Currently when the user reverts an inventory adjustment move line with a package the package contains extra line showing negative quantity of the product.

## Steps to produce:
- Install Inventory without demo data
- Settings Enable 'Packages'
- Create a product:
   - Cheese burger
   - On hand > Create a new quant
   - Package: 'Burgerbox' and 'On Hand Quantity`: 1 and save
- Set the On Hand quantity to zero and save
- History > Revert the Inventory adjustment line from WH/stock to Inventory adjustment by selecting it and reverting via actions.
- Products > Packages > BurgerBox

## Observed Behaviour:
After reverting an inventory adjustment that set the product's physical quantity to 0, the package contains two lines for the same product with quantities 1 and -1.

This is inconsistent because a package should not contain a product with a negative quantity.

The package should be restored to its original state and contain only the expected positive quantity.

## Root cause:
When the user reverts the move line, `action_revert_inventory` is called. This method creates the revert move and then marks that move as done at [1].

Marking the move as done subsequently marks all related move lines as done at [2]. During this process, the system first unreserves the quantity from the virtual location / inventory adjustment and then removes the quantity from that location (resulting in a -1 quantity move line at that location). This is performed through `_synchronize_quant`, which is responsible for synchronizing the physical inventory with the move line at [3].

The `_synchronize_quant` method uses the move line's `package_id` when updating the corresponding quant at [4]. As a result, `_update_available_quantity` creates a new quant with the following values at [5]:
```
{
    'product_id': 1,
    'location_id': 14,
    'lot_id': stock.lot(),
    'package_id': 1,
    'owner_id': res.partner(),
    'in_date': datetime.datetime(2026, 6, 22, 12, 42, 11),
    'quantity': -1.0,
}
```
This creates a quant with a negative quantity that is linked to the package because `package_id` is set on the newly created quant. Consequently, the move line with the negative quantity becomes associated with the package.

[1]-
https://github.com/odoo/odoo/blob/333c279be7cba9fda17d2818be36f4d96d8cd0f6/addons/stock/models/stock_move_line.py#L1016-L1035 
[2]-
https://github.com/odoo/odoo/blob/333c279be7cba9fda17d2818be36f4d96d8cd0f6/addons/stock/models/stock_move.py#L1956 [3]-
https://github.com/odoo/odoo/blob/333c279be7cba9fda17d2818be36f4d96d8cd0f6/addons/stock/models/stock_move_line.py#L662-L666 
[4]-
https://github.com/odoo/odoo/blob/333c279be7cba9fda17d2818be36f4d96d8cd0f6/addons/stock/models/stock_move_line.py#L678-L687 
[5]-
https://github.com/odoo/odoo/blob/333c279be7cba9fda17d2818be36f4d96d8cd0f6/addons/stock/models/stock_quant.py#L1130-L1143
## Solution:
Remove the source `package_id` when creating revert moves for inventory adjustment locations.

When an inventory adjustment sets a product's quantity to 0, the adjustment is completed without a destination package, meaning the product is effectively removed from the package. Therefore, the corresponding revert move should not retain the package as its source. Keeping the package as the source is inconsistent because package information should not exist on a virtual inventory adjustment location, and the original inventory adjustment removes the product from the package (there is no destination package).

By removing the source `package_id` from the revert move, the system avoids creating negative quants associated with the package during quant synchronization. This also ensures that, after the inventory adjustment is reverted, the quantities of products inside the package are restored correctly and match their state prior to the adjustment.

opw-6285739

Steps to reproduce:
- Create a vendor bill, one journal item: no product, price 100, no tax -> save
- Change the account of the line to one having a default included tax
- Remove the suggested tax -> save

Issue:
The tax is back on the line, applied on top of the price (100 -> 117.36), making the included tax look like an excluded one.

Cause:
`_inverse_account_id` is the inverse of `account_id` and its onchange, and retriggers `_compute_tax_ids` on every write of `account_id`. When the user removes the tax the onchange just suggested, javascript `_getChanges` does not identify the user-deleted tax and sends back only the new account_id <-> `tax_ids` is back to its server value. So the field is not protected by `_get_protected_vals` and the compute reapplies the default taxes of the account, overriding the user.

