### Description of the issue/feature this PR addresses:
Fixes an issue where timesheets lose their invoice reference when reversing and re-creating an invoice via a credit note.

### Current behavior before PR:
When reversing an invoice tied to timesheets and creating a replacement via a credit note, the timesheets linked to the original invoice have their timesheet_invoice_id cleared. Because the modify_moves function builds the replacement invoice directly via copy_data()/create(), it bypasses the normal sale order invoicing flow. As a result, the unbilled timesheets are left permanently unlinked from the newly created invoice, leaving the new invoice with no reference to the timesheets linked to the original sale order.

### Desired behavior after PR is merged:
When a replacement invoice is created, each timesheet is properly relinked to the corresponding line on the new invoice. This linkage matches on the sale order line (so_line) rather than line position, ensuring accuracy since line order and count are not guaranteed to be preserved between the original and modified invoices.

opw-6449995

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Forward-Port-Of: odoo/odoo#285232
Forward-Port-Of: odoo/odoo#283104

Issue:
When creating an asset using an "154.01.01 Vehicles", the depreciation values of the journal entries created are zero

Steps to reproduce:
1. Install l10n_mx and enter a Mexican company
2. Create an asset and setting the fixed asset account as 154.01.01 Vehicles
3. Click on "Compute Depreciation" and in the "Depreciation Board" page, all the journal entries created will not have any depreciation values

Cause:
In the l10n_mx chart of accounts, all accumulated depreciation accounts are set as "expense_depreciation" whereas they should be asset accounts because accumulated depreciation accounts are used to credit asset accounts to decrease asset values. If an account of type "expense_depreciation" is used, then when the depreciation account is credited and the expense account is debited, all financial events will occur within expense accounts, which is why no depreciation values were recorded as the asset balances stay the same. Additionally, the "asset_depreciation_account_id" field on "account.account" has a domain restricting selection to "account_type" of "asset_fixed" or "asset_non_current"

Solution:
Change the "account_type" of l10n_mx accumulated depreciation accounts from "expense_depreciation" to "asset_fixed" and l10n_mx accumulated amortization accounts from "expense_depreciation" to "asset_non_current"

Updrade PR: [odoo/upgrade/pull#10936](https://github.com/odoo/upgrade/pull/10936)

opw-6359618

Forward-Port-Of: odoo/odoo#283390
Forward-Port-Of: odoo/odoo#278185

# Before this commit:
The image controller performance test expects the admin partner to be
unpublished. In parallel test runs, the website_partner demo data
publishes the admin partner, causing the test to follow a different
code path and fail.

However, during parallel test execution, the website_partner module
installs its demo data, which updates the admin partner:

```
<record id="base.partner_admin" model="res.partner">
    <field name="is_published">True</field>
</record>
```

As a result, user_admin.website_published becomes True.

# After this commit:
The test explicitly restores the required precondition by setting the
admin partner's is_published value to False before executing the
performance check.

As a result, the image controller always follows the expected
"unpublished" code path, making the test deterministic regardless of
whether website_partner or any other module with demo data has already
been installed during parallel testing.

Runbot-241102

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

In the bank reconciliation widget, a reconciled transaction
matched with a bill and a loss exchange entry no longer displays
the bill reference next to the reconciled line.

Steps to reproduce:

- Have company currency USD and foreign currency EUR
- Create a vendor bill in foreign currency (100 EUR, rate 1 USD = 1 EUR)
- Create a bank transaction in the foreign currency for less than the
  bill total, at a rate making the amount higher in company currency
  (90 EUR, 108 USD at 1 EUR = 1.2 USD)
- Open the bank reconciliation widget, reconcile transaction and bill
- Unfold the reconciled transaction

Issue:

A bill reference is missing

Analysis:

The exchange difference line is reconciled with the transaction line.
Being reconciled with two lines (bill and exch), the widget relies
on `*_reconciled_lines_excluding_exchange_diff` to show the fields.
However, `count_reconciled_lines_excluding_exchange_diff`
has been defined as boolean instead of int, faulting the comparison.

opw-6479015

Forward-Port-Of: odoo/odoo#283855

When a relational property has a domain the server cannot evaluate, opening its definition editor crashes. The editor calls search_count on that domain to show how many records match, both when it opens and on every later render. The server raises a ValueError and the call has no error handling, so the whole editor goes down. The bad domain stays saved on the property, so reopening the editor fails the same way and the property can no longer be edited or deleted.

