Closed jamezpolley closed 6 years ago
Again (like the Mount Gambier web site) the raw data is a bit patchy, but I think I've improved it a fair bit to better infer the state and postcode where previously only a suburb name was available. So those changes should now prevent most validation errors.
The "Data too long for 'postcode' error seems to be because that particular application lists no suburb or hundred, just "FIFTH (NAP) STREET".
The GeoCoding API seems to translate that to Fifth St NYC, and so we get a five digit postcode.
I'm not sure what to do here; it seems like providing more hints in the address (maybe a default postcode or suburb, or at least "south australia") might be useful? "FIFTH (NAP) STREET port pirie" in Google Maps gets me a point on Fifth Street, Port Pirie, which is a little better.
But perhaps this suggests we should be changing the way we use the geocoder - perhaps giving it a bounding box or something like that might help us get better results?
I've further updated the scraper so it now uses the street name and hundred name information to derive the suburb name (if one hasn't been provided). It will now also omit any development application for which a valid suburb can't be determined (for example, if it has been omitted from the PDF containing the development application information).
I think this update should resolve all of the problems described in this issue.