is detailed below.
- # 1 reference
- # 3 place and description
- # 5 date received
- # 2 and 4 are just padding
- self._td_count = 0
-
- # This is just a flag to say that we are now ready to get the reference
- # from the next bit of data
- self._get_reference = False
-
- self._data = ''
-
- # this will hold the application we are currently working on.
- self._current_application = None
-
- # The object which stores our set of planning application results
- self._results = PlanningAuthorityResults(authority_name, authority_short_name)
-
- def handle_starttag(self, tag, attrs):
- # if we see a table tag, increment the table count.
- if tag == 'table':
- self._table_count += 1
-
- # we are only interested in other tags if we are in table 3.
- if self._table_count == 3:
-
- # If we are starting a
, create a new PlanningApplication object
- # for the application currently being processed
- if tag == 'tr':
- self._current_application = PlanningApplication()
-
- # if we see a td, increment the
count.
- if tag == 'td':
- self._td_count += 1
-
- # if we are in the first
, and we see a link,
- # then it is to the info page for this applicaion.
- if tag == 'a' and self._td_count == 1:
- for key, value in attrs:
- if key == 'href':
- url_end = value
- self._current_application.info_url = urlparse.urljoin(search_url,url_end)
-
- # We now know that the next bit of data is the reference
- self._get_reference = True
-
- # href is the only attribute we are interested in.
- break
-
- def handle_endtag(self, tag):
- # There is no need to do anything unless we are in table 3.
- if self._table_count == 3:
-
- # The end
indicates that the current application is finished.
- # Now we can fetch the info_page to get the address, postcode,
- # and description.
- # If we don't have a reference, then we are in the header row,
- # which we don't want.
- # There is no point in doing this if the date is not the requested one.
-
- if tag == 'tr' and \
- self._current_application.council_reference is not None and \
- self._current_application.date_received == self._requested_date:
-
- info_page_parser = SouthOxfordshireInfoURLParser()
- info_page_parser.feed(urllib2.urlopen(self._current_application.info_url).read())
-
- self._current_application.address = info_page_parser.address
- self._current_application.postcode = getPostcodeFromText(info_page_parser.address)
- self._current_application.description = info_page_parser.description
-
- # Add the current application to the results set
- self._results.addApplication(self._current_application)
-
- # At the end of the 5th
, self._data should contain
- # the received date of the application.
- if tag == 'td' and self._td_count == 5:
- app_year, app_month, app_day = tuple(time.strptime(self._data, "%d %B %Y")[:3])
- self._current_application.date_received = datetime.date(app_year, app_month, app_day)
- self._data = ''
- self._td_count = 0
-
- def handle_data(self, data):
- # There is no need to do anything if we aren't in table 3.
- if self._table_count == 3:
- # If we are in the first
, and the get_reference flag is set,
- # then the next data is the reference.
- if self._td_count == 1 and self._get_reference:
- self._current_application.council_reference = data
-
- # The comment url can now be made, as it depends only on the reference.
- # On this site, the link to the comment page is only displayed once
- # the planning authority has decided who is handling this application
- # and has opened consultations. The link below works straight away,
- # and also works for apps for which the consultation period is over.
- # I have no idea if anything is actually done with these comments if
- # it is followed too early...
- self._current_application.comment_url = comment_url %{'reference': self._current_application.council_reference}
-
- # Set the get_reference flag back to False.
- self._get_reference = False
-
- # If we are in the 5th
, then we need to collect all the data together
- # before we can use it. This is actually processed in handle_endtag.
- if self._td_count == 5:
- self._data += data
-
- def handle_entityref( self, ref ):
- # We might have some entity_refs to clear up.
- # there is no need to bother with this if we aren't in the results table.
- if self._table_count == 3 and self._td_count == 5:
- if ref == 'nbsp':
- self._data += ' '
-
-
- def getResultsByDayMonthYear(self, day, month, year):
- """This will return an ApplicationResults object containg the
- applications for the date passed in."""
-
- today = datetime.date.today()
- self._requested_date = datetime.date(year, month, day)
- delta = today - self._requested_date
-
- # to get the correct page, we need
- # page ((days mod 7) + 1)
- page_number = delta.days/7 + 1
-
- response = urllib2.urlopen(search_url %page_number)
-
- contents = response.read()
-
- self.feed(contents)
-
- return self._results
-
-
- def getResults(self, day, month, year):
- return self.getResultsByDayMonthYear(int(day), int(month), int(year)).displayXML()
-
-class SouthOxfordshireInfoURLParser(HTMLParser.HTMLParser):
- """This parser is to get the description and address out of the info page
- for a South Oxfordshire application."""
-
- def __init__(self):
- HTMLParser.HTMLParser.__init__(self)
-
- self.address = None
- self.description = None
-
- # These two states will be set to:
- # 0 - if we haven't yet got that bit
- # 1 - if we are currently working on it
- # 2 - if we have finished
- self._address_state = 0
- self._description_state = 0
-
- # We well need to know whether or not we are in a
- self._in_td = False
-
- # This is used for collecting together date which comes in several bits.