Wrap the search_count call in _updateMatchingRecordsCount (property_definition.js) in a try/catch and show no count when it fails. This is the only place the editor counts matching records, so guarding it here handles a bad domain from any source, the field selector or the code editor. The field selector still shows its warning on the invalid path, so the user can fix or delete the property.

Steps to reproduce:
1. On a model that has a Properties field, add a Many2one property and set its Model to a model that itself has a Properties field.
2. Open the property Domain and click New Rule.
3. In the field selector pick the Properties entry, then close the selector.

=> An error dialog appears and the property can no longer be edited or deleted.

Ticket [link](https://www.odoo.com/odoo/project.task/6101311)
opw-6101311

Forward-Port-Of: odoo/odoo#283802
Forward-Port-Of: odoo/odoo#259886

Steps to reproduce:
- Call your softphone with your smartphone.
- Let the smartphone ring forever. => At some point, you reach the voicemail and the softphone stops ringing but: - The softphone displays a weird message with emojis. - The call record stays "Trying to call" (and might *display* "Ended unexpectedly" in 19.2).

This commit fixes those two issues by showing a proper "Call missed" message and switching the call record status to "Missed". This is done in 19.2 and not before... because the behavior before is even more problematic:
- 19.1: the softphone keeps ringing and crash if you try to answer, that was mostly fixed thanks to [1] and this commit takes profit of those big ameliorations to fix the issue here.
- 19.0: you do not even reach voicemail (it stops without saying anything). This was apparently fixed as a side-effect of [2], which we do not consider worth even partially backporting as not critical.

[1]: https://github.com/odoo/enterprise/commit/942f32316ab02d8c739fe7fdd5ec2bdde472a68e
[2]: https://github.com/odoo/enterprise/commit/d24d7f3406ca47e7ac529d69957b0ad481d553bf

task-6453500

Forward-Port-Of: odoo/enterprise#127075

Currently, an error occurs when the user tries to update suggested products.

**Steps to Reproduce:**
 - Install the `website_sale` module.
 - Go to `Settings` and enable `Automate suggested products` under the `eCommerce` section.
 - Go to `Website` > `eCommerce` > `Products` > `Products`.
 - Create a `product` and, in the `eCommerce` tab, add a `category`.
 - Make sure the `product` is `not published`.
 - On the `product`, click the `gear icon` and select `Update suggested products`.

`ValueError: TypeError('unsupported operand types in: product.template() | None') while evaluating 'records.action_update_suggested_products()'`

When the user updates suggested products, the system updates the product's suggested 
products - optional, accessory, and alternative products [1]. While updating the alternative 
products [2], the system tries to find products based on the categories and attributes shared 
with the current product [3]. When retrieving products from the current product's category, 
it gets None [4] because the products linked to that category are unpublished and are 
therefore excluded by the domain [5]. Later, using this None value raise the error.

This commit ensures that when accessing a missing key in the category dictionary, then it falls 
back to an empty product recordset.

[1]: https://github.com/odoo/odoo/blob/21e3310a3eba2ce915b33cf10ee4c14bdda7aad4/addons/website_sale/models/product_template.py#L346

[2]: https://github.com/odoo/odoo/blob/21e3310a3eba2ce915b33cf10ee4c14bdda7aad4/addons/website_sale/models/product_template.py#L384-L386

[3]: https://github.com/odoo/odoo/blob/21e3310a3eba2ce915b33cf10ee4c14bdda7aad4/addons/website_sale/models/product_template.py#L471-L483

[4]- https://github.com/odoo/odoo/blob/21e3310a3eba2ce915b33cf10ee4c14bdda7aad4/addons/website_sale/models/product_template.py#L482

[5]- https://github.com/odoo/odoo/blob/21e3310a3eba2ce915b33cf10ee4c14bdda7aad4/addons/website_sale/models/product_template.py#L445-L456

sentry-7660509183

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Example of steps:

- Create a vehicle in Fleet
- Set 2 drivers: one with a start and end date, one with no end date
- Go back to the form view of the vehicle
- Click on the smart button "Drivers History" again
- Add an end date manually (without using the calendar), e.g. 062626
- Hit the Enter key, then save (won't work if you don't hit enter)
- Go back to the form view of the vehicle
- Enter the "Drivers History" again

=> The change has not been saved.

Also reproducible in Sales -> Milestone -> Deadline field

`onInputKeydown` closes the picker on Escape and on Enter the same way: by calling `saveAndClose()` directly, without first calling `updateValueFromInputs()` to parse the raw text typed in the input into the reactive `pickerProps.value`.

