Nota: If we replace sequence signaling for cache invalidation with pg
listen/notify in the future, we will use the same mechanism for more accurate
cron timing.
bzr revid: al@openerp.com-20121209170447-zs0k3jazokylwvar
As of 7.0, RNG validation is not possible for form views
that have a version attribute equal to "7.0", due to the
allowed usage of HTML syntax mixed with the regular OpenERP
view syntax. RNG validation is still enabled for regular
form views (@version missing or less than "7.0"), and for
all other views types.
Validation of 7.0 form views should be improved with the
addition of an assertion-based schema, still to be done.
The above is also complemented with an explicit call to fields_view_get()
during view installation, in order to immediately verify
that the updated view hierarchy does not cause any
issue when loaded along with its related views (i.e
parent and siblings, for inheriting views).
In addition to that, fields_view_get() will now only
consider loading views that belong to modules that have
already been loaded. This avoids a lot of validation errors
during a module update operation, which runs on top of
an existing database with all previous views visible,
even those whose module is not loaded yet.
bzr revid: odo@openerp.com-20120611122758-qcw9xdhupl24busq
When two requests arrive simultanously for the same uninitialized db,
the first request starts the db initialization, but the second one
immediately gets the partially uninitialized registry (actually just
created, so generally completely uninitialized), leading to an access
error in later code (as soon as a registry object is accessed).
Add a GRML (Global RegistryManager Lock) to ensure the RegistryManager
*never* returns a partially initialized registry.
The current implementation is simple (just lock all RegistryManager
methods before they manipulate registries), but overly broad. This is
an area which might be optimizable if there are perf/responsivity
issues (e.g. each Registry instance could have a lock, and the
RegistryManager would grab the instance's, allowing the inititlization
of registry A not to block registry B from being returned in heavily
concurrent uses).
However this is not an issue in multiprocessing scenarios, which are
being planned for the near future. So for now, being correct is
probably the best idea.
bzr revid: xmo@openerp.com-20110916075227-0zutzlxn2dcd94c4
This is necessary to preserve the model loading order
as defined in the module. This was broken when the
metaclass was introduced to make the explicit model
constructor call optional.
One consequence is that the order in which the classes
are declared in the module really defines their
initialization order, even if the constructors are
later called explcitly in a different order.
This latter case should be fairly rare, and easy to
fix too - simply putting the class declaration in the
right order.
bzr revid: odo@openerp.com-20110826161736-lgnpurtbcqtbseey
- make it more explcicit in osv_pool that afreshly instanciated object is added to the pool
- osv_pool.init_set() is called just once with False, so
no need to check if it is the first time it is called or
if is called with True
- and rename it in do_parent_store as it is what it does.
bzr revid: vmt@openerp.com-20110526210532-a8i91shptz5h4k48
Antony Lesuisse
Vo Minh Thu
Christophe Simonis
Olivier Dony
Xavier Morel