Solution:
Split the method in two: the onchange keeps suggesting the default taxes (when the user keeps them, the client sends `tax_ids` explicitly), the real inverse only keeps the analytic part.
The create path is not impacted: `tax_ids` is precomputed from the account, so imports of new bills/entries still get the default taxes.

opw-6088331

### Steps to reproduce

- Install `sale_timesheet_enterprise`.
- Go to **Tasks → All Tasks**.
- Switch to the **Gantt** view.
- Group by **Sale Order Item** (`sale_line_id`).

### Issue

A traceback is raised when loading the Gantt view with group by `sale_line_id`:

```text
ValueError: Invalid field 'planned_hours' on model 'project.task' for 'planned_hours:sum'
```

### Root cause

The Gantt progress bar computation for the `sale_line_id` grouping performs a `_read_group()` aggregation on the `planned_hours` field of `project.task`.

However, `planned_hours` was renamed to `allocated_hours` during the saas-16.5 migration, so the former field no longer exists on `project.task`. As a result, the aggregation raises a `ValueError`.

Migration reference:
https://github.com/odoo/upgrade/blob/e638c6ce00d9d8936d034ad7130fef51565b9195/migrations/project/saas~16.5.1.2/pre-migrate.py#L10

Issued PR: https://github.com/odoo/enterprise/pull/49685

### Fix

Use `allocated_hours`, the renamed equivalent of `planned_hours`, when computing the Gantt progress bar. This restores the Gantt view when grouping tasks by **Sale Order Item** and prevents the traceback.

### Summary

When `dev_mode` includes `reload`, `ThreadedServer`'s FSWatcher reacts to a file change by sending the process a `SIGHUP` to trigger a phoenix restart. `signal_handler` turns `SIGHUP` into `KeyboardInterrupt`, which `ThreadedServer.run()`'s wait-loop catches. The catch is too narrow — a reload `SIGHUP` can kill the process through **three** windows that all sit outside the wait-loop's `try/except`, so the exception escapes `run()`/`main()`. Under Docker's default `restart: no`, PID 1 dies and the container stays down.

### The three windows

1. **Teardown duplicate (exit 130).** One file change can emit several FS events; the FSWatcher's `if not odoo.phoenix:` dedup races across threads and fires more than one `SIGHUP`. The first begins the phoenix teardown; the second lands during `stop()` / `watcher.stop()` / `_reexec()` and `KeyboardInterrupt` escapes.
2. **Exec-gap (exit 129).** `os.execve()` resets caught signal handlers to their default disposition (`SIGHUP` terminates) but preserves `SIG_IGN`; a `SIGHUP` arriving after the exec but before the re-exec'd process re-installs its handler kills the process outright.
3. **Startup (exit 130).** In the re-exec'd process, a `SIGHUP` anywhere in the startup section that precedes the wait-loop — `start()`, `preload_registries()` **and** `cron_spawn()` — escapes `run()`.

### Reproducer (deterministic)

Boot a `ThreadedServer` (`--workers 0`) on any initialised db, then signal PID 1 a few times in quick succession:

```bash
docker exec <container> sh -c 'i=0; while [ $i -lt 8 ]; do kill -HUP 1; sleep 0.1; i=$((i+1)); done'
```

Unpatched the process exits 130 or 129. Patched it stays up after one clean phoenix reload. Verified live on 17.0 and 18.0: stock `server.py` dies; the patched `server.py` survives sustained bursts (20/20 across repeated reload cycles on each version); `SIGINT`/`SIGTERM` still exit 0.

