Hardening of commit:995610c065bf0242cb5023cfad2940234395739e
The JournalDashboardGraph requires nv, which is lazyloaded.
However it requires nvd3.js which is loaded after nv.d3.js.
A function called in destroy is defined by nv.d3.js,
which can make the crashhappen with the right (wrong) timing.
opw 1873749
The resequence mechanism works if the sequence is sorted by its handle field.
Therefore sorting on another field should lock the handle, while sorting on the
handle field should unlock the handle.
opw 1867049
With this commit, we introduce a new 'fail fast' feature to our qunit
test suite: when it is activated, the qunit test suite will immediately
stop after the first failed test.
It is accessible as a flag in the url (failfast), or by clicking the
checkbox in the UI. It is currently not activated by default.
Note that this commit also change the url for the runbot phantomjs test
in order to activate this feature. This allows us to increase the global
timeout for the js test suite without fear!
Backport of 611c836a46
This reverts commit 4cb585ab4f.
When creating from the kanban view:
- crm.stage
- hr.recruitment.stage
- project.task.type
- helpdesk.type
- mrp.eco.stage
We go from a dashboard of eg. "GroupRecord" named "group_id" in the
record. And when a stage is created, "default_group_id" is used. With
the change the stage was not linked to the current "grouping".
We could probably use "search_default_group_id" instead but the behavior
may be present in existing customization.
opw-1884217
closes#26943
Before this rev., there was a crash when a FieldDomain was reset
with a new domain (e.g. coming from an onchange). For instance, it
crashed in Email Marketing > create new Mass Mailing > select
mailing lists. Introduced by 68332a01.
opw-1882850
The resequence mechanism works if the sequence is sorted by its handle field.
Therefore sorting on another field should lock the handle, while sorting on the
handle field should unlock the handle.
opw 1867049
When a column is created in the kanban view, the context of the window
action is fully given but it may contain default_{field_name} context
keys that are intended for the action model.
eg. on a product.product view we could have {default_type:'product'} in
the context, if we grouped by company_id and created a new company, we
would the context is propagated and we would try to set type='product'
on the company (which causes an error).
Without the change, the added test fails with:
default_* should be removed from context, actual: "true", expected: "false"
opw-1879064
closes#26676
Be in a grouped kanban view with progressbar with limit 10:
- records 1 to 10 are loaded on a column
- do "Load more..." on a column => records 11 up to 20 are loaded
- do "Load more..." 2nd time => records 11 up to 20 are loaded
- do "Load more..." 3rd time => records 11 up to 20 are loaded
- ...
While in reality it should be:
- do "Load more..." 2nd time => records 21 up to 30 are loaded
- do "Load more..." 3rd time => records 31 up to 40 are loaded
- ...
This was caused by the progressbar reloading the column's group. This
set the loadMoreOffset to 0 thus forgetting the current state.
Without the change, added tests failed with:
records of column are loaded => actual: "1,2,2", expected: "1,2,3"
opw-1878359
closes#26628
Suppose you select a customer for a sale order.
You type the beginning of his name "think", and a name_search is triggered
which finds matching results, with first result "Think Big Systems".
There are three ways to select it: click, enter or tab.
Now there can be a warning set on this customer
("Bad client, only accept cash", or "good client, offer discount").
In that case, clicking on tab would trigger the onchange displaying the warning,
but it would also set floating to true, which means a dialog to create client
"think" would appear.
This happens if the onchange is delayed, and thus the call to reset the floating
state is triggered only after the focusout completed, instead of before.
Coauthored with @aab-odoo
opw 1866619
opw 1874475
The pivot view didn't take into account the widget set on the fields.
In particular this was a problem for float fields that store time values.
They are now displayed with float_time if declared as such in the view.
opw 1876445
The domain was instantiated without being given any evaluation context.
As a result, the uid variable was not defined,
crashing the js if present in a user-defined filter.
opw 1866852
Before this commit, when we applied two or more groupbys in line
graph, the groups displayed on the X axis where shifted to the left,
and the first one wasn't displayed at all.
related task: #1848289Closes: #25037
Let x_1, ..., x_a, x_{a+1}, ..., x_{a+k}, x_{a+k+1}, ..., x_{n} be a record
sequence ordered by its 'sequence' field.
Suppose the reordering occurs between elements x_{a+k} and x_{a+1}.
If the subsequence x_1, ..., x_a is ordered, then it does not change.
This is also true of the subsequence x_{a+k+1}, ..., x_{n}.
The resquence function makes exactly this assumption, and only sends the server
the subsequence x_{a+1}, ..., x_{a+k} to update its 'sequence' field.
The update of x is done as follows: it is index + offset, where index is the
position of x in the subsequence, and offset is the 'sequence' of the first
element of the sequence (i.e. x_{a+1}.sequence).
