Numeric types with a 0 precision are converted to decimal types instead of int

I am mirroring a mysql spatial database in postgis and querying the postgis data with this driver via a web api. The mirroring process takes mssql int types (small, normal, big) and inserts them into postgresql as numeric(5,0), numeric(10,0), etc which is equivalent to integer types. The npgsql driver returns these values as decimals instead of integer types.

Steps to reproduce

1. create a table with a schema containing numeric values with no precision e.g. numeric(5,0)
2. query the table and view the returned type e.g. 5.0
3. alter the field to the equivalent integer type e.g. smallint
4. query the table and view the returned type 5

The issue

The type will be decimal and int when both should be int. This causes problems when serializing to json as the numeric(5,0) is interpreted as a decimal when sent with an empty decimal space compared to the integer serialization without a decimal.

Further technical details

Npgsql version: 4.1.5 PostgreSQL version: 11.6 Operating system: Google Cloud SQL

[roji]

I don’t really think it’s useful to discuss whether it makes sense to use numeric(x, 0) - it’s certainly possible and may be useful in some (admittedly exotic) cases, where numbers beyond the range of bigint are being represented. May I ask if that’s the reason why numeric(x, 0) is being used? I’m asking because it’s likely that the performance of operations on numeric(x, 0) be inferior to the same operations on bigint.

Apart from that, regarding how numeric(x, 0) should be mapped: simply put, numeric(x,0) is still numeric; just as PostgreSQL maintains that distinction, so does Npgsql when mapping PG types to CLR types.

[YohDeadfall]

Okay, what will happen if I slightly modify a query some one wrote? For example, will do some math instead of reading a value. The answer is that it might brake a lot of code in the client application because instead of int there will be decimal.

The driver isn’t responsible for taking decisions on how to reinterpret data, but yours as a developer of a system which just uses the driver. The responsibility of the driver is to communicate with the server, receive and send data as is. It’s the single responsibility principle.