14208 - [MLK]: Memory leak of 16384 bytes from 1 block allocated in orkinHeap::Alloc(nsIMdbEnv *,DWORD,void * *)

Status: VERIFIED FIXED

(MailNews Core :: Database, defect, P3)

Description

[Suresh]

Overview Description: I found this Memory Leak on my recent Purify runs. I think
16 KB leak is big leak. I have already filed a bug# 10262 for Memory leak in
orkinHeap::Alloc. But the stack trace and the number of bytes looks different.

Davidmc, I remember you saying "MDB is designed to leak factory and environment
objects on purpose". If you think that this bug is related to your comment,
please mark this bug as Invalid. Sorry for the inconvenience.

Here is the Stack trace.

[W] MLK: Memory leak of 16384 bytes from 1 block allocated in
orkinHeap::Alloc(nsIMdbEnv *,DWORD,void * *)
    Distribution of leaked blocks
         16384 bytes from 1 block of 16384 bytes (0x08894de8) allocation number
379101
    Allocation location
        new(UINT)      [memcmp.asm:23]
        orkinHeap::Alloc(nsIMdbEnv *,DWORD,void * *) [orkinHeap.cpp:55]
                {
                  MORK_USED_1(ev);
                  mdb_err outErr = 0;
             =>   void* block = new char[ inSize ];
                  if ( !block )
                    outErr = morkEnv_kOutOfMemoryError;

        morkWriter::MakeWriterStream(morkEnv *) [morkWriter.cpp:240]
                            mWriter_Bud = bud;
                            stream = new(*heap, ev)
                              morkStream(ev, morkUsage::kHeap, heap, bud,
             =>                 morkWriter_kStreamBufSize, frozen);
                          }
                          else
                            bud->CutStrongRef(ev->AsMdbEnv());
        morkWriter::WriteMore(morkEnv *) [morkWriter.cpp:328]
                    if ( this->IsWriter() )
                    {
                      if ( !mWriter_Stream )
             =>         this->MakeWriterStream(ev);

                      if ( mWriter_Stream )
                      {
        ???            [ip=0x08853348]
        morkThumb::DoMore_Commit(morkEnv *) [morkThumb.cpp:464]
                  morkWriter* writer = mThumb_Writer;
                  if ( writer )
                  {
             =>     writer->WriteMore(ev);
                    mThumb_Total = writer->mWriter_TotalCount;
                    mThumb_Current = writer->mWriter_DoneCount;
                    mThumb_Done = ( ev->Bad() || writer->IsWritingDone() );
        ???            [ip=0x08853348]
        morkThumb::DoMore(morkEnv *,DWORD *,DWORD *,BYTE *,BYTE *)
[morkThumb.cpp:364]
                        this->DoMore_SessionCommit(ev); break;

                      case morkThumb_kMagic_CompressCommit: // 7 /* store method
*/
             =>         this->DoMore_CompressCommit(ev); break;

                      case morkThumb_kMagic_SearchManyColumns: // 8 /* table
method */
                        this->DoMore_SearchManyColumns(ev); break;
        orkinThumb::DoMore(nsIMdbEnv *,DWORD *,DWORD *,BYTE *,BYTE *)
[orkinThumb.cpp:225]
                  if ( ev )
                  {
                    morkThumb* thumb = (morkThumb*) mHandle_Object;
             =>     thumb->DoMore(ev, outTotal, outCurrent, outDone, outBroken);
                    outErr = ev->AsErr();
                  }
                  return outErr;
        nsMsgFolderCache::Close(void) [nsMsgFolderCache.cpp:426]
                        mdb_bool outBroken = PR_FALSE;     // is operation
irreparably dead and broken?
                        while (!outDone && !outBroken && ret == NS_OK)
                        {
             =>             ret = commitThumb->DoMore(GetEnv(), &outTotal,
&outCurrent, &outDone, &outBroken);
                        }
                        if(commitThumb)
                            commitThumb->CutStrongRef(m_mdbEnv);

Comment 1

[David :Bienvenu]

Wrong David - davidmc does mdb/mork - but you know, I fixed this a while ago, I
thought. David, did you crunch my change by accident? It was a ref-counting
problem - I can look for the e-mail I sent you.

Comment 2

[davidmc]

Yes it looks like I whacked your change as the result of some kind
of automerging in CVS, which got part of your fix and part of my
fix.  I coded a slightly different version of the fix.  My file
morkThumb.cpp now looks like this:

morkThumb::Make_LargeCommit(morkEnv* ev,
nsIMdbHeap* ioHeap, morkStore* ioStore)
{
  // ...
  nsIMdbFile_SlotStrongFile(file, ev, &outThumb->mThumb_File);

  outThumb->mThumb_Writer = writer; // pass writer ownership to thumb
  // ...
}

When it should only have the second assignment:


  // ...
  outThumb->mThumb_Writer = writer; // pass writer ownership to thumb
  // ...

I thought this was better the incrementing the refcount in the call
to nsIMdbFile_SlotStrongFile(), and then decrementing it immediately.

Comment 3

[David :Bienvenu]

this was fixed a while ago.

Comment 4

[Suresh]

I could see this leak in my recent purify posting at
www.mozilla.org/mailnews/purify.  Take a look at '18-November 1999 POP' log and
search for 'orkin'. Reopening the bug.

Comment 5

[Suresh]

Clearing resolution.

Comment 6

[David :Bienvenu]

I'll look at fixing this.

Comment 7

[David :Bienvenu]

I think the problem here is that the same change needs to be made to compress
commit. I'll check in a fix when the tree opens.

Comment 8

[David :Bienvenu]

fixed in morkThumb, as I indicated above.

Comment 9

[[e-mail] (gone - use [e-mail] instead)]

I haven't seen this stack since I've been running Purify.