REQUEST, RESPONSE, *args, **kwargs)
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Disposition', 'inline;filename="%s.pdf"'
values = {}
for cell in self.cells.keys() :
values[cell] = cell
pdf = self.pdftk.fillFormWithDict(str(self.data), values)
if RESPONSE :
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Length', len(pdf))
RESPONSE.setHeader('Content-Disposition',
'inline;filename="%s.template.pdf"' % (
self.title_or_id()))
return pdf
security.declareProtected('Change Images and Files', 'doEditCells')
def doEditCells(self, REQUEST):
""" This is the action to the 'Edit Cell TALES' tab. """
if SUPPORTS_WEBDAV_LOCKS and self.wl_isLocked():
raise ResourceLockedError, "File is locked via WebDAV"
for k, v in self.cells.items() :
self.setCellTALES(k, REQUEST.get(str(k), v))
self.__format_method__ = REQUEST.get("__format_method__")
self.__page_range__ = REQUEST.get("__page_range__")
message = "Saved changes."
if getattr(self, '_v_warnings', None):
message = ("<strong>Warning:</strong> <i>%s</i>"
% '<br>'.join(self._v_warnings))
return self.manage_cells(manage_tabs_message=message)
security.declareProtected('View', 'generatePDF')
def generatePDF(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" generates the PDF with form filled in """
values = self.calculateCellValues(REQUEST, *args, **kwargs)
context = {'here' : self.aq_parent, 'request' : REQUEST}
if hasattr(self, "__format_method__") \
and self.__format_method__ not in ('', None) :
compiled_tales = getEngine().compile(self.__format_method__)
format_method = getEngine().getContext(context).evaluate(compiled_tales)
# try to support both method name and method object
if not callable(format_method) :
format_method = self.restrictedTraverse(format_method)
if callable(format_method) :
for k, v in values.items() :
values[k] = format_method(v, cell_name=k)
else :
LOG("PDFForm", 0, 'format method (%r) is not callable' % format_method)
data = str(self.data)
if self.__page_range__ not in ('', None) :
compiled_tales = getEngine().compile(self.__page_range__)
page_range = getEngine().getContext(context).evaluate(compiled_tales)
if page_range :
data = self.pdftk.catPages(str(self.data), page_range)
pdf = self.pdftk.fillFormWithDict(data, values)
if RESPONSE :
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Length', len(pdf))
RESPONSE.setHeader('Content-Disposition', 'inline;filename="%s.pdf"'
% (self.title_or_id()))
return pdf
security.declareProtected('View', 'showCellNames')
def showCellNames(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" generates a pdf with fields filled-in by their names,
usefull to fill in settings.
"""
values = {}
for cell in self.cells.keys() :
values[cell] = cell
pdf = self.pdftk.fillFormWithDict(str(self.data), values)
if RESPONSE :
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Length', len(pdf))
RESPONSE.setHeader('Content-Disposition',
'inline;filename="%s.template.pdf"'%(
self.title_or_id()))
return pdf
security.declareProtected('Change Images and Files', 'doEditCells')
def doEditCells(self, REQUEST):
""" This is the action to the 'Edit Cell TALES' tab. """
if SUPPORTS_WEBDAV_LOCKS and self.wl_isLocked():
raise ResourceLockedError, "File is locked via WebDAV"
for k, v in self.cells.items() :
self.setCellTALES(k, REQUEST.get(str(k), v))
self.__format_method__ = REQUEST.get("__format_method__")
self.__page_range__ = REQUEST.get("__page_range__")
message = "Saved changes."
