The Python constraint method `_check_unique_sequence_number` performs an
SQL query, and therefore needs to `flush()` the some fields before doing
so. However, it uses a generic call to `flush()`. In some situations,
the call to `flush()` ends up into recursive flushing, and that's the
beginning of a nightmare:
- write on account.move.state,
- which calls account.move._check_unique_sequence_number(),
- which calls flush(),
- which computes purchase.order.line.qty_invoiced,
- which reads field purchase.order.line.invoice_lines,
- which calls flush() for account.move.line.move_name,
- which computes account.move.line.move_name,
- which computes account.move.name,
- which calls account.move._check_unique_sequence_number(),
- which calls flush(),
- which computes purchase.order.invoice_status,
- which uses the not-yet-computed value of purchase.order.line.qty_invoiced,
and computes an incorrect value.
We fix it simply by making the call to `flush()` more specific: we only
need to flush 4 fields. This avoids the nightmare above.
X-original-commit: cf5f50a733ddfd87985a39ffe7daffc48cfb582f
Raphael Collet
664ac15b59