[FIX] merge_models: psycopg2 param style + double-AND typo (marketing_card UPDATE)

[dnplkndll]

Bug

The merge_models() helper's card_campaign cross-reference UPDATE (active when marketing_card is installed) has two issues at openupgradelib/openupgrade.py:1210-1226:

query = """
    UPDATE card_card c
    SET res_id = mt.id, campaign_id = cc2.id
    FROM {model_table} mt, card_campaign cc, card_campaign cc2
    WHERE cc.res_model = {old_model} AND cc2.res_model = {new_model}    # ← bug 1
        AND mt.{ref_field} = c.res_id AND AND c.campaign_id = cc.id"""  # ← bug 2
logged_query(
    cr,
    sql.SQL(query).format(
        model_table=sql.Identifier(model_table),
        ref_field=sql.Identifier(ref_field),    # ← only 2 of 4 placeholders passed
    ),
    {"old_model": old_model, "new_model": new_model},   # ← psycopg2 query params, not sql.SQL.format kwargs
)

1. {old_model} / {new_model} are written as sql.SQL().format() placeholders, but they're passed via the psycopg2 query-params dict instead. Result at runtime: KeyError: 'old_model' from psycopg2/sql.py:254 in sql.SQL.format().
2. AND mt.{ref_field} = c.res_id AND AND c.campaign_id = cc.id — redundant AND (SQL syntax error if the first bug ever got past the formatter).

Source

Introduced by #436 ([FIX] merge_models: check table columns exist, merged 2026-02-09 by @MiquelRForgeFlow, approved by @pedrobaeza; @hbrunn requested changes on an unrelated base_automation concern that was resolved). Commit f7f5eab7 — [FIX] merge_models: correctly handle marketing_card. ~3.4 months old at time of report.

Why nobody caught it: the if is_module_installed(cr, 'marketing_card') gate means the buggy code path only fires when both marketing_card is installed and a model-merge migration runs. The OCA test seed at the time apparently didn't combine those.

Reproduction

Real-world hit: OCA/OpenUpgrade PR #5651 — when marketing_card is added to MODULES_NEW (so the seed has it installed), hr_recruitment/19.0.1.1/pre-migration.py:14's call to openupgrade.merge_models(env.cr, "hr.candidate", "hr.applicant", "candidate_id") triggers this code path and crashes with KeyError: 'old_model'.

Minimal psycopg2 repro (verified locally in the lab):

from psycopg2 import sql
buggy = '''
    WHERE cc.res_model = {old_model} AND cc2.res_model = {new_model}
        AND mt.{ref_field} = c.res_id AND AND c.campaign_id = cc.id'''
sql.SQL(buggy).format(ref_field=sql.Identifier("candidate_id"))
# → KeyError: 'old_model'

Fix

Switch the two value placeholders to psycopg2 param style (%(old_model)s / %(new_model)s, matching the dict already being passed) and drop the redundant AND:

query = """
    UPDATE card_card c
    SET res_id = mt.id, campaign_id = cc2.id
    FROM {model_table} mt, card_campaign cc, card_campaign cc2
    WHERE cc.res_model = %(old_model)s AND cc2.res_model = %(new_model)s
        AND mt.{ref_field} = c.res_id AND c.campaign_id = cc.id"""

Verification

• Bug reproduced standalone with psycopg2's sql.SQL().format() → KeyError: 'old_model'
• Fixed template formats cleanly → %(old_model)s / %(new_model)s remain as query params for cursor.execute, ref_field becomes a quoted identifier
• No-op functional change in behavior; same fields still pinned, same parameters still passed