postgres: int8/bigint reads return a string but the codec declares number (should be bigint for both output type and runtime value)

[SevInf]

Summary

Reading an int8 / bigint column returns a string at runtime, while the codec's declared output type is number. This is a silent type-soundness bug: every consumer sees number in TypeScript but receives a string value.

Evidence

• PgInt8Codec (packages/3-targets/3-targets/postgres/src/core/codecs.ts) is declared with wire/input type number and decode(wire: number): Promise<number> as a bare pass-through (no coercion).
• CodecTypes['pg/int8@1']['output'] therefore resolves to number, and generated contracts declare bigint-backed fields as number | null.
• At runtime, node-postgres decodes int8/bigint columns as strings (to avoid >2^53 precision loss). The pass-through decode returns that string verbatim.

Result: create({ big: 10000000000 }) then read yields "10000000000" (string), not 10000000000 (number).

Reproduction

Round-trip any BigInt field and compare the read-back value: it is a string, though typed number.

Observed in the test-porting integration tests ported-datatypes.test.ts (entries #68, #69, #78, #79, #80).

Correct fix (decided)

int8/bigint should be modelled as JS bigint for BOTH the declared codec output type AND the runtime value:

• PgInt8Codec.decode should return a JS bigint (parsing the node-postgres string), and its declared input/output types should be bigint (not number).
• This preserves full 64-bit precision and makes the declared type match the runtime value.

Affected tests

test/integration/test/sql-orm-client/ported-datatypes.test.ts — #68, #69 (round-trip) and #78, #79, #80 (int8 filters). They are marked it.fails referencing this issue; once fixed, update their assertions to bigint values and remove the .fails markers.

Provenance

Found while porting upstream data-type behaviours into prisma-next (tests-ports.md).