Created attachment 205657 [details] Java code demonstrating the problem I have a Java program (attached) that starts a LO process with --accept=socket option, connects to it using a socket and uses loadComponentFromURL to open an ODT file. This is NOT headless mode, the user then uses LO to do its work on the opened file. Doing this with up to LO 25.2 works perfectly, but LO 25.8 and 26.2 are unstable when started that way. Frequent crashes, even more frequent visual glitches from menus and other GUI elements. It happens with any ODT files that I've tried, all of which can be opened without problems when LO is started without the --accept=socket option - the one I am testing with has been created in LO 26.2 and is completely empty. Some crash dumps are created under %USERPROFILE%\AppData\Local\CrashDumps but I'm missing the pdb files (especially mergedlo) to analyse them (sorry...)
Created attachment 205658 [details] ODT file
Steps: 1. Download attachment (LOTest.java) and attachment (blank.odt). 2. Edit LOTest.java: 2.2. Line 19: change ODT_FILE constant to point to your blank.odt (for me, it became "C:\\Downloads\\blank.odt"). 2.1. Line 21: change SOFFICE_EXE constant, the first element of cmdArray, into a path to your LO. 3. Build and Run the result, using the libreoffice.jar found in your LO installation. E.g., my command lines were: build to C:\Downloads > javac.exe -cp "C:\Program Files\LibreOffice\program\classes\libreoffice.jar" C:\Downloads\LOTest.java -d C:\Downloads\build Run > java.exe -cp "C:\Downloads\build;C:\Program Files\LibreOffice\program\classes\libreoffice.jar" LOTest You may need 2 ou 3 executions to reproduce, but the crashes are very frequent. And sometimes, no crash but the GUI becomes erratic (and may crash after a while) By the way, disabling Skia may help for GUI visual glitches on 25.8, but it's not an option anymore on 26.2 Feel free to ask for anything I can do to help for this bug investigation !
Created attachment 205659 [details] Crash dump Analyse with WinDBG, without libreoffice symbols, from 26.2.1.1 LO Crash
Confirmed here. Regression - LO 24.8.7.2- : OK - LO 25.8.5.2 : KO - LO 26.2.1.2 : KO
- LO 25.2.7.2 : OK
Thanks for this confirmation! @(In reply to Antoine POURCHEZ from comment #4) > Confirmed here. Regression > - LO 24.8.7.2- : OK > - LO 25.8.5.2 : KO > - LO 26.2.1.2 : KO
It has probably been fixed. It's Start Version: 25.8.0.0.alpha0+ commit b5510122fd68e3d16117684f51062dfb03961b85 remove mpThreadGraphics from GDI backend It's Fixed. Version: 26.8.0.0.alpha0+ commit 90004b730eda4d06313f6e14603f1c1dad0f8f88 tdf#153554 fix crash when running python ODK tests
I can confirm it's fixed in the last daily master daily/master/Win-x86_64@tb103-1-TDF/2026-03-19_08.27.30/LibreOfficeDev_26.8.0.0.alpha0_Win_x86-64.msi No crash, no erratic UI, just a (maybe not related) truncation problem on the right and the bottom of the screen, fixed when resizing the window. Thanks for the investigation, sorry I didn't try the 26.8 alpha before !
I also confirm that it's OK in LE 26.8 ! Thank you for the correction. Is a backport possible in LO 25.2.7 (8.7?)? Otherwise, LO 25.2.7 has been EOL since November 2025, there is currently no non-EOL version of LO where this problem is not present.
@Saburo : Can a developer confirm that no port to an earlier version of LO 25.8.7 is planned for this fix, and therefore, LO25.8+ is completely unusable via batch tools (python or other)?
As of today, it appears that backports to versions 26.2 and 25.8 have not yet been released. I don't know if there is a plan.
The patch has been applied. 26.2.4 25.8.8