NAME

ogr_sql - The OGRDataSource supports executing commands against a datasource via the OGRDataSource::ExecuteSQL() method. While in theory any sort of command could be handled this way, in practice the mechanism is used to provide a subset of SQL SELECT capability to applications. This page discusses the generic SQL implementation implemented within OGR, and issue with driver specific SQL support.

Supported SQL syntax

OGR SQL supports the following pseudo-syntax:

SELECT <field-list> FROM <table_def>
     [LEFT JOIN <table_def> 
      ON [<table_ref>.]<key_field> = [<table_ref>.].<key_field>]*
     [WHERE <where-expr>] 
     [ORDER BY <sort specification list>]
<field-list> ::= <column-spec> [ { , <column-spec> }... ]
<column-spec> ::= <field-spec> [ <as clause> ]
                 | CAST ( <field-spec> AS <data type> ) [ <as clause> ]
<field-spec> ::= [DISTINCT] <field_ref>
                 | <field_func> ( [DISTINCT] <field-ref> )
                 | Count(*)
<as clause> ::= [ AS ] <column_name>
<data type> ::= character [ ( field_length ) ]
                | float [ ( field_length ) ]
                | numeric [ ( field_length [, field_precision ] ) ]
                | integer [ ( field_length ) ]
                | date [ ( field_length ) ]
                | time [ ( field_length ) ]
                | timestamp [ ( field_length ) ]
<field-func> ::= AVG | MAX | MIN | SUM | COUNT
<field_ref>  ::= [<table_ref>.]field_name
<sort specification list> ::=
              <sort specification> [ { <comma> <sort specification> }... ]
<sort specification> ::= <sort key> [ <ordering specification> ]
<sort key> ::=  <field_ref>
<ordering specification> ::= ASC | DESC
<table_def> ::= ['<datasource name>'.]table_name [table_alias]
<table_ref> ::= table_name | table_alias

SELECT

The SELECT statement is used to fetch layer features (analygous to table rows in an RDBMS) with the result of the query represented as a temporary layer of features. The layers of the datasource are analygous to tables in an RDBMS and feature attributes are analygous to column values. The simpliest form of OGR SQL SELECT statement looks like this:

SELECT * FROM polylayer

In this case all features are fetched from the layer named 'polylayer', and all attributes of those features are returned. This is essentially equivelent to accessing the layer directly. In this example the '*' is the list of fields to fetch from the layer, with '*' meaning that all fields should be fetched.

This slightly more sophisticated form still pulls all features from the layer but the schema will only contain the EAS_ID and PROP_VALUE attributes. Any other attributes would be discarded.

SELECT eas_id, prop_value FROM polylayer

A much more ambitious SELECT, restricting the features fetched with a WHERE clause, and sorting the results might look like:

SELECT * from polylayer WHERE prop_value > 220000.0 ORDER BY prop_value DESC

This select statement will produce a table with just one feature, with one attribute (named something like 'count_eas_id') containing the number of distinct values of the eas_id attribute.

SELECT COUNT(DISTINCT eas_id) FROM polylayer

Field List Operators

The field list is a comma separate list of the fields to be carried into the output features from the source layer. They will appear on output features in the order they appear on in the field list, so the field list may be used to re-order the fields.

A special form of the field list uses the DISTINCT keyword. This returns a list of all the distinct values of the named attribute. When the DISTINCT keyword is used, only one attribute may appear in the field list. The DISTINCT keyword may be used against any type of field. Currently the distinctness test against a string value is case insensitive in OGR SQL. The result of a SELECT with a DISTINCT keyword is a layer with one column (named the same as the field operated on), and one feature per distinct value. Geometries are discarded. The distinct values are assembled in memory, so alot of memory may be used for datasets with a large number of distinct values.

SELECT DISTINCT areacode FROM polylayer

There are also several summarization operators that may be applied to columns. When a summarization operator is applied to any field, then all fields must have summarization operators applied. The summarization operators are COUNT (a count of instances), AVG (numerical average), SUM (numericla sum), MIN (lexical or numerical minimum), and MAX (lexical or numerical maximum). This example produces a variety of sumarization information on parcel property values:

SELECT MIN(prop_value), MAX(prop_value), AVG(prop_value), SUM(prop_value), 
       COUNT(prop_value) FROM polylayer WHERE prov_name = 'Ontario'

As a special case, the COUNT() operator can be given a '*' argument instead of a field name which is a short form for count all the records though it would get the same result as giving it any of the column names. It is also possible to apply the COUNT() operator to a DISTINCT SELECT to get a count of distinct values, for instance:

SELECT COUNT(DISTINCT areacode) FROM polylayer

Field names can also be prefixed by a table name though this is only really meaningful when performing joins. It is further demonstrated in the JOIN section.