- self._data = ''
-
- def handle_starttag(self, tag, attrs):
- # If we see the start of a
and we are still interested in some data
- # then set the td flag to true, and blank the data
- if tag == 'td' and (self._address_state < 2 or self._description_state < 2):
- self._in_td = True
- self._data = ''
-
- def handle_endtag(self, tag):
- if tag == 'td' and (self._address_state < 2 or self._description_state < 2):
- # If we are working on the description,
- # set description from _data and note that we need to work on it no more.
- if self._description_state == 1:
- self.description = self._data
- self._description_state = 2
-
-
- # If we are working on the address,
- # set address from _data and note that we need to work on it no more.
- elif self._address_state == 1:
- self.address = self._data
- self._address_state = 2
-
- # If we see data which says 'Descripton',
- # then set the description state to working.
- elif self._data.strip() == 'Description':
- self._description_state = 1
-
- # If we see data which says 'Location',
- # then set the addresss state to working.
- elif self._data.strip() == 'Location':
- self._address_state = 1
-
- # Note that we are leaving the
- self._in_td = False
-
- def handle_data(self, data):
- # if we are in a td, and we are still interested in the data for something,
- # append the current bit to self._data
- if self._in_td and (self._address_state < 2 or self._description_state < 2):
- self._data += data
-
-
-# TODO
-
-# find out what time of day this is run - does it matter that
-# we aren't being careful with daylight saving time etc.
-
-# Can we check that scraped email address really is
-# an email address?
diff --git a/CGI/Southampton.cgi b/CGI/Southampton.cgi
deleted file mode 100755
index 01188c3..0000000
--- a/CGI/Southampton.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Southampton City Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Southampton City Council"
-authority_short_name = "Southampton"
-base_url = "http://publicaccess.southampton.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Spelthorne.cgi b/CGI/Spelthorne.cgi
deleted file mode 100755
index f507ad7..0000000
--- a/CGI/Spelthorne.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Spelthorne Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Spelthorne Borough Council"
-authority_short_name = "Spelthorne"
-base_url = "http://phoenix.spelthorne.gov.uk/PublicAccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Stevenage.cgi b/CGI/Stevenage.cgi
deleted file mode 100755
index 25bf3ca..0000000
--- a/CGI/Stevenage.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Stevenage Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Stevenage Borough Council"
-authority_short_name = "Stevenage"
-base_url = "http://publicaccess.stevenage.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Sunderland.cgi b/CGI/Sunderland.cgi
deleted file mode 100755
index bf8bc15..0000000
--- a/CGI/Sunderland.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Sunderland City Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Sunderland City Council"
-authority_short_name = "Sunderland"
-base_url = "http://www.sunderland.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Test Valley.cgi b/CGI/Test Valley.cgi
deleted file mode 100755
index f782eaf..0000000
--- a/CGI/Test Valley.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Test Valley Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Test Valley Borough Council"
-authority_short_name = "Test Valley"
-base_url = "http://publicaccess.testvalley.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Tonbridge.cgi b/CGI/Tonbridge.cgi
deleted file mode 100755
index a9256f9..0000000
--- a/CGI/Tonbridge.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Tonbridge and Malling Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Tonbridge and Malling Borough Council"
-authority_short_name = "Tonbridge"
-base_url = "http://publicaccess.tmbc.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/West Norfolk.cgi b/CGI/West Norfolk.cgi
deleted file mode 100755
index 47e3dce..0000000
--- a/CGI/West Norfolk.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Kings Lynn and West Norfolk Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Kings Lynn and West Norfolk Borough Council"
-authority_short_name = "West Norfolk"
-base_url = "http://online.west-norfolk.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/Woking.cgi b/CGI/Woking.cgi
deleted file mode 100755
index 6c8f51f..0000000
--- a/CGI/Woking.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for Woking Borough Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "Woking Borough Council"
-authority_short_name = "Woking"
-base_url = "http://caps.woking.gov.uk/publicaccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml
diff --git a/CGI/York.cgi b/CGI/York.cgi
deleted file mode 100755
index ed59990..0000000
--- a/CGI/York.cgi
+++ /dev/null
@@ -1,31 +0,0 @@
-#!/usr/bin/python
-
-# This is the parser for City of York Council.
-# it is generated from the file CGITemplate
-
-import cgi
-import cgitb
-#cgitb.enable(display=0, logdir="/tmp")
-
-
-form = cgi.FieldStorage()
-day = form.getfirst('day')
-month = form.getfirst('month')
-year = form.getfirst('year')
-
-
-authority_name = "City of York Council"
-authority_short_name = "York"
-base_url = "http://planning.york.gov.uk/PublicAccess/"
-
-import PublicAccess
-
-parser = PublicAccess.PublicAccessParser(authority_name,
- authority_short_name,
- base_url)
-
-xml = parser.getResults(day, month, year)
-
-print "Content-Type: text/xml" # XML is following
-print
-print xml # print the xml