### Fix

Minimal, in `signal_handler` + `run()` + `_reexec()`; `SIGINT`/`SIGTERM` untouched; one new instance attribute, no new module globals:

- **Teardown duplicate:** ignore a `SIGHUP` once `quit_signals_received` is set (a restart/shutdown is already pending; the re-exec reloads fresh code).
- **Startup:** a per-instance `in_preload` flag marks the entire startup section (`start()` + `preload_registries()` + `cron_spawn()`); a `SIGHUP` there sets the phoenix flag + counter and returns instead of raising, so the wait-loop exits right after startup and runs the normal restart.
- **Exec-gap:** `signal.signal(signal.SIGHUP, signal.SIG_IGN)` just before `os.execve` so a `SIGHUP` in the gap is dropped rather than terminating the process.

### Related

- #21209 (merged) — introduced the phoenix flag; did not guard these windows.
- #206898 (merged), #207930 (open) — PreforkServer reload.

### CLA

Covered by Codeforward B.V.'s corporate CLA; #269240 adds me to its contributor list (pending merge).

In Multiwebsite settings, when the public user interactions needs email generation (appointment or event flow), the email links are generated with a base url that does not corresponds to the one from which the request started.

Case 1:
- Have website A and website B
- Create an appointment page website A
- Log in via website B
- As public user, make an appointment in Website A
- Check the generated email

Issue: button links in the email will redirect to the wrong website, so users will encounter an issue when managing the appointment. This occurs because when an user log in, the system parameter 'web.base_url' is updated with the current url.
This parameter is then used as fallback when we need to retrieve the base url without an active record

Case 2:
- Have website A and website B
- Create an event and assign it to website B
- As public user, access the event and register to it
- Check the generated email

Issue: button links in the email will redirect to the wrong website, so users will encounter an issue when managing the event.
This occurs because the record `event.registration` has no website_id field and the base url is taken from the company default website (website A)

opw-4146760
opw-4336369

Description of the issue/feature this PR addresses:

Current behavior before PR:

Desired behavior after PR is merged:




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

**Steps to reproduce:**

- Create a product A, tracked by lots
- Create a kit product, include a component A
- Change the UoM to 0.5
- Go to the PoS, order this kit product
- Also order the component A, with a quantity of 2
- Pay for it, ask for an invoice
- Go to the created picking
- The Demand column is correctly computed and is 0.5
- The Quantity column is wrong and is 2

**Why the fix:**

When getting the data from https://github.com/odoo/odoo/blob/e0d84c7fbb270d0d1f82572daefa96c2978d3785/addons/point_of_sale/models/stock_picking.py#L283 
we always get the component's line, as the move's product is the component, even if it used to be the kit product's move. This is because when exploding a kit's moves, it gets the kit's component as a product instead of keeping the kit product.

This was introducing a weird behavior because we took the quantity from the component line, and not from the kit line, meaning the kit would always have the same quantity as the component.

We now check if the move is actually a kit product's move, and if it is we adapt the qty to correct one by fetching the correct line's qty, and adapting it with the correct UoM.

Changing the line in itself would not work, as the kit itself is not tracked by lots, so we would not enter https://github.com/odoo/odoo/blob/e0d84c7fbb270d0d1f82572daefa96c2978d3785/addons/point_of_sale/models/stock_picking.py#L284 
and the move line would not be correctly created.

opw-6153000

`_l10n_tr_nilvera_add_pdf_to_invoice` writes the response from `client.request('GET', '.../pdf')` directly into `ir.attachment.raw`. The Nilvera client sets `Accept: application/json` on the session and calls `response.json()` by default, so the returned value is a Python `str` holding the base64-encoded PDF body, not raw binary bytes.

The previous code wrote to the base64-aware `datas` field, which auto-decoded its input. An earlier fix switched to `raw` to work around a `binascii.Error` from Python 3.14's stricter base64 validation in the `datas` auto-decode path. That switch silently changed what ends up on disk (`datas` decodes its input, `raw` does not)

Storing that string in the binary `raw` field encodes it as UTF-8, so the file on disk ends up as the literal ASCII of the base64 text. The attachment is served as `application/pdf` but the browser receives base64 ASCII and cannot preview or download the PDF.

Call `b64decode(response)` before storing so the attachment contains the actual PDF bytes.

OPW-6302803

Description of the issue/feature this PR addresses:

Current behavior before PR:

Desired behavior after PR is merged:




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