QGIS » QGIS Application

Hi Etienne and Nyall,

same problem here with a 3 columns primary keys (QGIS 3.2 on ubuntu 18.04).
And I think I understood the problem.

QGIS apply a unique constraint to each of the three columns, as it does for a simple ( one column) primary key.

I found a workaround to this problem, but not a real clean solution : I defined a rank() integer column (needed by QGIS) as the primary key of the table, and set a UNIQUE constraint over the three columns.

Now I can edit an new line to my table,

Yes, I can confirm what Mathieu said. The primary key is done for each column, individually, not checking all columns together.

CREATE TABLE test_two_pk
(
    field1 character varying(50) NOT NULL,
    field2 character varying(50) NOT NULL,
    CONSTRAINT sources_pkey PRIMARY KEY (field1, field2)
);

CREATE TABLE test_one_pk
(
    id serial PRIMARY KEY NOT NULL,
    field1 character varying(50) NOT NULL,
    field2 character varying(50) NOT NULL,
    CONSTRAINT sources UNIQUE (field1, field2)
);

With these two tables:
in QGIS 2.18, we can add features in both tables.
In QGIS 3.3, we can add features only in `test_one_pk`. It's impossible in `test_two_pk`.

Can we tag this ticket as a regression? As it was working in QGIS 2.

So Nyall, yes I think disabling these checkboxes would be fine. The best would be to check on the QGIS side these multiple primary keys, this is the bonus ;-)

Same issue here. Being able to disable the checkbox would be an acceptable workaround for the time being but I'd strongly prefer a more robust solution. Training other users to catch the issue and mess with checkboxes in the layer setup isn't a sustainable long term solution.