It is easy to see that we need another hypothesis: that the sequence numbers are
unique. To show it, suppose that the sequence of x_1, ..., x_a is 0, then all
x_{a+1}, ..., x_{a+k} have sequence 1, and lastly all x_{a+k}, x_{a+k+1}, ...,
x_{n} have sequence 2. Then after the call to the server resequence, the
subsequence x_{a+1}, ..., x_{a+k} would have sequence numbers 1, 2,..., k.
Since the x_{a+k+1}, ..., x_{n} subsequence have 'sequence' value 1, this would
obviously be incorrect.
We fix the reordering of the data in the js frontend to be consistent with
what is done by the server.
Note that this is only pertinent for sequences of reordering moves, since
otherwise the data comes directly from the server.
opw 1867049
You:
- open a grouping in a list view
- check a line of this grouping
- open another grouping
=> all checkbox are unchecked
With this commit, 0982ceaa is improved so the selection is only reset
when the list has been reloaded.
opw-1865736
closes#26213
The 'drag and drop' function defined in test_utils should be aware of four
different cases, moving up or down, above or below another row.
We need an offset of one pixel for the function to work when moving down
(this is a case of broken symmetry because of <=)
Related to opw 1867049 regarding sequences of resequence moves.
The value of the statinfo widget is rendered by the qweb after a t-esc.
Say if the python rounding returned 14.0000000001
(supposed to be a two digits precision number through the magic of floats)
then the statinfo widget would print it as-is.
We now format the value at assignation.
opw 1865426
In a one2many, if the user clicked the trash button while editing another line,
the row was not unselected.
Thus the modal was not triggered, which entailed that the currently edited line
was always saved.
In particular, it allowed to give empty values to required fields.
opw 1834346
Suppose there is a one2many field embedded in a one2many,
(say we are viewing record A with one2many field B itself with a one2many C).
Furthermore, there is a second page of the one2many field B.
If an onchange is applied on B, then the server replies with a 1 command for all
B records attached to the currently viewed object A.
This includes a list of commands [[5], [4, C_id]*] for each B record,
in particular B records that are on the second page.
Since they are on the second page, they might not have been fetched by the
frontend; in that case the corresponding C_ids are thus unknown.
As a consequence, on save the frontend sends a command 0 to create the records
for all unknown C_ids.
Since it doesn't have any value to give to the 0 command, the row creation
usually crash in the backend.
Another wrong behaviour is fixed at the same time: if the server sends an update
on the list of C ids with 4 commands.
In that case, if the record has not been prefetched, we have no way to know
if the list changed or not.
To solve this, when we receive the 4 commands by the server, we send back 4
commands.
coauthored by @aab-odoo
opw 1835936
Before this commit, there was a traceback in the following scenario:
Let's have a kanban view with groupby and a domain.
If we clear the domain then clear the groupby, there could be a crash.
This crash occurs because when we clear the groupby, it performs a `reload`,
and assumes that the datapoint contains grouped data. However, if this
computation takes some time, it is possible to overwrite the datapoint so that
there is no longer grouped data.
This commit fixes the issue by enforcing at most 1 execution of the method
`reload` active at any time. Consequently, successive reloads operations are
now much slower.
Note that it uses the single non-reentrant mutex on the basic_model to enforce
this rule, which means:
- No more concurrent `reload` and `save`.
- If an execution having the mutex in `reload` come accross the same mutex,
it is blocked forever. This is hopefully not the case at the moment,
however a reentrant mutex should be considered in the future.
Have a view with a field as:
<field name="stuck_in_the_middle" attrs="{'readonly': [('with_you', '=', False)]}"/>
Before this commit, the contents of attrs were parsed and merged into the modifiers
and the attrs attribute of the node was deleted
which might cause divergence between reality and tests
After this commit, the mock server does what the server does in this case
closes#25583
Let's assume a one2many A inside a form view, such that the one2many
views (list, not editable, and form) are defined inline, and there is an
onchange on A. There is another inline one2many B editable list
displayed in the inner form view (and B isn't in the list). There are
two records on the one2many A, both of them having one record in their
one2many B. Edit the first record of A, and edit it's subrecord in B.
Click on save. This triggers an onchange on A and the onchange returns a
command 5 (DELETE_ALL), and two commands 1 (UPDATE). Now edit the second
record of A, and simply close the dialog (e.g. 'Discard'). Save the main
record: the record in one2many B of the second record of A has been duplicated.
This occured because the code in BasicModel messed up information coming
from the onchange (UPDATE command) and from the read (performed when
opening the second record), thus producing a CREATE command instead of
an UPDATE one.
