[Fix] not to make typeorm generate alter query on geometry column when that column was not changed

[ohjeyong]

Problem

Let's assume the entity like this.

import { Point, Polygon } from 'geojson';

class SomeTable1 {
    ...
    @Column({type: 'geometry', srid: 4326, spatialFeatureType: 'Polygon'})
    my_polygon!: Polygon;
}

class SomeTable2 {
    ...
    @Column({type: 'geometry', srid: 4326, spatialFeatureType: 'Point'})
    my_place!: Point;
}

Typeorm always think the geometry field has been changed even though I did not change it.
If synchronize option is true, typeorm always run below query whenever server started.

query: ALTER TABLE "some_table2" ALTER COLUMN "my_place" TYPE geometry(Point,4326)

If synchronize option is false, when I execute typeorm migration:generate, it "always" make new migration file below even after I execute typeorm migration:run.

    public async up(queryRunner: QueryRunner): Promise<any> {
        await queryRunner.query('ALTER TABLE "some_table2" ALTER COLUMN "my_place" TYPE geometry(Point,4326)')
    }

    public async down(queryRunner: QueryRunner): Promise<any> {
        await queryRunner.query('ALTER TABLE "some_table2" ALTER COLUMN "my_place" TYPE geometry(POLYGON,4326)')
    }

When I run query below,

SELECT type 
FROM geometry_columns 
WHERE f_table_schema = 'public' 
AND f_table_name = 'some_table2 
and f_geometry_column = 'my_place';

it returns

type
POINT

It means, my_place column's geometry type is POINT on postgresql which is correct, but as we can see in the result of typeorm migration:generate, typeorm think this column is not POINT but POLYGON.
Therefore, typeorm keep trying to alter column's geometry type even though it set correctly.
This is not a critical bug, but typeorm execute useless alter query if synchronize option is true. It is pretty annoying when execute typeorm migration:generate if synchronize option is false.

Cause

Let's see the code of loadTables method in PostgresQueryRunner.ts.

if (tableColumn.type === "geometry") {
    const geometryColumnSql = `SELECT * FROM (
    SELECT
        "f_table_schema" "table_schema",
        "f_table_name" "table_name",
        "f_geometry_column" "column_name",
        "srid",
        "type"
        FROM "geometry_columns"
) AS _
    WHERE ${tablesCondition} AND "column_name" = '${tableColumn.name}'`;
        const results: ObjectLiteral[] = await this.query(geometryColumnSql);
        tableColumn.spatialFeatureType = results[0].type;
        tableColumn.srid = results[0].srid;
}

The geometryColumnSql execute the query like this.

SELECT * FROM (
    SELECT
    "f_table_schema" "table_schema",
    "f_table_name" "table_name",
    "f_geometry_column" "column_name",
    "srid",
    "type"
    FROM "geometry_columns"
    ) AS _
WHERE (blahblah) OR (blahblah) OR ... OR ("table_schema" = 'public' AND "table_name" = 'some_table1') OR ("table_schema" = 'public' AND "table_name" = 'some_table2') AND "column_name" = 'my_place'

The result is

table_schema	table_name	column_name	srid	type
public	some_table1	my_polygon	4326	POLYGON
public	some_table2	my_point	4326	POINT

This is the problem.
Let's go back to the code. results[0].type is always POLYGON. Therefore even though the column my_pointgeometry type is POINT, typeorm always think it's type is POLYGON.

Solution

First, we need to fix WHERE condition by wrapping OR condition in a parenthesis like this.

WHERE
    (
        (blahblah) OR (blahblah) OR ... OR ("table_schema" = 'public' AND "table_name" = 'some_table1') OR ("table_schema" = 'public' AND "table_name" = 'some_table2')
    )
    AND "column_name" = 'my_place'

Second, add table name on WHERE condition so that if we define same column name with different geometry type, the queryRunner can handle this properly.

WHERE
    (
        (blahblah) OR (blahblah) OR ... OR ("table_schema" = 'public' AND "table_name" = 'some_table1') OR ("table_schema" = 'public' AND "table_name" = 'some_table2')
    )
    AND "column_name" = 'my_place' AND "table_name" = 'some_table2'

Then, this query will always return only one correct column info.

table_schema	table_name	column_name	srid	type
public	some_table2	my_point	4326	POINT

Extra

The default postgis srid value is 0, which means, if you omit the Column option srid like this,

    @Column({type: 'geometry', spatialFeatureType: 'Point'})
    my_place!: Point;

postgis set this column's srid to 0.
However in ColumnMetadata.ts,

if (options.args.options.srid)
    this.srid = options.args.options.srid;

srid value zero can't pass this if statement, so that migration process think there are changes.
Related issue

By fixing the condition of if statements like this options.args.options.srid !== undefined we can solve this problem.

[pleerock]

Wow haven't seen such a detailed PR description for a while! Thanks!

[krazibit]

This fix created a new issue.
The new clause added AND "table_name" = '${table.name}'
But table.name includes the schema name as prefix making the query return no result

[thynson]

This PR has not been included in v0.3.0-rc.19 yet. Any plan to make a new rc release? @pleerock

[pleerock]

@thynson I'll do. Just need to merge master into next first.

[ohjeyong]

@pleerock
This PR cause problem when schema is set in ormconfig.
The PR(#5614) can fix this issue without any problem. Please check. Thanks.

[thynson]

I'll do. Just need to merge merge master into next first.

Pinging for that 0.2.25 released. 🚀🚀🚀

[BlakeGardner]

For anyone else looking for a solution to this issue, this was affecting me with Postgres using the following code:

  @Column({
    type: 'point',
    spatialFeatureType: 'Point',
    srid: 4326,
  })
  point: string;

For whatever reason, changing my column to this fixed the issue for me:

import { Point } from 'typeorm';


  @Column('geometry', {
    spatialFeatureType: 'Point',
    srid: 4326,
  })
  point: Point;

Not entirely sure if this is still considered a bug, but it works.