There were several issues with the resequence method.
First the lookup in the model's fields to determine if there is a
'sequence' field (to know whether or not an RPC should be done)
was incorrect when resequencing columns as the lookup was done for
the wrong model. Actually, we don't have the list of fields of the
related model, so this lookup shouldn't exist at all, and the RPC
should be done in all cases (the controller correctly handles the
absence of 'sequence' field).
Second, the JS sort of the data didn't work either as we didn't
manipulate the dataPoints but only their ids (the internal
representation of the dataPoints is flat, so the data of elements
of type 'list' only contains the id of its dataPoints).
Returning a rejected deferred isn't the correct way to notify the
MockServer that something went wrong, throwing an error is. This
error is then catched, and the RPC deferred is rejected with two
arguments: an object describing the error and a JQuery event.
When the user quick creates a record in a grouped kanban view, a
name_create RPC is done. This RPC may fail, and there were
several issues with the fail handler.
First, those fail handlers take 2 params: an object describing
the error, and a JQuery event. The event was the first argument
in the handler, so it crashes on event.preventDefault().
Second, when the name_create fails, a FormViewDialog in edit
should be open. Former code opened a SelectCreateDialog.
This commit also ensures that the default values are correctly
given to the form in the dialog, through the context.
When reloading a grouped Kanban view (e.g. after specifying a new
domain), the internal state of the (sub-)groups is kept so that
it can be restored (like which are the open groups and the limit
and offset inside the groups).
However, for models and fields whose read_group returns empty
groups (e.g. project.task grouped by stade_id) this was wrong if
the specified domain filtered out all records of an open group,
because as its data wasn't reloaded (because there is no record),
the old data were still in place, so the filtered out records still
appeared in the Kanban.
The `FieldDomain` was sending string domains instead of array domains
to the server (by its associated `BasicModel._fetchSpecialDomain`
function). The server does not do the conversion string -> array for
domains, so the RPC broke everytime.
This commit also modifies the `MockServer` so that tests also break
when sending string domains.
It also improves the way failed rpc are handled. Indeed when an rpc
fails, its deferred is rejected with two arguments: the error object,
and a JQuery event (that will display the error if not
preventDefaulted). The mockRPC didn't pass the second one and thus
there was a crash when reaching code that e.g. calls preventDefault
on that event. Note: this also removes the stringify/parse of the
error object as it is not a correct simulation of a server error
object anyway.
* Data manager load_view returns a viewInfo object
* ViewInfo contains: the arch, fields and fieldsInfo objects, and some extra informations
* The fields come from the server and are frozen, without any view attributes (depend of the model).
* The fieldsInfo contains every view attributes and informations.
This rev. allows to specify some kwargs of read_group directly in
an Object given to kwargs(), instead of explicitly specify them
using the corresponding method (e.g. to specify the context in the
kwargs instead of by calling .withContext().
When calling a route using the new RPC query builder (e.g. by
doing "this._rpc('/a/route').params({a: 'a'}).exec()"), if one of
the route kwargs was 'model', 'method', 'args' or 'kwargs', the
given value was overriden by undefined.
The session can me mocked in the test environment (if a session
is given in the params). However, when it is not, it crashed
if one of the instantiated widgets requests the session with
getSession().
This rev. adds a fallback so that when this happens, the real
session (web.session) is given.
When reloading a view by clicking on its icon in the view switcher,
some params of the view (e.g. limit) were reset to their original
value. So if the user previously updated that param (e.g. by
editing the pager), these changes were lost.
This rev. ensures that those original params are only given to the
view at its initialization, but not on reload.
When the pager is disabled, the user can't go to the next or
previous page. It shouldn't be able to manually edit the pager's
values neither.
Also
- converts pager.js to the new JS coding guidelines
- adds two tests
One might need all the preparation stuff that InputField does on
its input but might need something else as its tagName and handle
the input differently.
This commit gets the input preparation code out from the rendering
part of the widget so one can render a widget the way he likes and
call _prepareInput on his input.
This is used in monetary field which requires all the feature of
InputField but with the input actually being one level deeper in
the DOM than the standard InputField.
Some fields react differently depending on the values in web.config.
For example, the phone field will display a clickable phone link if
it detects a small screen (e.g. a smartphone) but won't do it on a
bigger screen.
To be able to test fields which react differently depending on the
values in web.config, we need a way to mock it. To do that, we
modify web.config in place with the value given for the test, and
we put back the original config when the field widget is destroyed.
Input fields should not be re-rendered by the onchange mechanism
as they might exhibit formatting that you do not want to have when
typing.
The list view was not propagating the original event to the lower
functions like the form view does, so I added it there.
read_group returns a __domain for each group, which is the
conjunction of the domain given for the read_group with the domain
matching the given group.
The MockServer didn't reproduce that behavior.
Moreover, the BasicModel did again that conjunction, leading to
domains with redundant subdomains.
* board, google_spreadsheet
Commit 8e0c7ccc77f358601b0701536fcb33842ca324a6 got rid of the
CompoundDomain class but did not adapt the code correctly.
- Domains have to be concatened, not regrouped in a big array
- String domains have to be converted to array
- "OR" joining must normalize subdomains first (add explicit "AND"s)
Before this rev., the count attribute of list dataPoints was
basically the length of the data Array. It means that for an
ungrouped list, it was the number of records in the list, and for
a list containing groups, it was the number of groups.
This caused a problem in the list view with nested groups, as the
counter in the group header is supposed to display the number of
records in the group, whether or not the group contains other
subgroups or directly records.
This commit changes a bit the semantics of the count attribute,
such that it always represents the number of records inside the
list (i.e. the sums of records inside subgroups if the list is
grouped).
Before this commit, the relational data of a record was only
fetched once, and then retrieved from the cache.
This was premature optimization, that can lead to bugs as those
relational data displayed in the webclient might be desynchronized
with the actual data in DB.
Before: the method '_buildOnchangeSpecs' returned all fields, but some
models like 'res.users' have redefined the fields_get and add some fake
fields which cause a crash.
Christophe Matthieu