This is fine for Escape (its purpose is to discard the input), but not for Enter: when the calendar popover was never opened (i.e. the value was typed by hand instead of picked visually), `saveAndClose()` calls `apply()` directly, which only pushes `pickerProps.value` to `onApply`. Since that value was never refreshed from the input's text, `apply()` sees no change and silently returns without calling `onApply`, so the typed value is lost.

Every other confirmation path (`onInputChange`, the popover's `onClose`, and `Ctrl+Enter`) already calls `updateValueFromInputs()` before proceeding, so plain Enter was the only path missing it.

To fix this we call `updateValueFromInputs()` before `saveAndClose()` in the Enter case, like every other confirmation path already does.

opw-6511435

**Description of the issue/feature this PR addresses:**

When selecting Unsplash images from a relational field (which uses `CustomMediaDialog`), the images would fail to process or link correctly to the target record.

In the web editor, Unsplash image selections are handled by patching the base `MediaDialog`. When a user selects an Unsplash image, that patch intercepts the save action and uses the `unsplash` service to notify the backend. The server then fetches the external Unsplash URL and converts it into a native `ir.attachment` record before the frontend completes the save.

Because `CustomMediaDialog` is a distinct component used for relational fields, it bypassed the existing `MediaDialog` patch entirely and lacked this specialized fetch-and-convert logic.

This commit introduces a parallel patch specifically for `CustomMediaDialog`. It intercepts `imageSave`, routes any Unsplash records through the `unsplash` service, and replaces the raw Unsplash records with the newly generated Odoo attachments before executing the underlying save.

**Steps to reproduce:**

- POS > Products > Products > choose any product > click the ‘edit’ button in the image > search something, e.g. ‘burger’ > add Unsplash Access Key and Application ID when prompted > select one of the resulting Unsplash images

**Current behavior before PR:**

- Error when saving an unsplash image when editing product images

**Desired behavior after PR is merged:**

- No error when saving an unsplash image when editing product images

opw-6445843

**Steps to Reproduce:**
- Install the Time Off app and Belgian Payroll (l10n_be_hr_payroll).
- Create and validate an accrual allocation for currently logged-in user with: No end date Any start date An accrual plan configured with a carry-over milestone Any Time Off Type
- Create another allocation for the same employee, using the same accrual plan and same Time Off Type, but with a different start date that is not in the future.
- Select the accrual plan. The error is raised immediately during the onchange.

**Issue:**
- When the accrual allocation onchange computes the accrued balance, a temporary allocation is used internally to simulate the accrual computation.
- This temporary record is discarded after the computation.
- The discarded temporary record can remain pending for computed field recomputation.
- When the onchange later triggers recomputation, the stale temporary NewId can cause: `KeyError: <NewId origin=18>`

**Root Cause:**
- Temporary allocations created with 'new(origin=allocation)' can add computed fields to `env.transaction.tocompute`.
- `invalidate_recordset()` clears the temporary record's cache but does not remove the `NewId` from the pending recomputation queue.
- Since the temporary record has no database row to recompute from, the NewId can later be picked up during recomputation.
- This can lead to a KeyError when the framework tries to access cached data for the discarded temporary record.

**Solution:**
- Properly discard temporary allocations after the accrual simulation.
- In addition to invalidating the cache, remove the temporary NewId from `env.transaction.tocompute` using `remove_to_compute()`.
- Use this cleanup for temporary allocations created with `new(origin=...)`.

**Result**
- Prevents the KeyError during accrual allocation onchange.
- Allows users to create another allocation with the same accrual plan without triggering the RPC error.

**opw-6390559**

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Forward-Port-Of: odoo/odoo#283772

Before this commit, MobileWebSuite.test_unit_mobile failed at random on the mail test "HtmlMail add icon and save inline html", which taps the save button of an html_mail field and waits for the write:

    [verifySteps] expected the following steps
    > Expected: [
      "web_save",
    ]
    > Received: []

This happens because a tap held longer than `LONG_TAP_DELAY` counts as a long press, and `_pointerUp` then dispatches no click at all. That delay spans the synthetic pointerdown and pointerup of a single `click()`, so the work a listener does during the press counts towards that delay: tapping save blurs the html field, which converts the whole editor content to inline styles, and under the load of a runbot build that conversion alone takes longer than the delay.

One solution could have been to raise `LONG_TAP_DELAY`, but a listener slower than the new value suppresses the click again.