Using the field name alias

OGR SQL supports renaming the fields following the SQL92 specification by using the AS keyword according to the following example:

SELECT select *, OGR_STYLE AS 'STYLE' FROM polylayer

The field name alias can be used as the last operation in the column specification. Therefore we cannot rename the fields inside an operator, but we can rename whole column expression, like:

SELECT COUNT(areacode) AS 'count' FROM polylayer

We can optionally omit the AS keyword in the field name aliases, like:

SELECT *, OGR_STYLE 'STYLE' FROM polylayer

Changing the type of the fields

OGR SQL supports changing the type of the columns by using the SQL92 compliant CAST operator according to the following example:

SELECT *, CAST(OGR_STYLE AS character(255)) FROM rivers

Currently casting to the following target types are supported:

1.

character(field_length)

2.

float(field_length)

3.

numeric(field_length, field_precision)

4.

integer(field_length)

5.

date(field_length)

6.

time(field_length)

7.

timestamp(field_length)

Specifying the field_length and/or the field_precision is optional. Conversion to the 'integer list', 'double list' and 'string list' OGR data types are not supported, which doesn't conform to the SQL92 specification.

Field List Limitations

1.

Field arithmetic, and other binary operators are not supported, so you can't do something like:

SELECT prop_value / area FROM invoices

2.

Lots of operators are missing.

WHERE

The argument to the WHERE clause is a fairly simplistic logical expression used select records to be selected from the source layer. In addition to its use within the WHERE statement, the WHERE clause handling is also used for OGR attribute queries on regular layers.

A WHERE clause consists of a set of attribute tests. Each basic test is of the form fieldname operator value. The fieldname is any of the fields in the source layer. The operator is one of =, !=, <>, <, >, <=, >=, LIKE and ILIKE and IN.

Most of the operators are self explanitory, but is is worth nothing that != is the same as <>, the string equality is case insensitive, but the <, >, <= and >= operators are case sensitive. Both the LIKE and ILIKE operators are case insensitive.

The value argument to the LIKE operator is a pattern against which the value string is matched. In this pattern percent (%) matches any number of characters, and underscore ( _ ) matches any one character.

    String             Pattern       Matches?
    ------             -------       --------
    Alberta            ALB%          Yes
    Alberta            _lberta       Yes
    St. Alberta        _lberta       No
    St. Alberta        %lberta       Yes
    Robarts St.        %Robarts%     Yes
    12345              123%45        Yes
    123.45             12?45         No
    N0N 1P0            %N0N%         Yes
    L4C 5E2            %N0N%         No

The IN takes a list of values as it's argument and tests the attribute value for membership in the provided set.

    Value              Value Set            Matches?
    ------             -------              --------
    321                IN (456,123)         No
    'Ontario'          IN ('Ontario','BC')  Yes
    'Ont'              IN ('Ontario','BC')  No
    1                  IN (0,2,4,6)         No

In addition to the above binary operators, there are additional operators for testing if a field is null or not. These are the IS NULL and IS NOT NULL operators.

Basic field tests can be combined in more complicated predicates using logical operators include AND, OR, and the unary logical NOT. Subexpressions should be bracketed to make precidence clear. Some more complicated predicates are:

SELECT * FROM poly WHERE (prop_value >= 100000) AND (prop_value < 200000)
SELECT * FROM poly WHERE NOT (area_code LIKE 'N0N%')
SELECT * FROM poly WHERE (prop_value IS NOT NULL) AND (prop_value < 100000)

WHERE Limitations

1.

The left of any comparison operator must be a field name, and the right must be a literal value. Fields cannot currently be compared to fields.

2.

Fields must all come from the primary table (the one listed in the FROM clause, and must not have any table prefix ... they must just be the field name.

3.

No arithmetric operations are supported. You can't test 'WHERE (a+b) < 10' for instance.

4.

All string comparisons are case insensitive except for <, >, <= and >=.

