[Libreoffice-commits] core.git: sw/qa sw/source

[Michael Stahl (via logerrit)]

 sw/qa/extras/layout/data/fdo43573-2-min.docx |binary
 sw/qa/extras/layout/layout.cxx               |   16 ++++++++++++++++
 sw/source/core/layout/fly.cxx                |   14 +++++++++-----
 3 files changed, 25 insertions(+), 5 deletions(-)

New commits:
commit ed12269c42f75f553bb8a8770923406f7824e473
Author:     Michael Stahl <michael.stahl at allotropia.de>
AuthorDate: Wed May 5 12:21:42 2021 +0200
Commit:     Michael Stahl <michael.stahl at allotropia.de>
CommitDate: Wed May 5 16:55:23 2021 +0200

    tdf#142080 sw: layout: fix infinite loop in CalcContent()
    
    On page 9, SwSectionFrame::Format() calls CalcContent() and that formats
    all its content frames, then on SwTextFrame 91
    SwObjectFormatter::FormatObj() fails becuase it moved forward.
    
    With commit c799de145f7e289f31e3669646e5bd12814e6c5e this now sets the
    o_rbPageHasFlysAnchoredBelowThis to true, which prevents a call to
    SwLayouter::InsertMovedFwdFrame(), and the flys anchored in next frames
    aren't moved off the page at this time.
    
    Then the loop starts over at the beginning of the SwSectionFrame, and
    frame 91 will be formatted again because the loop tries to format the
    first frame on the next page to see if it will move back; now the
    MoveBwd() isn't prevented any more so the result is the same failure
    in SwObjectFormatter::FormatObj().
    
    Fix this by ignoring the bRestartLayoutProcess in case the current frame
    was originally on the next page and didn't move back (or, as is the case
    here, moved back and then forward again); it should just be formatted
    again on the next page.  Once that happens, it will eventually be
    entered into SwLayouter::InsertMovedFwdFrame() too.
    
    This happens to fix another problem with this bugdoc too: the first
    column of the section on page 9 is empty.  This also happens in LO 6.4
    but not LO 6.1.
    
    An alternative would be to move the flys anchored below off the page as
    is done in SwLayAction::FormatContent() now but sections can be in flys
    or footnotes or headers so perhaps it should be done only at the
    top-level.
    
    (regression from c799de145f7e289f31e3669646e5bd12814e6c5e)
    
    Change-Id: I0965aebb4e3cec687f4e70f8d5e3aa8a55da3393
    Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115144
    Tested-by: Jenkins
    Reviewed-by: Michael Stahl <michael.stahl at allotropia.de>

diff --git a/sw/qa/extras/layout/data/fdo43573-2-min.docx b/sw/qa/extras/layout/data/fdo43573-2-min.docx
new file mode 100644
index 000000000000..429b7948ed02
Binary files /dev/null and b/sw/qa/extras/layout/data/fdo43573-2-min.docx differ
diff --git a/sw/qa/extras/layout/layout.cxx b/sw/qa/extras/layout/layout.cxx
index ddb17e6d9c30..88dd06078ec2 100644
--- a/sw/qa/extras/layout/layout.cxx
+++ b/sw/qa/extras/layout/layout.cxx
@@ -1306,6 +1306,22 @@ CPPUNIT_TEST_FIXTURE(SwLayoutWriter, testTdf116486)
     CPPUNIT_ASSERT_EQUAL(OUString("4006"), aTop);
 }
 
+CPPUNIT_TEST_FIXTURE(SwLayoutWriter, TestTdf142080)
+{
+    // this caused an infinite loop
+    load(DATA_DIRECTORY, "fdo43573-2-min.docx");
+
+    xmlDocUniquePtr pLayout = parseLayoutDump();
+    // check the first paragraph on page 9 with its fly; the colum was empty too
+    assertXPath(pLayout, "/root/page[9]/body/section[1]/column[1]/body/txt[1]/Text[1]", "Portion",
+                "De kleur u (rood) in het rechtervlak (R), de kleur r (wit) beneden (D),");
+    SwTwips nPage9Top = getXPath(pLayout, "/root/page[9]/infos/bounds", "top").toInt32();
+    assertXPath(
+        pLayout,
+        "/root/page[9]/body/section[1]/column[1]/body/txt[1]/anchored/fly[1]/notxt/infos/bounds",
+        "top", OUString::number(nPage9Top + 1460));
+}
+
 CPPUNIT_TEST_FIXTURE(SwLayoutWriter, testTdf128198)
 {
     SwDoc* pDoc = createDoc("tdf128198-1.docx");
diff --git a/sw/source/core/layout/fly.cxx b/sw/source/core/layout/fly.cxx
index 84e52c6de46c..989870bfcb92 100644
--- a/sw/source/core/layout/fly.cxx
+++ b/sw/source/core/layout/fly.cxx
@@ -1466,6 +1466,7 @@ void CalcContent( SwLayoutFrame *pLay, bool bNoColl )
         do
         {
             pLast = pFrame;
+            bool const wasFrameLowerOfLay(pLay->IsAnLower(pFrame));
             if( pFrame->IsVertical() ?
                 ( pFrame->GetUpper()->getFramePrintArea().Height() != pFrame->getFrameArea().Height() )
                 : ( pFrame->GetUpper()->getFramePrintArea().Width() != pFrame->getFrameArea().Width() ) )
@@ -1607,7 +1608,10 @@ void CalcContent( SwLayoutFrame *pLay, bool bNoColl )
 
                 // #i28701# - restart layout process, if
                 // requested by floating screen object formatting
-                if ( bRestartLayoutProcess )
+                if (bRestartLayoutProcess
+                    // tdf#142080 if it was aleady on next page, and still is,
+                    // ignore restart, as restart could cause infinite loop
+                    && (wasFrameLowerOfLay || pLay->IsAnLower(pFrame)))
                 {
                     pFrame = pLay->ContainsAny();
                     pAgainObj1 = nullptr;
@@ -1671,10 +1675,10 @@ void CalcContent( SwLayoutFrame *pLay, bool bNoColl )
                         pFrame->InvalidatePos_();
                 }
             }
-          // Stay in the pLay
-          // Except for SectionFrames with Follow: the first ContentFrame of the Follow
-          // will be formatted, so that it gets a chance to load in the pLay.
-          // As long as these Frames are loading in pLay, we continue
+          // Stay in the pLay.
+          // Except for SectionFrames with Follow: the first ContentFrame of the
+          // Follow will be formatted, so that it gets a chance to move back
+          // into the pLay.  Continue as long as these Frames land in pLay.
         } while ( pFrame &&
                   ( pLay->IsAnLower( pFrame ) ||
                     ( pSect &&