This commit counts a tap as long only when the test releases the pointer itself, with `pointerUp`, so the work a listener does during a `click()` no longer suppresses the click.

https://runbot.odoo.com/odoo/error/946581

Forward-Port-Of: odoo/odoo#285640
Forward-Port-Of: odoo/odoo#284971

### Issue before this commit:
When a CFDI invoice was sent through a mass action (Send & Print executed by OdooBot), the unit of measure in the generated CFDI appeared in English instead of the configured language (e.g. Spanish.

### Steps to reproduce the issue:
1.Download Accounting, Contacts and l10n_mx
2. Switch to Spanish (MX) language
3. Set the language of "Inmoviliaria CVA" and "XENON INDUSTRIAL ARTICLES" to Spanish (MX)
4. In contacts select Archived in filters and switch OdooBot language to Spanish (MX)
5. Create and confirm two invoices in the database, one for each customer but don't send these invoices
6. Go to the list view and select these two invoices, and click on "Send and print" and select CFDI
7. Check any of the XML files generated in any of the invoices (in the CFDI tab of the invoice)
8. See that the UoM in the XML file, will be set in english rather than Spanish (MX)

### Cause of the issue:
The context under which the batch action runs does not contain the lang key. As a result, translated fields (such as product_uom_id.name) were read in the source language (English) instead of the executing user's language, because nothing in the CFDI generation chain explicitly forced the correct lang into the context.

### Reason to introduce the fix:
A CFDI must always report translated fields in the correct language. The fix ensures the invoice is read with the executing user's language when the context doesn't already specify one, so translated fields are consistently correct across both flows.

opw-6399860

Forward-Port-Of: odoo/enterprise#129829
Forward-Port-Of: odoo/enterprise#125698

### Problem

When an AI agent includes a document source whose underlying `ir.attachment`
has `res_field` set (e.g., an invoice PDF generated from a Sale Order),
triggering **Update With AI** from **Website → Site → Optimize SEO**
raises a `KeyError` in `_build_rag_context`.

### Steps to Reproduce

1. Open the **AI** app.
2. Configure the **Odoo Agent**.
3. Add a source → **Add From Documents**.
4. Select a document whose underlying `ir.attachment` has `res_field` set
   (e.g., an invoice PDF generated from a Sale Order).
5. Go to **Website → Site → Optimize SEO**.
6. Click **Update With AI**.
7. Observe the following error:

```
KeyError: 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'
```

[Video](https://drive.google.com/file/d/1AAg0AvC3TRQzMLhI9hWFuSarOTHzQE-b/view?usp=sharing)

### Root Cause

In `ai/models/ai_agent.py`, `_build_rag_context()` retrieves the
`ai.agent.source` records corresponding to the embeddings by searching on the
attachment checksum:

```python
agent_sources = self.env["ai.agent.source"].search([
    ("attachment_id.checksum", "in", embeddings_attachment_checksums),
    ("agent_id", "=", self.id),
])
```

This domain traverses `attachment_id.checksum`, which internally calls
`ir.attachment._search()`.

As part of the standard attachment search behavior,
`ir.attachment._search()` automatically injects a
`('res_field', '=', False)` filter unless:

- `skip_res_field_check=True` is set in the context,
- the domain explicitly references `id` or `res_field`, or
- `bypass_access` is enabled.

```python
domain = Domain(domain)
if (
    not self.env.context.get("skip_res_field_check")
    and not any(d.field_expr in ("id", "res_field") for d in domain.iter_conditions())
    and not bypass_access
):
    disable_binary_fields_attachments = True
    domain &= Domain("res_field", "=", False)
```

[Reference](https://github.com/odoo/odoo/blob/19.0/odoo/addons/base/models/ir_attachment.py#L630)

Because of this implicit filter, attachments with `res_field` set are excluded
from the search. Consequently, `agent_sources` does not contain all the sources
corresponding to the retrieved embeddings.

Later, `_build_rag_context()` builds a checksum-to-source mapping:

```python
source_map = {
    source.attachment_id.checksum: source
    for source in agent_sources
}

for embedding in similar_embeddings:
    checksum = embedding.attachment_id.checksum
    agent_source = source_map[checksum]
```

Since `source_map` is built from the incomplete `agent_sources` recordset, it is
missing entries for attachments filtered by `ir.attachment._search()`.
However, `similar_embeddings` still contains embeddings for those attachments.

As a result, the lookup:

```python
agent_source = source_map[checksum]
```

raises a `KeyError`.