ORDER BY

The ORDER BY clause is used force the returned features to be reordered into sorted order (ascending or descending) on one of the field values. Ascending (increasing) order is the default if neither the ASC or DESC keyword is provided. For example:

SELECT * FROM property WHERE class_code = 7 ORDER BY prop_value DESC
SELECT * FROM property ORDER BY prop_value 
SELECT * FROM property ORDER BY prop_value ASC
SELECT DISTINCT zip_code FROM property ORDER BY zip_code

Note that ORDER BY clauses cause two passes through the feature set. One to build an in-memory table of field values corresponded with feature ids, and a second pass to fetch the features by feature id in the sorted order. For formats which cannot efficiently randomly read features by feature id this can be a very expensive operation.

Sorting of string field values is case sensitive, not case insensitive like in most other parts of OGR SQL.

JOINs

OGR SQL supports a limited form of one to one JOIN. This allows records from a secondary table to be looked up based on a shared key between it and the primary table being queried. For instance, a table of city locations might include a nation_id column that can be used as a reference into a secondary nation table to fetch a nation name. A joined query might look like:

SELECT city.*, nation.name FROM city 
     LEFT JOIN nation ON city.nation_id = nation.id

This query would result in a table with all the fields from the city table, and an additional 'nation.name' field with the nation name pulled from the nation table by looking for the record in the nation table that has the 'id' field with the same value as the city.nation_id field.

Joins introduce a number of additional issues. One is the concept of table qualifiers on field names. For instance, referring to city.nation_id instead of just nation_id to indicate the nation_id field from the city layer. The table name qualifiers may only be used in the field list, and within the ON clause of the join.

Wildcards are also somewhat more involved. All fields from the primary table (city in this case) and the secondary table (nation in this case) may be selected using the usual * wildcard. But the fields of just one of the primary or secondary table may be selected by prefixing the asterix with the table name.

The field names in the resulting query layer will be qualified by the table name, if the table name is given as a qualifier in the field list. In addition field names will be qualified with a table name if they would conflict with earlier fields. For instance, the following select would result might result in a results set with a name, nation_id, nation.nation_id and nation.name field if the city and nation tables both have the nation_id and name fieldnames.

SELECT * FROM city LEFT JOIN nation ON city.nation_id = nation.nation_id

On the other hand if the nation table had a continent_id field, but the city table did not, then that field would not need to be qualified in the result set. However, if the selected instead looked like the following statement, all result fields would be qualified by the table name.

SELECT city.*, nation.* FROM city 
    LEFT JOIN nation ON city.nation_id = nation.nation_id

In the above examples, the nation table was found in the same datasource as the city table. However, the OGR join support includes the ability to join against a table in a different data source, potentially of a different format. This is indicated by qualifying the secondary table name with a datasource name. In this case the secondary datasource is opened using normal OGR semantics and utilized to access the secondary table untill the query result is no longer needed.

SELECT * FROM city 
  LEFT JOIN '/usr2/data/nation.dbf'.nation ON city.nation_id = nation.nation_id

While not necessarily very useful, it is also possible to introduce table aliases to simplify some SELECT statements. This can also be useful to disambiguate situations where ables of the same name are being used from different data sources. For instance, if the actual tables names were messy we might want to do something like:

SELECT c.name, n.name FROM project_615_city c
  LEFT JOIN '/usr2/data/project_615_nation.dbf'.project_615_nation n 
            ON c.nation_id = n.nation_id

It is possible to do multiple joins in a single query.

SELECT city.name, prov.name, nation.name FROM city
  LEFT JOIN province ON city.prov_id = province.id
  LEFT JOIN nation ON city.nation_id = nation.id

JOIN Limitations

1.

Joins can be very expensive operations if the secondary table is not indexed on the key field being used.

2.

Joined fields may not be used in WHERE clauses, or ORDER BY clauses at this time. The join is essentially evaluated after all primary table subsetting is complete, and after the ORDER BY pass.

3.

Joined fields may not be used as keys in later joins. So you could not use the province id in a city to lookup the province record, and then use a nation id from the province id to lookup the nation record. This is a sensible thing to want and could be implemented, but is not currently supported.

4.

Datasource names for joined tables are evaluated relative to the current processes working directory, not the path to the primary datasource.

5.

These are not true LEFT or RIGHT joins in the RDBMS sense. Whether or not a secondary record exists for the join key or not, one and only one copy of the primary record is returned in the result set. If a secondary record cannot be found, the secondary derived fields will be NULL. If more than one matching secondary field is found only the first will be used.

