Today, after importing the geodata for all the postcodes in the UK I had to be able to determine which postcodes were in the closest vicinity to a fixed point.

This was a little work, the CSV wasn't in the best of condition; which had me banging my head against the wall with FasterCSV with it's stringent type ruling but eventually fixed that.

What I did was grab the first record that closely matches what I'm after and then build up the query to find it's closest neighbours.

zipcode = 'SW1'
distance = '100'

point = Geodata.find(:all, :conditions => ["postalcode = ?", zipcode], :limit => 1).first
puts point.postalcode

lng = point.longitude.to_f / 180 * Math::PI
lat = point.latitude.to_f / 180 * Math::PI

sql = "SELECT DISTINCT postalcode, place, 
  (6367.41*SQRT(2*(1-cos(RADIANS(latitude))*cos('#{lat}')*(sin(RADIANS(longitude))*sin('#{lng}')+cos(RADIANS(longitude))*cos('#{lng}'))-sin(RADIANS(latitude))* sin('#{lat}')))) 
  AS distance 
  FROM geodatas 
  WHERE (6367.41*SQRT(2*(1-cos(RADIANS(latitude))*cos('#{lat}')*(sin(RADIANS(longitude))*sin('#{lng}')+cos(RADIANS(longitude))*cos('#{lng}'))-sin(RADIANS(latitude))*sin('#{lat}'))) <= '#{distance}')
  ORDER BY distance"

results = Geodata.find_by_sql(sql)

results.each {|row| puts "postcode:#{row.postalcode} place:#{row.place} distance:#{row.distance.to_f.round}km"}

Probably still got some bugs but a stab in the right direction.