### Solution

Bypass the implicit `res_field` filter when searching `ai.agent.source`:

```python
agent_sources = (
    self.env["ai.agent.source"]
    .with_context(skip_res_field_check=True)
    .search([
        ("attachment_id.checksum", "in", embeddings_attachment_checksums),
        ("agent_id", "=", self.id),
    ])
)
```

This ensures that all `ai.agent.source` records matching the requested
attachment checksums are returned, including those referencing attachments with
`res_field` set. As a result, `source_map` contains all expected entries and
`_build_rag_context()` no longer raises a `KeyError`.

opw-6379816

Forward-Port-Of: odoo/enterprise#125780

Steps to reproduce
------------------
1. Open a restaurant PoS with a preparation display.
2. Add a product, put the internal note "A" on the line, press Send.
3. Change the note from "A" to "B", then open the note again and click "discard", that will clear it.
4. Press Send.

-> The line is fully cancelled on the preparation display and the same quantity is sent again as a new order.

Expected Behavour
-----------------
The existing line should be updated in-place instead of cancelling and re-creating a new one.

Why it's happening
------------------
In `_process_preparation_changes` we calculate a key for every line and the `note` is inside this key.

For the order lines we take `line.note or "[]"`, but for the note history we take `note['new'] or ''`, in that case, the keys don't match, so we cancel the current one and create a new line!!

The fix
-------
We use `"[]"` now, in many places (for completness), a follow up of 484ef2e8b2b.

opw-6467346

Forward-Port-Of: odoo/enterprise#128601

Steps to reproduce:
- Restaurant config with two POS devices on the same session
- Device A: open a table, add two products, press Order (the order is synced as a draft, its lines get server ids)
- Device A: remove both lines, without syncing
- Device B: touch the same order, so device A re-reads it from the server through the synchronisation websocket
- Device A: pay and validate the order

Issue:
The order is saved as paid, with its total and its payment, but without any orderline. The lines are deleted on the server: odoo.models.unlink: deleted pos.order.line records with IDs: [...]

Cause:
Removing an orderline queues an unlink command in
models.commands['pos.order'].unlink['lines_<order id>'] (delete_ in related_models.js). That command is only discarded by clearCommands(), which syncAllOrders() calls after a successful sync, so it stays pending in between.

Any read of the order in that window puts the line back: a deleted record counts as missing in missingRecursive(), so it is fetched again and loadData() re-creates it with its server id.

serialize() then emits both the update of the live line and the still pending unlink, and sync_from_ui writes
lines: [[1, id, {...}], [3, id]]. The ORM applies commands in order, and pos.order.line.order_id is ondelete='cascade', so Command.UNLINK deletes the line right after writing it.

Fix:
Skip a removal command when the record is still linked to the parent at serialization time. A record cannot be both linked and removed in the same payload, so the pending command is stale and dropping it keeps the local state. Genuine removals, where the record is no longer linked, are still sent.

opw-6401146

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Forward-Port-Of: odoo/odoo#285427
Forward-Port-Of: odoo/odoo#284695

In this PR, https://github.com/odoo/odoo/pull/186250, the `product_uom` field was renamed to `product_uom_id`. However, in the `delivery` module, `product_uom_id` is dropped from the values when the delivery line is created. 
This causes `product_uom_id` to be recomputed.

This commit reintroduces `product_uom_id` in the values to prevent the field from being recomputed.

**Description of the issue/feature this PR addresses:**

For a strange reason, when a module inherits from `sale.order.line` and adds some computed fields with `precompute=True`. `price_unit`, `price_subtotal`, and `price_total` are computed incorrectly.

I have attached a module to demonstrate the issue.

https://github.com/user-attachments/assets/ebdd8695-c9d8-477b-b5cf-ba6d8d41e84a

Without this change, the test fails, and Odoo incorrectly recomputes the fields, as shown in the video.
<img width="1232" height="515" alt="image" src="https://github.com/user-attachments/assets/27370f1f-e7b7-4102-a606-0181b9d1a97a" />

When the ORM computes fields marked as `precompute=True`, in this function `_add_precomputed_values` https://github.com/odoo/odoo/blob/0d44f26d9b0fb1c1a5db463cf1f8dd0d3c72ba26/odoo/orm/models.py#L4836, `price_unit` is 0, but the records get `price_unit` from the product. Therefore, when [_compute_amount](https://github.com/odoo/odoo/blob/0d44f26d9b0fb1c1a5db463cf1f8dd0d3c72ba26/addons/sale/models/sale_order_line.py#L855) is called, the values are computed with an incorrect `price_unit`.
<img width="1087" height="940" alt="image" src="https://github.com/user-attachments/assets/92feadf0-7f93-4acc-8f1a-931db0655fcc" />