This rev. ensures to properly mix the read data with the onchange
result, by transfering the changes over the list datapoint created after
the read.
opw-1836785
closes#25330
Revision on b09f0c99
The intent of the previous commit was to no drop records in a list
on discard in the following cases:
- when they have been created from a `default_get`
- when they have been created from an `onchange` that result
from a `default_get`
Before the commit above, the kind of records that were abandoned
in those cases are stated as invalid, which happens when a record
has a required field that is empty.
The case of an `onchange` that is not involved in a `default_get`
was not considered in the fixes above. In fact, this is the intent
behaviour in most cases. However, if the items in the list are
created from such an `onchange`, we should not drop them.
This commit fixes the bug where records in a list that are created
from an `onchange` are wrongly abandoned. Here is the new logic to
abandon a record on discard in a list:
- the record that is marked as "do not abandon" should not be
abandoned;
- a record that is registered as a "not new addition" in the
list should not be dropped;
- a record that is registered as a "new addition" in the list,
and has been updated afterwards, should not be dropped;
- a record that is not new should not be dropped;
opw-806650
Have a model A with a o2m to B.
In the list have a field B > m2o > C
Also have a sequence field (widget handle) on B
Then Create a record A
Arrange yourself to have two new records B popping right away (through an onchange)
in the o2m. Their C field have to be empty
Resequence those two B lines with the handle
Before this commit, there was a traceback when trying to do an name_get on the C fields
Since both of them are empty, there is no record in the localData
Hence, no model. And also no records anyway, so no need to do a name_get
After this commit, there is no traceback, the name_get is avoided when no data is to be fetched
OPW 1853088
closes#24988
Model A has a o2m to model B
Model B has a m2o parent_id on B
In A's form, add a B item,
then in the m2o, create record
Save and New
Click on the m2o
Before this commit, the python crashed on name_search.
This was because we sent the virtual ids to it
After this commit, there is no crash
OPW 1850212
closes#24948
Have a decorated (conditional on field) editable list, with a default_order.
Have more records than the pager allows.
Add an item, and hit escape to cancel addition.
Before this commit, the cancellation crashed because the first record on the other page
was trying to evaluate the decoration for its row, which is wrong in the first place.
This was because creating a record increases the limit of the list, to be able to see the record
being created.
At cancellation though, that limit was never decreased.
After this commit, the whole flow happens without problem.
OPW 1844495
closes#24682
Let's assume the following scenario in a Kanban view with a default
filter. The user removes the filter, and quickly adds a new one.
On a slow network, it may happen that the second reload request
returns before the first one. On odoo.com, this is easy to
reproduce on the tasks Kanban view.
Before this rev., when this occured, the result of the first
request was displayed, whereas it should have been ignored as
another request was done later on.
Revision on https://github.com/odoo/odoo/commit/9a9da5763b38a3bce54e9c4870af0032e38cacc3
The intent of the commit above was to no abandon records in a list on
discard when these records have been created from a `default_get`.
Also, the `default_get` may also perform an `onchange` that could create
more records. The records from this `onchange` must also be included in
the "do not abandon" rule.
However, the rule was too relax, so that any following `onchanges` from
a user interaction was also considered into this rule. As a result, it was
not possible to abandon such records.
With this fix, we restrict the rule to `default_get` (already the case) and
only the `onchange` right after a `default_get`. Any record created on an
`onchange` from a user interaction after a `default_get` should be abandoned
on discard.
Representation of creating the list:
savepoint
|----------|-----------|-------||--------|------->
create default_get onchange1 onchange2
record in the list is created by default_get or onchange1
=> do not abandon on discard
record in the list is created by onchange2
=> abandon on discard
Before this commit, some records in a list did not show the name of their
many2one fields.
This case occurs when the records in the list are created from a default_get,
and the records do not fit in a single page. In that case, the records in the
first page correctly display the names of their many2one fields, but not the
records in the other pages.
Fixes#24547
Before this commit, when a list contains records with many2one fields that
contain the same values, duplicate ids were passed to the server on rpc
`name_get`.
For instance, if 40 records in a list refer to a Unit of Measure with id "1"
whose displayed_name is "unit(s)", the server receives a long list of ids with
simply the value "1". Also, the server responds with a list of displayed_name,
in this case a list of 40 strings "unit(s)".
With this commit, we always pass unique ids to the server on rpc `name_get`.
When sorting records according to a many2one field in a list view
embedded in a form view, the js would sort according to their ids,
in a seemingly arbitrary order.
Thus sorting records according to a field with values in ("a", "b", "c")
could be sorted as "a", "a", "c", "b" instead of "a", "a", "b", "c".
We make the comparison on the name of the field instead.
opw 1838860
Recently, rev. c0809dd41 removed a protection in the reload of
Kanban views with progressbars. This protection ensured that the
manual reload of the progressbars was only done when one of the
records of the Kanban view was updated (and not when a column, or
the whole view was).
