https://bugs.kde.org/show_bug.cgi?id=522261

--- Comment #5 from Carlos Acevedo <[email protected]> ---
Hi, Jean-Baptiste.
Unfortunately, the original reproduction scenario is still present. The
development build starts, becomes unresponsive during startup, and
eventually generates an APPCRASH. The new executable has a different
timestamp and different crash offsets, confirming that the new build was
indeed tested.

I should mention that I am not a programmer, I am actually a musician, haha
:)
This investigation took me quite a while because I had to learn many things
along the way, but if my testing and reports can help improve Kdenlive, I
am very happy to contribute.



This is the Validation Report, in case it is useful for you:




# Validation Report – Development Build 12941

## Validation of the proposed fix for the Windows startup crash

---

# Test environment

* Development Build: 12941
* Operating System: Windows 11
* User profile: HP Z2
* Display configuration: Three HP E221 monitors
* Original reproduction scenario restored before testing

---

# Baseline

Before executing the development build, the original configuration that
consistently reproduced the crash was restored.

Relevant configuration:

```ini
[sdl]
project_monitor_fullscreen=1:CNC5040FTX
```

This is the same configuration used during the original investigation.

---

# Validation objective

Determine whether the development build eliminates the original startup
crash under exactly the same conditions that reproduced the bug in the
released version.

---

# Test result

## Original reproduction scenario

Result:

**FAIL**

Observed behavior:

* Kdenlive starts.
* Startup does not complete.
* The application becomes unresponsive.
* An APPCRASH is eventually generated.

---

# Windows Error Reporting

The development build generates a new crash report.

Main observations:

* Exception: `0xC0000005`
* Fault module: `kdenlive.exe`
* New executable timestamp
* New fault offset:
  `0x000000000079A890`

This confirms that the tested executable is the new development build and
not the previous release.

---

# WinDbg observations

The new dump still reports an:

```
INVALID_POINTER_READ
```

The call stack continues to involve the same general family of operations
observed during the original investigation, including:

```
QWidget::setParent
QQuickWidget::event
QObject::~QObject
```

The offsets naturally differ because this is a different executable, but
the overall execution pattern remains very similar.

At this point we cannot conclude whether the failure occurs in exactly the
same internal function, since no symbols are available for the development
build.

However, the observed execution pattern remains consistent with the
previous investigation.

---

# Workaround validation

After the failed startup test, the previously discovered workaround was
restored:

```ini
[sdl]
project_monitor_fullscreen=
```

Result:

**PASS**

Kdenlive starts normally.

The application remains usable.

No startup crash occurs.

---

# Validation matrix

Stable Release

```
project_monitor_fullscreen=1:CNC5040FTX
→ APPCRASH

project_monitor_fullscreen=
→ Starts normally
```

Development Build 12941

```
project_monitor_fullscreen=1:CNC5040FTX
→ APPCRASH

project_monitor_fullscreen=
→ Starts normally
```

---

# Experimental conclusions

The following statements are supported by reproducible testing.

* The original reproduction scenario remains valid.
* Clearing `kdockLayout` alone does not eliminate the crash.
* Restoring the original fullscreen configuration reproduces the crash in
the development build.
* Clearing only `project_monitor_fullscreen` prevents the crash both in the
stable release and in Development Build 12941.
* The previously discovered workaround remains fully effective.
* The overall WinDbg stack remains consistent with the original
investigation, particularly around `QWidget::setParent`,
`QQuickWidget::event`, and `QObject::~QObject`.

Based on these observations, the proposed patch does not resolve the
original reproduction scenario.

---

# Additional observations

During the source-code analysis performed throughout the investigation, we
identified a number of possible instrumentation ideas that might help
isolate the exact point where the invalid pointer originates.

These ideas were **not experimentally validated**, because generating an
instrumented Windows build would have required setting up a complete
KDE/Craft development environment, while the priority shifted to validating
the development build you kindly provided.

They are therefore included only as possible debugging directions, not as
conclusions or proposed fixes.

Possible instrumentation ideas include:

* Delaying the automatic call to `slotSwitchFullScreen()` until the user
interface has completed initialization.
* Instrumenting `Monitor::Monitor()` and `slotSwitchFullScreen()`.
* Logging object state immediately before and after:

  * `m_glWidget->setParent(nullptr)`
  * `showFullScreen()`
* Recording:

  * parent hierarchy
  * `window()`
  * `quickWindow()`
  * screen index
  * screen serial number
  * monitor ownership
* Temporarily isolating the effects of `setParent(nullptr)` and
`showFullScreen()` independently to determine which operation ultimately
leads to the invalid pointer observed by WinDbg.

These are only potential debugging experiments that emerged from the code
analysis and are provided in case they are useful during further
investigation.

---

# Final conclusion

The development build successfully changes the executable under test (new
timestamp and new offsets), confirming that the proposed patch was
evaluated.

However, under the exact same conditions that originally reproduced the
bug, the application still crashes during startup.

The previously identified workaround continues to prevent the crash.

>From the QA perspective, the original reproduction scenario is therefore
**not yet resolved**.

We hope these results help narrow down the remaining issue, and we would be
happy to perform additional validation on future development builds if
needed.







El dom, 28 jun 2026 a las 5:27, Jean-Baptiste Mardelle (<
[email protected]>) escribió:

> https://bugs.kde.org/show_bug.cgi?id=522261
>
> Jean-Baptiste Mardelle <[email protected]> changed:
>
>            What    |Removed                     |Added
>
> ----------------------------------------------------------------------------
>              Status|REPORTED                    |NEEDSINFO
>          Resolution|---                         |WAITINGFORINFO
>
> --- Comment #4 from Jean-Baptiste Mardelle <[email protected]> ---
> Thanks for your detailed report. I don't have a Windows multiple monitor
> system
> to test right now, but pushed a possible fix in master.
> Could you try the latest development builds that contain my fix to confirm
> if
> this fixes the problem ?
> Try installing the daily build from here:
> https://origin.cdn.kde.org/ci-builds/multimedia/kdenlive/master/windows/
>
> Thanks in advance for your help.
>
> --
> You are receiving this mail because:
> You reported the bug.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to