**Steps to reproduce the issue:**

- Install the attached module. [sale_precompute.zip](https://github.com/user-attachments/files/31265993/sale_precompute.zip)
- Configure a delivery carrier as free for orders over 1, and set the fixed price to 5, for example.
- Create a sales order and add a product with a value greater than 1.
- Add the shipping method. The price should be 0. In the sales order line, `price_unit` is 0, but `price_subtotal` and `price_total` are equal to 5 (the product's sale price).

For more context, this module is a simple example extracted from the OCA `product_secondary_unit` module, which adds a mixin with these fields: https://github.com/OCA/product-attribute/blob/18.0/product_secondary_unit/models/product_secondary_unit_mixin.py. In the `sale_order_secondary_unit` module, `sale.order.line` inherits from this mixin.

You can see the error in this PR: https://github.com/OCA/sale-workflow/pull/4535.
https://github.com/OCA/sale-workflow/actions/runs/32259842816/job/96090194021?pr=4535#step:8:509

I understand that this requires a deeper investigation into precompute to solve the underlying issue, but I propose setting `product_uom_id` in the `_prepare_delivery_line_vals` method as a temporary solution while the final solution is being investigated.

I understand that this field should not have been removed from `_prepare_delivery_line_vals`; the referenced PR only renamed the field and did not intend to remove the value from this method.
@Tecnativa @pedrobaeza @kcv-odoo @Feyensv coudl you please review this?

---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Forward-Port-Of: odoo/odoo#283551

**Description of the issue/feature this PR addresses:**

`weeknumber` takes a `first_week_day` override, documented as `(0 = Monday, ..., 6 = Sunday). If None, derived from the locale.`, but tests it with `if not first_week_day`, so Monday (`0`) is discarded and the locale's own day is used.

`resource` passes `int(get_lang(self.env).week_start) - 1`, which is exactly `0` for Monday, at three call sites. #259600 added the parameter for the opposite case, a Monday locale with a Sunday-start user, and that direction works; this one never has.

**Current behavior before PR:**

With `en_US` (first day Sunday) and week start set to Monday, `resource_resource.py:299` builds `week_start_date` from Monday while `:303` buckets by Sunday. Monday 2026-08-24 to Sunday 2026-08-30 comes back as W35 for six days and W36 for the Sunday, so one week's hours are split across two buckets and the `hours_per_week` cap lands on the wrong days.

**Desired behavior after PR is merged:**

The override is honoured and those seven days are all W35.

I checked this with a partition property: every day of a `first_week_day`-aligned week must share one `(year, week)`. Over 9 locales x 7 values of `first_week_day` x 12 years, 275940 combinations, 5 locales failed and every failure was in the `first_week_day=0` bucket. After the change there are none.

I also diffed every output before and after, 315360 combinations of locale, `first_week_day` and date: 4460 rows change, all of them `first_week_day=0` on a non-Monday locale. `None` and values 1 to 6 are byte identical, so the change is contained to the broken case.

Targeting 19.0 because that is the oldest branch carrying the parameter; 17.0 and 18.0 do not have it.

AI-assisted: I used Claude to sweep the parameter space and write the tests.

Forward-Port-Of: odoo/odoo#285706
Forward-Port-Of: odoo/odoo#285508

Issue
-----
When adding all of the catalog's suggestions at once, the uom specified on the product's vendor lines is not respected.

Steps to reproduce
-----
- Create a product
	- Add a vendor line in a different uom
- Create a past outgoing shipment of 100 of the product
- Create a PO (same vendor as vendor line)
	- Open the Catalog
	- Click "Add All" in the suggestion section on the left
- Go back to the PO

> The product is in units instead of the different uom

Cause
-----
Clicking the button calls `action_purchase_order_suggest` in Python directly

https://github.com/odoo/odoo/blob/3aec1c317aad547b1ad0c85a21cc53f735c5cbfd/addons/purchase_stock/models/purchase_order.py#L131-L134

whereas clicking on a product will go through `addProduct` in JS

https://github.com/odoo/odoo/blob/3aec1c317aad547b1ad0c85a21cc53f735c5cbfd/addons/purchase_stock/static/src/product_catalog/record/kanban_record.js#L20-L28

Which leads to `_update_order_line_info` where the purchase line is created

https://github.com/odoo/odoo/blob/f59783d43f891fea14d03f67d7c017d7ffbbe5a7/addons/purchase/models/purchase_order.py#L1322-L1328

The uom is then retrieved in the create call through `_suggest_quantity`

https://github.com/odoo/odoo/blob/f59783d43f891fea14d03f67d7c017d7ffbbe5a7/addons/purchase/models/purchase_order_line.py#L547-L549

Other issue
-----
The quantity of the product also doesn't match the suggestion since the product `suggested_qty` is in the product's uom and not the vendor ones.

