When the user chooses as product image a file which is not an image, the
message "Could not display the selected image" is displayed. However, at
saving, a traceback is thrown since the file chosen is uploaded anyway.
If the image cannot be displayed, the image field is cleared.
opw-672206
the widget FieldMany2ManyTags does a rpc to fetch the fields (just to
know if 'color' is in the model...). It was done in a mutex and used
later in a mutex. This commit simply use the willStart method to make
sure the data is available before rendering.
In 9.0's 05adb7f the `<div class="address_format"/>` wrapping an address
city, state_id and zip fields was removed.
There is a hack on fields_view_get which transform this block depending
on the company country, so without the block the feature isn't enabled.
This commit adds a class on the new block `<div class="address_format"/>`
used to wrap the whole address which will change the sizing of the city,
state_id and zip fields as it was before version 9.
To fully follow what was done previously, the zip is aldo moved before
the city if the address format needs it.
Thus we now have as before the three format as follows:
[default format]
city state zip
[format .o_city_state (brazil)]
city state
zip
[format .o_zip_city (belgium, netherlands, ...)]
zip city
state
Another change is brought by this commit, before most element attributes
from city, state_id and zip fields were lost whilst doing the hack, now
they are all kept.
closes#10692
opw-666567
The implementation of `__len__` & `__nonzero__`
were not defined for the py types `wrapping_dict`
and `wrapping_list`.
For instance, this prevented an empty list
to be considered as `False` in domains
evaluated by the web client.
e.g., in the automatic partner merge wizard,
the domain
```
[('id', 'in', partner_ids and partner_ids[0] and partner_ids[0][2] or False)]
```
is used.
When the `partner_ids` list was empty (meaning the "partners to merge" list
was empty in the wizard), it wasn't considered as
`False`, and it therefore attempted to get
`partner_ids[0]`, which resulted to a crash, as
the list was empty.
opw-672095
Closes#11343
Courtesy of Xavier Morel (xmo@odoo.com)
- Issue: Edit sale.order form view, add new line, change product who have
taxes and click quickly on save (with slow network) => error message.
Because some deferreds are lost.
- reduce number of reads and onchange for x2m widgets
- remove before_save method and use commit_value (the issue was that if the
before_save and the change trigger an onchange, the onchange deferred is not
checked before saving the content.
When using the left/riht arrow to switch of record
in a form view,
the rendering of the `char_domain` widget was done too
early, before all fields are ready / set, and if the
domain of the previous record could not be applied
on the current record, it leaded to a traceback.
This revision introduce a new deffered,
because I coudln't find one that was doing
what I was looking for:
- `is_initialized` is a defferred which is
resolved when the form view has finished
its rendering for the first time
- `reload_mutex` is a mutex used only when reloading or
switch to left/right record.
While the one needed in this case is a deferred
which is resolved when the record has finished
being rendered, wether when its when coming
from the list view to the form view (and it's
not the first time the form view is loaded,
e.g. list view -> form view -> list view -> form view, another record),
or on the switch of left/right record.
opw-671594
Test to bind save and discard buttons of an editable tree view are done
only once when initializing it.
But if the view is by default grouped at the beginning, it will be then
"not editable" but only removing the groupby will change this.
So this commit add a case so if the view is grouped when initialized,
the buttons are always bound.
closes#11198
opw-666794
When the user click on save, 'is_valid' is called on each field.
The o2m test all sub-field (apply modifier and check value).
But, if a modifier change the read-only attribute, the field is reinitialized and call 'render_value' without check 'no_rerender' value.
The relational fields call the db in render_value, and the x2m is invalid because a request is pending.
(change X2ManyListView to keep the 'no_rerender' initial value)
opw-653120
The Dialog class destroys the modal (and so removes it from the DOM)
when the dialog closes but leaves the modal-open class on the body
element if there are still other modal(s) in the DOM.
The problem is that (maily on the frontend side) there may be others
modal(s) in the DOM (not handled by the Dialog class) but which are
not shown to the user, and in that case the modal-open class should
be removed.
DO NOT FORWARD-PORT: the functionality is not used anymore in v9
Commit aaf9badb filters the returned ids by keeping the domain in the
request. However, `this.dataset.domain` might be empty, while
`self.dataset.get_domain()` will retrieve it from the model.
Since `get_domain` is not always defined on the dataset, we keep
`this.dataset.domain` as a fallback. The fix is not great, and a better
solution should be found in master branch when the web refactoring is
done.
opw-666755
The copied textarea used in autoresize was a valid target for focus when
using the tab key. This is a major issue in editable list views,
because when the user leaves a textarea, the list view jumps up, with no
visible focused field.
Translated text is stripped of starting and ending spaces on Transifex which
makes the result text look bad (as translators do not put spaces in their
translation either).
Add explicitely a space to be sure it's present in the translation.
Will have no effect on English as it's HTML
When clicking on a pivot view cell, the gotten view were default views.
