https://bugs.documentfoundation.org/show_bug.cgi?id=150064

Michael Weghorn <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
                 CC|                            |[email protected]
           Assignee|[email protected] |[email protected]
                   |desktop.org                 |

--- Comment #3 from Michael Weghorn <[email protected]> ---
Confirmed, it's quite well reproducible when e.g. saving attachment 181334 as
`sd/qa/unit/sdlayouta11y.cxx` and doing a quick hack to run that along with
e.g. CppunitTest_sd_png_export_tests.mk:

diff --git a/sd/CppunitTest_sd_png_export_tests.mk
b/sd/CppunitTest_sd_png_export_tests.mk
index 015557873961..73bce5d85660 100644
--- a/sd/CppunitTest_sd_png_export_tests.mk
+++ b/sd/CppunitTest_sd_png_export_tests.mk
@@ -20,6 +20,7 @@ $(eval $(call
gb_CppunitTest_use_common_precompiled_header,sd_png_export_tests))

 $(eval $(call gb_CppunitTest_add_exception_objects,sd_png_export_tests, \
     sd/qa/unit/PNGExportTests \
+    sd/qa/unit/sdlayouta11y \
 ))

Then, with that in place, run the test in a loop until it fails:

    while make CppunitTest_sd_png_export_tests; do echo OK; done

or, recording the runs with rr so it can be replayed for debugging:

    while make CppunitTest_sd_png_export_tests RR=1; do echo OK; done

>From a first analysis, it looks like the (or at least one) problem is in
ChildrenManagerImpl::MergeAccessibilityInformation
(svx/source/accessibility/ChildrenManagerImpl.cxx), where the order of children
can change.

Might be a regression from

    commit 8f9fd6806ccfbf381a383efe5d143ead86ee49de
    Date:   Wed Jun 29 19:47:20 2022 +0200

        tdf#137544 reduce cost of ChildrenManagerImpl::Update

        there are 2 O(n^2) algorithms here, reduce them to O(n log n) by
        pre-sorting

        Change-Id: Ib3912155cda62cac95b5037528e23ef3c686a7e8
        Reviewed-on: https://gerrit.libreoffice.org/c/core/+/136655
        Tested-by: Jenkins

I'll look into it a little closer myself first to see whether it's actually a
regression and whether any good ideas come to mind rather quickly while doing
so.

Tested with LO master as of commit 28dc1e713cfc5b5ea38e15f032aba72d05e40b33.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to