-----
Ticket:
opw-6417665

Forward-Port-Of: odoo/odoo#284841
Forward-Port-Of: odoo/odoo#280934

Before this commit, a build whose unit tests suite fails collects a second runbot error on top of the real one:

    odoo.addons.web.tests.test_js.WebSuite.test_unit_desktop.browser:
    Error received after termination: [MEMINFO] tests done (after GC)
    - used: 220535896 - total: 341990868 - limit: 4395630592

Note that the figures themselves are fine (220MB used of a 4.4GB limit): the error is not about memory, only about when the line is logged.

This happens because the runner logs that line after `stop()`, and `stop()` is what reports the suite result: the server settles the test on that report and kills the browser. The final major GC outlasts the report. A passing build escapes because the browser dies within milliseconds; a failing one stays up for the screenshot, long enough for the GC to finish and the line to land after termination.

This commit fixes the issue by moving `stop()` after the final cleanups and their memory log: the [MEMINFO] line is still logged, but before the result report, while the server still waits on the browser.

https://runbot.odoo.com/odoo/error/229655

Forward-Port-Of: odoo/odoo#284957

**Problem:**
Opening a Project Update activity from the activity menu (the clock icon in the systray) lists every project update of every user, instead of the ones carrying the current user's activities.

**Steps to reproduce:**
1. Schedule an overdue activity on a project update
2. Have a colleague create another project update with no activity
3. Click the clock icon in the systray
4. Under "Project Update", click the "Late" count

**Current behavior:**
The list opens unfiltered and shows all project updates, including the ones of other users and the ones carrying no activity at all.

**Expected behavior:**
The list shows only the updates carrying the user's own activities, restricted to the bucket that was clicked.

**Cause of the issue:**
The activity menu never builds a "my activities" domain. `openActivityGroup` in `mail/static/src/core/web/activity_menu.js` narrows the generic act_window it opens purely through context keys — `search_default_filter_activities_my` plus `search_default_activities_overdue` / `_today` / `_upcoming_all` — and the only domain it forwards comes from `_get_activity_groups`, which is limited to `[('active', 'in', [True, False])]`. A `search_default_<name>` key is resolved against a filter of that name in the model's search view, and is silently dropped when no such filter exists. `project.update`'s search view never declared them, so every key the menu sends is discarded and the action opens on an empty domain.

**Fix:**
`project.project` and `project.task` — the addon's two other `mail.activity.mixin` models — already declare this block of invisible activity filters, as does every other model reachable from the activity menu. Declaring them on `project.update` is what makes the menu's context keys resolvable, and keeps the model consistent with the rest of the codebase instead of special-casing `project.update` on the client side.

opw-6416397

Forward-Port-Of: odoo/odoo#281511

### Issue:
A warning `IndexError: tuple index out of range` is logged on RunBot when upgrading from 18.0 to 19.0 for employees without any version

### Cause:
`_get_version` on `hr.employee` falls back to `versions[0]` when no version matches the given date
But if the employee has no versions at all, `versions` is empty and `versions[0]` raises an `IndexError`

### Steps to reproduce:
-  Create a database with `hr_attendance` installed in 18.0
- Upgrade to 19.0

Before the fix, a warning is logged during the upgrade

### Notes:
An employee without any version is a transient state during upgrade
`hr/saas~18.4.1.1/end-migrate.py` backfills a version for every employee still missing one (`current_version_id IS NULL`) once the whole upgrade chain is done
Since it only runs at the very end, an employee can still be found without a version by earlier steps (e.g. modules reloading their demo data), which is when this warning was logged
Confirmed on an upgraded RunBot database that employees without a version before the upgrade do end up with one once fully completed

runbot-241187

Forward-Port-Of: odoo/odoo#284467

Steps to produce:
---
- Create a company with `Argentina` as the country.
- Switch to the newly created company.
- Create a new website for the `Argentina company`.
- Create a pricelist with a `23%` discount on the sales price for all products.
- Create a new product, set the sale price to 100, remove all tax, and publish.
- Go to the website and switch to the newly created Argentina company website.
- Go to the Shop page and open the product.