For example when on the opportunity pivot view, after clicking on a cell
we we would get to the crm.lead default view instead of the opportunity
one.
This commit keep references to current action views when clicking on the
cell.
closes#10768
opw-668472
With 003d85b instead of getting the name of the relation
field_relation we get field_relation/id which will get us an xmlid
instead.
But the data about related fields are not gotten all at once when
opening the export modal. They are gotten by clicking on the arrow of
the related record.
It is in this data that the /id will be appended for the field, but when
getting a saved export list, this data may not be available.
This commit immediately add the `/id` when a field is put in "Fields to
export" column, so this inconsistency is no more.
closes#10640fixes#10327
opw-665994
When editing a line, leave the buttons visible.
Also remove the 3px gap between fields so that readonly field texts
does not move when entering the edition mode.
* Fix the grid display of images (use the column property)
* Use Dialog class
* Allow removing a cover image
* Add image selection before choosing it (or double click to choose)
+ add less mixins in web for columns property
The module system allow modules to return a deferred. When that
deferred resolves, the module is considered to have succesfully loaded,
and the result value is its value. Then, the boot.js code used a
setTimeout (via _.defer) to load all its dependencies.
This is not wrong per se, but it empties the call stack, which give
other code the opportunity to interfere. In particular, it is an issue
in phantomjs tests, because it causes a window between the
domcontentloaded event (which is a key event in website) and the actual
loading of all dependent modules. This execution window can be used to
start some tours, but the code of those tours is waiting the next tick
to load, so horrible crashes happens.
This commit calls immediately process_jobs to avoid those issues.
The dataset of an x2m could be set as changed on a focus event, but if
the form is currently not in editing, it could stay blocked in this
state.
This commit add a check so blur and focus on x2m are better coupled.
opw-665401
When editing a form with the arabic language, the date is formatted in
arabic. But the version of the Moment.js library we use for this had
inconsistencies and parsing the arabic date could not work.
Hence, saving would not work since date field values would be invalid.
This commit cherry-pick http://momentjs.com/ commits which solves that:
from version 1.8.4: https://github.com/moment/moment/commit/c9023b5
from version 1.9.0: https://github.com/moment/moment/commit/bd328d5closes#10634
opw-660453
This commit does not actually fix qweb2.js, but it works around a
phantomjs bug. The problem with phantomjs is that in some situations,
the xmlhttprequests are dropped with a status 0. It is most likely
related to the issue:
https://github.com/ariya/phantomjs/issues/11195
In our testing framework, it seems to appear when requests sent before
the 'load' event of phantomjs are received after, and also when those
requests are listened with the onreadystatechange handler.
The bug does not appear if we use addEventListener with the 'load' event (at
least, not for a few builds)
This commit does not worsen qweb, and it is also a small improvement in
style (in my opinion).
The definition of "total" as a global variable is done too soon in the
JS. Depending on the order in which the JS modules are loaded, it is
very unlikely that the translations are already loaded at that moment.
opw-666577
In most backend views using CTRL+P to use browser printing would only
print an header without the content.
stumbled-uppon-when-checking: opw-659430 opw-660763
The action "Technical Translations" was not retrieving the correct translations.
Since 9.0, the translations names no longer contains the name of the model (in
the form "field:<model>,<name>") but are records on ir.model.fields (in the form
"model:ir.model.fields,field_description:<model>_<name>").
The translations of type selection are style a different type (type 'selection').
Create a method that will find the linked fields and retrieve the correct
translations.
Fixes tw:reis_pt/status/687419541199187969
In most backend views using CTRL+P to use browser printing would only
print an header without the content.
stumbled-uppon-when-checking: opw-659430 opw-660763
The problem occured when the user quickly clicked several times on the view
switcher or the breadcrumbs. The function switch_mode() of the view manager
was called once per click. This function's role is to prepare the requested
view, and once it is ready, to remove the former one from the DOM and insert
the new one instead. It is thus asynchronous as it waits for the
fields_view_get and search_read RPCs' response.
Unfortunately, when those calls to switch_mode() overlapped, the former view
reference was lost and it was thus never removed, producing a weird situation
where two views were displayed one above the other. This may happen when a
view takes a while to load, or simply when the user double-clicks on the
breadcrumbs or view switcher.
This rev. makes sure that calls to switch_mode() can never overlap by refusing
any other call when there is still one processing.
We also added a protection against breadcrumbs and view switcher double-clicks.
opw 658924
When an error occured when switching from a view to another (e.g. switching
to a form view for an id that doesn't exist, by editing the url), active_view
and view_stack were updated even though the switch failed. The view manager
thus lost the reference to the real current active view, which couldn't be
detached anymore. This resulted in two views being displayed at the same time
if the user kept navigating through this view manager.
This is a backport of odoo/enterprise@0b1aafa6b and odoo/enterprise@8e691be.
Nicolas Martinelli