postgres: Fix inability to create model with known int primary key

[philband]

Problem and a partial fix already described in #586

[aeneasr]

I don't think this is correct. Primary keys should not be writeable! If this was changed, it should also be reflected in the other dialects, not just psql.

[philband]

I don't think this is correct. Primary keys should not be writeable! If this was changed, it should also be reflected in the other dialects, not just psql.

I would argue that as a default, they should be auto-generated. However, there are situations where it is necessary to overwrite this behavior.

Example: I need to import some basic data from a different database that already has existing references and needs matching primary keys for consistency.

Currently, pop silently ignores the primary key set on the import - producing results that can not be intended and to the best of my knowledge, this behavior is not documented anywhere.

[aeneasr]

That might be true - however the submitted code does not behave as you probably expect which is why the test suites are failing. Primary keys should not be writable after the fact. It is true that they could be set before. However, for auto increments, this is a bad idea because the table's ai index is left unchanged, causing future duplicates and other potential issues. If you intend on importing int ai PKs, please do so using SQL instruction sets to avoid common problems of replicating writes instead of ex/importing full schemas.

[philband]

Okay, that explanation makes sense to me, thank you for clarifying. Closing.