Question

I wrote a simple web application that lets you mark flea market stands on a google map.
Each stand is stored in a sqlite3 database with its geolocation and other information.
This is the CREATE statement for the stands table:

CREATE TABLE stands (
id INTEGER PRIMARY_KEY,
name TEXT,
address TEXT,
u REAL,
v REAL,
);

u and v are respectively Latitude and Longitude.
Additionally I have a cities table that stores the name and geographic bounds of each city which host a stand. This is used to let users quickly navigate between cities.

CREATE TABLE cities
(name TEXT PRIMARY_KEY,
u_min REAL,
u_max REAL,
v_min REAL,
v_max REAL);

When a new stand is added, a new row is added to the cities table or if the stand is in a known city, only the bounds of the city are updated if needed.

Here are some sample stands:

592077673|Kierrätystori Rovaniemellä|Urheilukatu 1, 96100 Rovaniemi, Suomi|66.4978306921681|25.7220569153442
1321495145|Kruununhaka|Liisankatu, 00170 Helsinki, Suomi|60.1742596|24.9555782    
571688977|Viikki asukastalo LAVAn edusta|Biologinkatu 5, 00790 Helsinki, Suomi|60.2342312|25.04058
563089951|Hämeentie 156|Hämeentie 156, 00560 Helsinki, Suomi|60.2130467082539|24.9785856067459    
518892420|Joensuu - Ilosaari|Siltakatu 1, 80100 Joensuu, Finland|62.5990455742272|29.7706540507875

and cities:

    Rovaniemi|66.4978306921681|66.4978306921681|25.7220569153442|25.7220569153442
Helsinki|60.1577049447137|60.2556221042622|24.9216988767212|25.0662129772156
Järvenpää|60.4513724|60.4513724|25.0819323000001|25.0819323000001
Joensuu|62.5990455742272|62.5990653244875|29.7706540507874|29.7706540507875
Vantaa|60.2731724937748|60.2731724937748|24.9571491285278|24.9571491285278

The issue I’m having is retrieving the the number of stands per cities.
So far I’ve been using the following query:

SELECT cities.name AS city,
  cities.u_min,
  cities.u_max,
  cities.v_min,
  cities.v_max,
  count(stands.id) AS count
FROM cities
LEFT OUTER JOIN stands
  ON ((stands.u BETWEEN cities.u_min AND cities.u_max)
    AND(stands.v BETWEEN cities.v_min AND cities.v_max))
GROUP BY cities.name;

This returns:

Helsinki|60.1577049447137|60.2556221042622|24.9216988767212|25.0662129772156|9
**Joensuu|62.5990455742272|62.5990653244875|29.7706540507874|29.7706540507875|0**
Järvenpää|60.4513724|60.4513724|25.0819323000001|25.0819323000001|1
Rovaniemi|66.4978306921681|66.4978306921681|25.7220569153442|25.7220569153442|1
Vantaa|60.2731724937748|60.2731724937748|24.9571491285278|24.9571491285278|1

Which is not correct as the city named Joensuu does have 1 stand in its boundaries:

518892420|Joensuu - Ilosaari|Siltakatu 1, 80100 Joensuu, Finland|62.5990455742272|29.7706540507875

But the following query returns the expected stand:

SELECT * FROM stands where u between 62.5990455742272 and 62.5990653244875 and v between 29.7706540507874 and 29.7706540507875;

I really can’t understand what is going wrong here.
Any help would be greatly appreciated.

By the way, I imported this database to Mysql and the same thing happens so I doubt this is a sqlite3 bug.

Answer 1

I think this has to do with floating point precision error. One of the ways to deal with such problems is to introduce a small number and add it to your boundaries to make them a little wider – that eliminates precision errors.

One approach is to widen boundaries in every query directly:

SET @e = 0.0000000000001;

SELECT cities.name AS city,
  cities.u_min,
  cities.u_max,
  cities.v_min,
  cities.v_max,
  count(stands.id) AS count
FROM cities
LEFT OUTER JOIN stands
  ON ((stands.u BETWEEN cities.u_min - @e AND cities.u_max + @e)
    AND(stands.v BETWEEN cities.v_min - @e AND cities.v_max + @e))
GROUP BY cities.name;

Another approach is to store the widened boundaries in the cities table:

SET @e = 0.0000000000001;

UPDATE cities
SET cities.u_min = cities.u_min - @e,
  cities.u_max = cities.u_max + @e,
  cities.v_min = cities.v_min - @e,
  cities.v_max = cities.v_max + @e;

P.S. I am not sure if the variable syntax works in SQLite, but if doesn’t, just substitute all @e with 0.0000000000001.