Issue:
---
- The tax-excluded price (`Precio s/Imp. Nac.`) displayed on the shop catalog card differs from the tax-excluded price displayed on the product detail page.

Root cause:
---
- In `_get_additionnal_combination_info`, [1] returns the unit price with the pricelist discount already applied. However, when `combination_info['has_discounted_price']` [2] is `True`, the method applies the discount again manually, resulting in the pricelist discount being applied twice on the product detail page.

Solution:
---
- Remove the redundant discount calculation block. This ensures that the tax-excluded price displayed on the product detail page matches the price shown on the shop catalog card.

[1]https://github.com/odoo/odoo/blob/7f9560bd0ff66882459593a2d043c0197ceec0bb/addons/l10n_ar_website_sale/models/product_template.py#L61-L62
[2]https://github.com/odoo/odoo/blob/7f9560bd0ff66882459593a2d043c0197ceec0bb/addons/l10n_ar_website_sale/models/product_template.py#L71-L74

opw-6480531
---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Forward-Port-Of: odoo/odoo#283761

Steps:
1. Turn on AvaTax for a company (connect it to Avalara).
2. In Settings > General Settings > Invoicing, set the Document Layout to Boxed (Box), just for a clear view.
3. Create a customer invoice:
   - Use the customer with an AvaTax fiscal position and proper address details.
   - Add a product with an AvaTax category.
   - Add a section line (a divider/heading row).
   - Compute Taxes for the AvaTax.
4. Print the invoice as a PDF.
5. Look at the section/divider row. Its cell border does not line up with the other rows. On longer invoices, this can also cause a big blank gap at the bottom of a page.

Cause:
When an invoice is computed by AvaTax, the Taxes column is hidden from the PDF (this is intentional, since AvaTax shows tax differently).

But a different part of the same template decides how wide the section row should be. This part does not know the Taxes column was hidden, so it still counts it. That makes the section row's width wrong by one column, which breaks the table layout in the PDF.

Fix:
Instead of hiding the Taxes column only in one place, fix it at the source: the variable that decides (does this invoice show a Taxes column) now also checks if the invoice is an AvaTax invoice. Every other part of the template that uses this variable (the header, the tax cells, and the section row width) automatically gets the right answer, since they all read from the same place.

Result:
- Section rows on AvaTax invoices now have the correct width.

opw - 6455674

Forward-Port-Of: odoo/enterprise#129097

Steps to reproduce:
----------------------------------------
1. Install `whatsapp_sign` and Contact modules
2. Go to Sign > templates
3. Make two sign requests for the same partner
4. Go to partner record and try to change the email (remove or add a character)

Observation:
----------------------------------------
Traceback occurs:
```
  File '/home/odoo/odoo/enterprise/whatsapp_sign/models/sign_request_item.py', line 101, in _send_signature_access_message
    is_whatsapp = self.sign_request_id.send_channel == 'whatsapp'
  File '/home/odoo/odoo/community/odoo/orm/fields.py', line 1657, in __get__
    record.ensure_one()
  File '/home/odoo/odoo/community/odoo/orm/models.py', line 5942, in ensure_one
    raise ValueError('Expected singleton: %s' % self)
ValueError: Expected singleton: sign.request(3, 4)
```

Issue:
----------------------------------------
In `_send_signature_access_message()` which accessed `self.sign_request_id.send_channel` directly on the full multi-record recordset. When `res_partner.write()` detects an email change, it searches for all sent sign request items of that partner and calls `send_signature_accesses()` on the combined recordset. Accessing `sign_request_id` on this multi-record set resolved to multiple sign requests, and the subsequent field access triggered `ensure_one()`.

Solution:
----------------------------------------
* Refactored `_send_signature_access_message()` to iterate over each item individually, checking `send_channel` per item's own `sign_request_id`. Items are partitioned into `email_items` and `whatsapp_items` batches. Email items are delegated to `super()` and WhatsApp items are processed individually.

* Refactored to cleaner and more readable code and correctly handles mixed-channel recordsets where items may belong to different sign requests with different send channels.

opw-6387586

Forward-Port-Of: odoo/enterprise#128953
Forward-Port-Of: odoo/enterprise#125075