It seems to randomly save a bad value into the cache for star_date(time) field.
For already impacted database, it can be easily fixed with a simple update.
update calendar_event set start_datetime=start, stop_datetime=stop where allday = 'f';
update calendar_event set start_date=start, stop_date=stop where allday = 't';
Than can cause errors for google synchro.
Read(['start', 'start_datetime', 'start_date']) make the cache corrupt
for computed field start_datetime or start_date.
Now we don't read (for nothing ?) these 2 fields.
We can remove the skip for the test #500fabf1
Before this commit, a simple :
x = self.create({'name':'subject', 'allday': False, 'start': foo, 'stop': bar})
create randomly an event withtout start_datetime
Observed via google sync, for some events being moved around during
sync the stop_datetime ends up being before the start_datetime failing
the sync. Not (explicitly) reading the computed fields from the old
record properly recomputes them during copy.
Underlying issue linked to (stored) computed fields when multiple
fields are computed by the same method *and* the creation provides a
subset of these fields explicitly: depending on the order in
which *recompute* iterates the fields, it's possible that some of the
fields provided explicitly get extracted (from the cache) before
fields not-provided trigger a recomputation (and overwrite) of
computed fields.
In this case `recompute` iterates [start_datetime, start_date,
stop_date, stop_datetime], if `start_datetime` and `stop_datetime` are
provided explicitly to create, first recompute will check for
`start_datetime`, find it in the cache and extract it, then it checks
for `start_date`, does not find it in the cache, calls
`_compute_dates` which writes all four fields, then it extracts the
recomputed `start_datetime`, creating an incoherence: `start_datetime`
is the one from the old event while `stop_datetime` is the one from
the update/computation. This may lead to invisible corruption if the
event is moved forwards (e.g. creates a multi-day event) and triggers
a validation error if the event is moved backwards (stop_datetime is
now before start_datetime).
ping @rco-odoo multi-field computes may need some special handling
with respect to invalidations & sequencing & the like (e.g. check
their caching as a unit or something).
OPW-771886
A recurring event with a start date and a time duration had a wrong start and
stop date computed.
Introduced at 76b724294a.
Previously the r_date contained only the start_date but now it contains both
start and stop dates. This implies that the method 'get_recurrent_ids' could
use 'get_search_fields' with a stop_date (depending on the domain order) and
returns incorrect dates.
This commit ensure to launch the get_search_fields with the start date.
Closes#15922
Add a 3 use cases with tests that were failing before.
datetime filters were not working correctly on stop filters
The `_get_recurrent_date_by_event` was only computing recurrency on the start
field of an event which can make a difference. A recurring event
start=2017-01-01 09:00:00; stop=2017-01-01 10:00:0
was not matched in a search ('stop', '>', '2017-01-01 09:30:00')
The filter != was not handled in the recurring use cases.
e.g. making the search 'start is set' in the web client is equal to
('start', '!=', False)
Searching with only a start filter matching recurring events but not the
first ("real") occurence, was returning no result if no stop filter was present.
start=2017-01-01 09:00:00; count=5
had no result in a search ('start', '>', '2017-01-02')
Christophe Simonis