The use-case that motivated this fix is the deletion of a project task
with many subtasks. The field 'project_id' on tasks is recursively
computed, and some automated action must be executed when its value
corresponds to a given project.
The issue occurs when the domain of automated actions is evaluated by
method search(), because the latter flushes the fields to search on,
which are also the ones being recomputed. Combined with the fact that
recursive fields are not computed in batch, this leads to a huge amount
of recursive calls between the automated action and flush().
The execution of task.unlink() looks like this:
- mark 'project_id' to compute on subtasks
- delete task
- flush()
- recompute 'project_id' on subtask1
- call compute on subtask1
- in action, search([('id', 'in', subtask1.ids), ('project_id', '=', pid)])
- flush(['id', 'project_id'])
- recompute 'project_id' on subtask2
- call compute on subtask2
- in action, search([('id', 'in', subtask2.ids), ('project_id', '=', pid)])
- flush(['id', 'project_id'])
- recompute 'project_id' on subtask3
- call compute on subtask3
- in action, search([('id', 'in', subtask3.ids), ('project_id', '=', pid)])
- flush(['id', 'project_id'])
- recompute 'project_id' on subtask4
...
closesodoo/odoo#80141
X-original-commit: e2788b580ef15ef3083ac919737a46d850143832
Signed-off-by: Raphael Collet (rco) <rco@openerp.com>
Raphael Collet
1de5dc4b96