However, removing it highlighted a bug linked to the manual reload
of progressbars, which messed up the list of ids in the
environment. As a consequence, for instance, if the user archived
the records of a column, and then opendc a record in form view,
it couldn't use the pager anymore.
Besides fixing this bug, this rev. re-introduces the protection,
and refines it to prevent 2 useless RPCs when the whole Kanban
view was reloaded.
opw~1841276
opw~1841342
Assume a one2many field inside a form, displaying field A in its
list view, and a field B in its (dialog) form view. Moreover, there
is an action button in the dialog, which updates the value of B.
Before this rev., B wasn't correctly reloaded when the button was
clicked, because the BasicModel reloaded the fields on the list
view (i.e. field A), whereas it should reload the fields in both
views.
This fix is twofold. First, we force the viewType to 'form' when
the record is reloaded, instead of the default ('list'), so that
fields in the form view are correctly reloaded. Second, when we
open the form view in the dialog, we add the fields of the list
view to the dict of fields of the form view, so that the form
is aware of those fields and can ask the model to reload them
(still without displaying them).
Fixes#24189
Have a One2Many
The views of that o2m define two inline views: the list and the form
in the list view, put a field that isn't in the form and used in a domain
Try to add an item to the list
The form pops up
Fill in the data and save & close
Before this commit:
It crashed when evaluating the domain on a field using the field that is not in the form
After this commit:
Subviews share their fields in a special object on the parent record
In our case, the form view checks in that shared resource if the fields it is supposed to initialize
are there.
Hence, this flow doesn't crash anymore
OPW 1837511
closes#24350
The way the calendar view selected its form view was broken for two
reasons:
- any value given in the form_view_id attribute was kept as string,
which means that the string value would be sent to the server, and
this is not good.
- if no value was set, the calendar view then simply did a do_action
with the form view id set to false, which means that the default form
view is used. This is mostly ok, except when we are in the context of
an action with a form view which is not the default one. In that
case, we clearly prefer the form view from the action.
For example, before this commit, the form view in the timesheet
application (community) is not the same as the default form view on
account.analytic.line.
Note that this commit also adds a small tweak to the mock server to
better simulate errors like the web client does (in session.js).
Before this, the progressbar values were not updated (both count and progress)
after archiving all the records of a column.
Multiple things have been done here to update it.
Firstly, a optimization had been done to avoid reloading the progressbar when
updating a column ; this optimization was not correct due to this precise use
case. This will update the progressbar.
Secondly, the `aggregateValues` are reset when putting empty groups back in the
datapoint. This attribute is used to compute the counter if a `sum_field` is set
on the progressbar widget. It needs to be reset because the loop that computes
it won't iterate on empty groups.
Eventually, the mockRPC of the progressbar route was not taking the `group_by`
argument into account (probably a typo).
When doing several grouping quickly (or if the system is slow) on the
pivot view, we may try to display data that now doesn't corresponds to
the current selected groupBys.
This commit ensures that when we load data of the pivot view, we drop
the previous data loading that are currently being carried out.
Without the fix, the added test failed with:
TypeError: Cannot read property 'length' of undefined
because the test does:
- select one level of group by
- select two levels of group by
- resolve the first level of group by
- resolve the second level of group by
Without the fix, we would try to display the data for one level of group
by when the pivot view had two levels selected.
With the fix, displaying the first level of group by was discarded once
a second one was being requested.
opw-1825892
closes#24396
When a user creates a new column in a kanban view, leave the action and
comes back, the order is not preserved. With this commit, we force a
call to resequence to ensure that the order does not change.
The initial rev. odoo/odoo@2d3ecf2 that was supposed to handle this use case
was in fact only dealing with one command as default value for nested x2m, not
more.
The basic model can now handle multiple commands.
Closes#24241
and keep all those filters at last. (PR#20463)
e.g. If field used in color attribute of calendar view does not have value then such sidebar filters will not have string besides filter checkbox
in such cases we should set Undefined string and put all those filters at last in filter list
Currently we can see issue in calendar of manufacturing orders in db with demo data
Related to Issue #1778360
Co-authored-by: Mohammed Shekha <msh@openerp.com>
Here is a scenario that could happen before this commit:
1. go to a list view with more than one record, say 3
2. click on the 3rd record to open the form view
3. the pager says 3/3
4. click on edit button
5. click on discard button
6. the pager says 1/3 (but the record displayed is still the same)
It is often necessary to restore the offset for the sub records, because
the number of pages in a one2many could have changed. However, it is
not a good idea to do that for the main record, since it interferes with
the usual flow of operations.
MOENS Alexandre