[Libreoffice-commits] core.git: vcl/source

[Tor Lillqvist]

 vcl/source/gdi/virdev.cxx |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

New commits:
commit 6025fd8970444794e0a9dde76f12d0dd1ec8b778
Author: Tor Lillqvist <tml at iki.fi>
Date:   Sun Apr 7 10:42:01 2013 +0300

    Avoid hang on make subsequentcheck on Linux
    
    Acquire the yield mutex in the VirtualDevice destructor only in a dbgutil
    build on OS X. (I added that acquire yesterday to fix an assertion when
    exiting on OS X.) Seems to fix a hang in "make subsequentcheck" on Linux. No
    idea what is going on exactly, sigh.
    
    Change-Id: Ifcb1897720998d0725a8db0d5f8c75848c6a30f7

diff --git a/vcl/source/gdi/virdev.cxx b/vcl/source/gdi/virdev.cxx
index 559a192..2e8415d 100644
--- a/vcl/source/gdi/virdev.cxx
+++ b/vcl/source/gdi/virdev.cxx
@@ -184,9 +184,14 @@ VirtualDevice::~VirtualDevice()
 
     ImplSVData* pSVData = ImplGetSVData();
 
-    // OutputDevice::ImplReleaseGraphics() requires the yield mutex to
-    // be held, hmm.
+#if defined(DBG_UTIL) && defined(MACOSX)
+    // OutputDevice::ImplReleaseGraphics() requires the yield mutex to be
+    // held, hmm. For some reason this is a problem only in a dbgutil build on
+    // OS X, when this stuff gets called when soffice is about to exit, from
+    // editeng/source/editeng/eerdll.cxx. If this code is used on Linux, we
+    // again get a hang in some/all of the subsequentchecks.
     pSVData->mpDefInst->AcquireYieldMutex( 1 );
+#endif
 
     ImplReleaseGraphics();