Bug 72864

Created attachment 4221 [details]
test PS file which reliably exhibits this bug

I plunged into the code today.  The problem with the main viewer turned out to be caused by the following.

KPSWidget::gs_input(), which controls the parceling-out of data to GhostScript, is supposed to be called only when the KProcess _process is ready to accept some more data.  The way this is enforced is that the KProcess emits a signal when it is ready, and that signal is connected to gs_input(), which then sends the data to the KProcess pointed to by the member variable _process.  However, sometimes a stale KProcess can emit this signal although _process has been changed.  On receipt of this signal gs_input() tries to send data to _process, which isn't yet ready to accept it, and the data gets lost.

A workaround for this problem is to modify KPSWidget::gs_input() to KPSWidget::gs_input(KProcess *), which can then keep track of which KProcess sent the signal, and just do nothing if it was a KProcess other than _process.  On my system this workaround fixes the problems with the main viewer, although I still very occasionally have problems with the thumbnails.

A patch implementing this workaround will be attached.

Created attachment 4233 [details]
workaround for problem with main viewer

Subject: kdegraphics/kghostview

CVS commit by luis_pedro: 

Fixes bug 72864 by correctly identifying if it's our KProcess which is ready.

Patch by Andy Neitzke. Thanks a lot.

CCMAIL:72864-close@bugs.kde.org


  M +10 -4     kpswidget.cpp   1.138
  M +1 -1      kpswidget.h   1.76


--- kdegraphics/kghostview/kpswidget.cpp  #1.137:1.138
@@ -180,5 +180,5 @@ bool KPSWidget::sendPS( FILE* fp, unsign
     // Start processing the queue.
     if( _stdinReady )
-        gs_input();
+        gs_input(_process);
 
     return true;
@@ -371,5 +371,5 @@ bool KPSWidget::startInterpreter()
              this, SLOT( gs_output( KProcess*, char*, int ) ) );
     connect( _process, SIGNAL( wroteStdin( KProcess*) ),
-             this, SLOT( gs_input() ) );
+             this, SLOT( gs_input( KProcess* ) ) );
 
     kapp->flushX();
@@ -449,8 +449,14 @@ void KPSWidget::gs_output( KProcess*, ch
 }
 
-void KPSWidget::gs_input()
+void KPSWidget::gs_input( KProcess* process )
 {
     kdDebug(4500) << "KPSWidget: gs_input" << endl;
 
+    if (process != _process)
+    {
+      kdDebug(4500) << "KPSWidget::gs_input(): process != _process" << endl;
+      return;
+    }
+    
     _stdinReady = true;
     const unsigned bufferSize = 8192;
@@ -483,5 +489,5 @@ void KPSWidget::gs_input()
             _currentRecord->begin += bytesRead;
             _currentRecord->len -= bytesRead;
-            if( _process && _process->writeStdin( _inputBuffer, bytesRead ) )
+            if( process && process->writeStdin( _inputBuffer, bytesRead ) )
                 _stdinReady = false;
             else

--- kdegraphics/kghostview/kpswidget.h  #1.75:1.76
@@ -197,5 +197,5 @@ protected:
 
 protected slots:
-    void gs_input();
+    void gs_input( KProcess* );
     void gs_output( KProcess*, char* buffer, int len );
     void interpreterFailed();

This has been working for me since the patch was applied, but I just tried out 3.3 Beta 1 from the Onebase LiveCD and the misbehavior seems to be back -- I tried downloading a bunch of postscript/PDF files from arxiv.org, and every time, the first page appeared blank.  In fact, the first page remained blank even after I went to the second page and back to the first.  The thumbnails also appeared all blank, except for the thumbnail corresponding to the first page, which showed a thumbnail of the last page.

This behavior seems similar to what was observed a while ago in this bug report, and it seems like it might have a similar origin, so I'm reopening this bug, but I could open a new one if that's preferable.

This is not the case in Okular.