if getattr(self, '_v_warnings', None):
message = ("<strong>Warning:</strong> <i>%s</i>"
% '<br>'.join(self._v_warnings))
return self.manage_cells(manage_tabs_message=message)
security.declareProtected('View', 'generatePDF')
def generatePDF(self, REQUEST=None, RESPONSE=None, *args, **kwargs) :
""" generates the PDF with form filled in """
values = self.calculateCellValues(REQUEST, *args, **kwargs)
context = {'here' : self.aq_parent, 'request' : REQUEST}
if hasattr(self, "__format_method__") and self.__format_method__ not in ('', None) :
compiled_tales = getEngine().compile(self.__format_method__)
format_method = getEngine().getContext(context).evaluate(compiled_tales)
# try to support both method name and method object
if not callable(format_method) :
format_method = self.restrictedTraverse(format_method)
if callable(format_method) :
for k, v in values.items() :
values[k] = format_method(v, cell_name=k)
else :
LOG("PDFForm", 0, 'format method ("%r") is not callable' % format_method)
data = str(self.data)
if self.__page_range__ not in ('', None) :
compiled_tales = getEngine().compile(self.__page_range__)
page_range = getEngine().getContext(context).evaluate(compiled_tales)
if page_range :
data = self.pdftk.catPages(str(self.data), page_range)
pdf = self.pdftk.fillFormWithDict(data, values)
if RESPONSE :
RESPONSE.setHeader('Content-Type', 'application/pdf')
RESPONSE.setHeader('Content-Length', len(pdf))
RESPONSE.setHeader('Content-Disposition', 'inline;filename="%s.pdf"'
% (self.title_or_id()))
return pdf
index_html = generatePDF
__call__ = generatePDF
security.declareProtected('View', 'calculateCellValues')
def calculateCellValues(self, REQUEST=None, *args, **kwargs) :
""" returns a dict of cell values """
# values to be returned
values = {}
# list of values that need to be reevaluated (i.e. they depend on the
# value of a cell that was not already evaluated when evaluating them )
uncalculated_values = []
for cell_name in self.cells.keys() :
not_founds = []
value = self.evaluateCell(cell_name, REQUEST = REQUEST,
cell = SafeMapping(CalculatedValues(
values, cell_name, not_founds)))
if len(not_founds) != 0 :
uncalculated_values.append(cell_name)
else :
values[cell_name] = value
# now we iterate on the list of uncalculated values, trying
# to evaluate them again, if an iteration doesn't decrement
# the length of this list, there are some circular references
# and we cannot continue.
while 1 :
uncalculated_values_len = len(uncalculated_values)
if uncalculated_values_len == 0 :
return values
for cell_name in uncalculated_values :
not_founds = []
value = self.evaluateCell(cell_name, REQUEST = REQUEST,
cell = SafeMapping(CalculatedValues(
values, cell_name, not_founds)))
if len(not_founds) == 0 :
uncalculated_values.remove(cell_name)
values[cell_name] = value
if len(uncalculated_values) == uncalculated_values_len :
raise "Circular reference", "unable to evaluate cells " \
+ `uncalculated_values`
security.declareProtected('View', 'getCellNames')
def getCellNames(self, REQUEST=None) :
""" returns a list of cell names """
names = self.cells.keys()
names.sort()
return names
security.declareProtected('Change Images and Files', 'setCellTALES')
def setCellTALES(self, cell_name, TALES):
""" changes the TALES expression that will be used to evaluate
cell value """
if type(TALES) != types.StringType :
LOG("PDFForm", 100,
'TALES is not a string for cell "%s", it is = "%s"'
%(cell_name, `TALES`))
raise ValueError, 'TALES must be a string'
self.all_cells[str(cell_name)] = self.cells[str(cell_name)] = TALES
security.declareProtected('View', 'getCellTALES')
def getCellTALES(self, cell_name):
""" returns the TALES expression associated with this cell """
return self.cells[str(cell_name)]
security.declareProtected('View', 'evaluateCell')
def evaluateCell(self, cell_name, REQUEST=None, **kwargs):
""" evaluate the TALES expression for this cell """
cell_name = str(cell_name)
# we don't pass empty strings in TALES engine
# (and this also raises the KeyError for non existant cells)
if not self.cells[cell_name] :
return None
context = {'here' : self.aq_parent, 'request' : REQUEST}
context.update (kwargs)
compiled_tales = getEngine().compile(self.cells[cell_name])
value = getEngine().getContext(context).evaluate(compiled_tales)
return value
security.declareProtected('Change Images and Files', 'setAllCellTALES')
def setAllCellTALES(self, new_cells) :
""" set all cell values from a dict containing { name: TALES } """
for cell_name, cell_TALES in new_cells.items() :
self.setCellTALES(cell_name, cell_TALES)
security.declareProtected('View', 'getFormatMethodTALES')
def getFormatMethodTALES(self):
""" returns the TALES expression for the format method attribute """
# backward compat
if not hasattr(self, "__format_method__") :
self.__format_method__ = ""
return self.__format_method__
security.declareProtected('Change Images and Files', 'setFormatMethodTALES')
def setFormatMethodTALES(self, TALES):
""" sets TALES expression for the format method attribute """
self.__format_method__ = str(TALES)
security.declareProtected('View', 'getPageRangeTALES')
def getPageRangeTALES(self):
""" returns the TALES expression for the page range attribute """
return self.__page_range__
security.declareProtected('Change Images and Files', 'setPageRangeTALES')
def setPageRangeTALES(self, TALES):
""" sets TALES expression for the page range attribute """
self.__page_range__ = str(TALES)
return pdf
index_html = generatePDF
__call__ = generatePDF
security.declareProtected('View', 'calculateCellValues')
def calculateCellValues(self, REQUEST=None, *args, **kwargs) :
""" returns a dict of cell values """
# values to be returned
values = {}
# list of values that need to be reevaluated (i.e. they depend on the
# value of a cell that was not already evaluated when evaluating them )
uncalculated_values = []
for cell_name in self.cells.keys() :
not_founds = []
value = self.evaluateCell(cell_name, REQUEST = REQUEST,
cell = SafeMapping(CalculatedValues(
values, cell_name, not_founds)))
if len(not_founds) != 0 :
uncalculated_values.append(cell_name)
else :