SPECIAL FIELDS

The OGR SQL query processor treats some of the attributes of the features as built-in special fields can be used in the SQL statements likewise the other fields. These fields can be placed in the select list, the WHERE clause and the ORDER BY clause respectively. The special field will not be included in the result by default but it may be explicitly included by adding it to the select list. When accessing the field values the special fields will take pecedence over the other fields with the same names in the data source.

FID

Normally the feature id is a special property of a feature and not treated as an attribute of the feature. In some cases it is convenient to be able to utilize the feature id in queries and result sets as a regular field. To do so use the name FID. The field wildcard expansions will not include the feature id, but it may be explicitly included using a syntax like:

SELECT FID, * FROM nation

OGR_GEOMETRY

Some of the data sources (like MapInfo tab) can handle geometries of different types within the same layer. The OGR_GEOMETRY special field represents the geometry type returned by OGRGeometry::getGeometryName() and can be used to distinguish the various types. By using this field one can select particular types of the geometries like:

SELECT * FROM nation WHERE OGR_GEOMETRY='POINT' OR OGR_GEOMETRY='POLYGON'

OGR_GEOM_WKT

The Well Known Text representation of the geometry can also be used as a special field. To select the WKT of the geometry OGR_GEOM_WKT might be included in the select list, like:

SELECT OGR_GEOM_WKT, * FROM nation

Using the OGR_GEOM_WKT and the LIKE operator in the WHERE clause we can get similar effect as using OGR_GEOMETRY:

SELECT OGR_GEOM_WKT, * FROM nation WHERE OGR_GEOM_WKT
   LIKE 'POINT%' OR OGR_GEOM_WKT LIKE 'POLYGON%'

OGR_STYLE

The OGR_STYLE special field represents the style string of the feature returned by OGRFeature::GetStyleString(). By using this field and the LIKE operator the result of the query can be filtered by the style. For example we can select the annotation features as:

SELECT * FROM nation WHERE OGR_STYLE LIKE 'LABEL%'

CREATE INDEX

Some OGR SQL drivers support creating of attribute indexes. Currently this includes the Shapefile driver. An index accelerates very simple attribute queries of the form fieldname = value, which is what is used by the JOIN capability. To create an attribute index on the nation_id field of the nation table a command like this would be used:

CREATE INDEX ON nation USING nation_id

Index Limitations

1.

Indexes are not maintained dynamically when new features are added to or removed from a layer.

2.

Very long strings (longer than 256 characters?) cannot currently be indexed.

3.

To recreate an index it is necessary to drop all indexes on a layer and then recreate all the indexes.

4.

Indexes are not used in any complex queries. Currently the only query the will accelerate is a simple 'field = value' query.

DROP INDEX

The OGR SQL DROP INDEX command can be used to drop all indexes on a particular table, or just the index for a particular column.

DROP INDEX ON nation USING nation_id
DROP INDEX ON nation

ExecuteSQL()

SQL is executed against an OGRDataSource, not against a specific layer. The call looks like this:

OGRLayer * OGRDataSource::ExecuteSQL( const char *pszSQLCommand,
                                      OGRGeometry *poSpatialFilter,
                                      const char *pszDialect );

The pszDialect argument is in theory intended to allow for support of different command languages against a provider, but for now applications should always pass an empty (not NULL) string to get the default dialect.

The poSpatialFilter argument is a geometry used to select a bounding rectangle for features to be returned in a manner similar to the OGRLayer::SetSpatialFilter() method. It may be NULL for no special spatial restriction.

The result of an ExecuteSQL() call is usually a temporary OGRLayer representing the results set from the statement. This is the case for a SELECT statement for instance. The returned temporary layer should be released with OGRDataSource::ReleaseResultsSet() method when no longer needed. Failure to release it before the datasource is destroyed may result in a crash.

Non-OGR SQL

All OGR drivers for database systems: MySQL, PostgreSQL and PostGIS (PG), Oracle (OCI), SQLite, ODBC and ESRI Personal Geodatabase (PGeo) override the OGRDataSource::ExecuteSQL() function with dedicated implementation and, by default, pass the SQL statements directly to the underlying RDBMS. In these cases the SQL syntax varies in some particulars from OGR SQL. Also, anything possible in SQL can then be accomplished for these particular databases. Only the result of SQL WHERE statements will be returned as layers.