values[cell_name] = value
# now we iterate on the list of uncalculated values, trying
# to evaluate them again, if an iteration doesn't decrement
# the length of this list, there are some circular references
# and we cannot continue.
while 1 :
uncalculated_values_len = len(uncalculated_values)
if uncalculated_values_len == 0 :
return values
for cell_name in uncalculated_values :
not_founds = []
value = self.evaluateCell(cell_name, REQUEST = REQUEST,
cell = SafeMapping(CalculatedValues(
values, cell_name, not_founds)))
if len(not_founds) == 0 :
uncalculated_values.remove(cell_name)
values[cell_name] = value
if len(uncalculated_values) == uncalculated_values_len :
raise ValueError, "Circular reference: unable to evaluate cells " \
+ `uncalculated_values`
security.declareProtected('View', 'getCellNames')
def getCellNames(self, REQUEST=None) :
""" returns a list of cell names """
names = self.cells.keys()
names.sort()
return names
security.declareProtected('Change Images and Files', 'setCellTALES')
def setCellTALES(self, cell_name, TALES):
""" changes the TALES expression that will be used to evaluate
cell value """
if type(TALES) != types.StringType :
LOG("PDFForm", 100,
'TALES is not a string for cell "%s", it is = "%s"'
%(cell_name, `TALES`))
raise ValueError, 'TALES must be a string'
self.all_cells[str(cell_name)] = self.cells[str(cell_name)] = TALES
security.declareProtected('View', 'getCellTALES')
def getCellTALES(self, cell_name):
""" returns the TALES expression associated with this cell """
return self.cells[str(cell_name)]
security.declareProtected('View', 'evaluateCell')
def evaluateCell(self, cell_name, REQUEST=None, **kwargs):
""" evaluate the TALES expression for this cell """
cell_name = str(cell_name)
# we don't pass empty strings in TALES engine
# (and this also raises the KeyError for non existant cells)
if not self.cells[cell_name] :
return None
context = {'here' : self.aq_parent, 'request' : REQUEST}
context.update (kwargs)
compiled_tales = getEngine().compile(self.cells[cell_name])
value = getEngine().getContext(context).evaluate(compiled_tales)
return value
security.declareProtected('Change Images and Files', 'setAllCellTALES')
def setAllCellTALES(self, new_cells) :
""" set all cell values from a dict containing { name: TALES } """
for cell_name, cell_TALES in new_cells.items() :
self.setCellTALES(cell_name, cell_TALES)
security.declareProtected('View', 'getFormatMethodTALES')
def getFormatMethodTALES(self):
""" returns the TALES expression for the format method attribute """
# backward compat
if not hasattr(self, "__format_method__") :
self.__format_method__ = ""
return self.__format_method__
security.declareProtected('Change Images and Files', 'setFormatMethodTALES')
def setFormatMethodTALES(self, TALES):
""" sets TALES expression for the format method attribute """
self.__format_method__ = str(TALES)
security.declareProtected('View', 'getPageRangeTALES')
def getPageRangeTALES(self):
""" returns the TALES expression for the page range attribute """
return self.__page_range__
security.declareProtected('Change Images and Files', 'setPageRangeTALES')
def setPageRangeTALES(self, TALES):
""" sets TALES expression for the page range attribute """
self.__page_range__ = str(TALES)
InitializeClass(PDFForm)
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment