Hi Brandon,
OK, I see the problem now. It happens when a Tree has a basket created
before the file has reached 2 GBytes and is written when the file
is above 2 Gbytes. The fix is now in CVS.
Thanks for reporting this problem and sorry for the long time it took
to identify the cause.
Rene Brun
Brandon Kohn wrote:
>
> Yes the int will be negative, but the test is:
>
> if(testValue < 0 && fVersion <= 1000)
>
> fVersion should be 1003 if the seekKey is over 2GB. and this test should
> fail. Otherwise you are writing a key out with a version marking it as below
> the 2GB transition when the seekKey is greater (overflows the int). This is
> definately an error.
>
> Brandon
>
> ----- Original Message -----
> From: "Rene Brun" <Rene.Brun@cern.ch>
> To: "Brandon Kohn" <blk@maia-institute.org>
> Cc: "Roottalk List" <roottalk@pcroot.cern.ch>
> Sent: Thursday, August 19, 2004 11:16 AM
> Subject: Re: test program on linux
>
> > Hi Brandon,
> >
> > As I already reported to you, I cannot reproduce your problem.
> >
> > Note that if you modify TKey as you suggest, YOU MUST GET AN ERROR.
> > The line
> > Int_t testValue = (Int_t)fSeekKey;
> > is not correct. Testing testValue<0 will automatically happen when
> > your file reaches 2 GBytes.
> >
> > Rene Brun
> >
> > > Brandon Kohn wrote:
> > >
> > > Hello Rene,
> > >
> > >
> > > Here is a test program/makefile bit for linux that will reproduce the
> problem
> > > I'm seeing. Have you had a chance to run any more tests?
> > >
> > > Thanks much!
> > >
> > > Brandon
> > >
> > > ---Below is a message that bounced back when I tried to post on
> roottalk.
> > >
> > > Hello rooters,
> > >
> > > I've been experiencing some sporadic trouble with key writing on files
> over
> > > 2 gigabytes, and I would like to get some confirmation from an external
> > > party that the bug I'm seeing is indeed happening on other machines.
> > >
> > > In order to see the bug, you must make a small change to TKey.cxx (a
> simple
> > > conditional test with a message output to a file and stdout.)
> > >
> > > here is the diff for my TKey.cxx from the current cvs HEAD:
> > >
> > > RCS file: /user/cvs/root/base/src/TKey.cxx,v
> > >
> > > retrieving revision 1.44
> > >
> > > diff -r1.44 TKey.cxx
> > >
> > > 937a938,950
> > >
> > > > ///This will allow you to check when the error occurs with a debug
> break
> > > or file output.
> > >
> > > > ///HACK: try to correct incorrectly identified seekKeys.
> > >
> > > > Int_t testValue = (Int_t)fSeekKey;
> > >
> > > > if(testValue < 0 && fVersion <= 1000)
> > >
> > > > {
> > >
> > > > std::cout << "It happened!" << std::endl;
> > >
> > > > std::ofstream ofs("c:/errorfileroot.txt",std::ios::app);
> > >
> > > > ofs << "File pointer using int when long64." << std::endl;
> > >
> > > > ofs << "File end value: " << gFile->GetEND() << std::endl;
> > >
> > > > ofs << "File name: " << gFile->GetName() << std::endl;
> > >
> > > > ofs << "On object: " << this->GetName() << std::endl;
> > >
> > > > ofs.close();
> > >
> > > > }
> > >
> > > Or simply replace the TKey::Streamer method with the following...
> > >
> > >
> //_______________________________________________________________________
> > > void TKey::Streamer(TBuffer &b)
> > > {
> > > //*-*-*-*-*-*-*-*-*Stream a class object*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> > > //*-* =========================================
> > > Version_t version;
> > > if (b.IsReading()) {
> > > b >> fNbytes;
> > > b >> version; fVersion = (Int_t)version;
> > > b >> fObjlen;
> > > fDatime.Streamer(b);
> > > b >> fKeylen;
> > > b >> fCycle;
> > > if (fVersion > 1000) {
> > > b >> fSeekKey;
> > > b >> fSeekPdir;
> > > } else {
> > > Int_t seekkey, seekdir;
> > > b >> seekkey; fSeekKey = (Long64_t)seekkey;
> > > b >> seekdir; fSeekPdir= (Long64_t)seekdir;
> > > }
> > > fClassName.Streamer(b);
> > > fName.Streamer(b);
> > > fTitle.Streamer(b);
> > > } else {
> > > b << fNbytes;
> > > ///This will allow you to check when the error occurs with a debug
> break or
> > > file output.
> > > ///HACK: try to correct incorrectly identified seekKeys.
> > > Int_t testValue = (Int_t)fSeekKey;
> > > if(testValue < 0 && fVersion <= 1000)
> > > {
> > > std::ofstream ofs("c:/errorfileroot.txt",std::ios::app);
> > > ofs << "File pointer using int when long64." << std::endl;
> > > ofs << "File end value: " << gFile->GetEND() << std::endl;
> > > ofs << "File name: " << gFile->GetName() << std::endl;
> > > ofs << "On object: " << this->GetName() << std::endl;
> > > ofs.close();
> > > fVersion += 1000;
> > > }
> > > version = (Version_t)fVersion;
> > > b << version;
> > > b << fObjlen;
> > > if (fDatime.Get() == 0) fDatime.Set();
> > > fDatime.Streamer(b);
> > > b << fKeylen;
> > > b << fCycle;
> > > if (fVersion > 1000) {
> > > b << fSeekKey;
> > > b << fSeekPdir;
> > > } else {
> > > b << (Int_t)fSeekKey;
> > > b << (Int_t)fSeekPdir;
> > > }
> > > fClassName.Streamer(b);
> > > fName.Streamer(b);
> > > fTitle.Streamer(b);
> > > }
> > > }
> > >
> > > After making this change, recompile root.. and then compile and run the
> test
> > > that I've included. The test simply creates a TFile and begins filling
> it
> > > with both large and small instances of a CBinaryLargeObject class (which
> is
> > > included). It it essential that you run the case exactly as I have
> included
> > > it as I've found that the problem only occurs with a mixture of size
> types
> > > for the Binary Large Objects.
> > >
> > > I've included a tar.gz'd version of the example that hopefully
> illustrates
> > > the bug...
> > >
> > > Please let me know if you are able to replicate this.
> > >
> > > --Kindest Regards
> > >
> > > ____________________________________________________________
> > > Brandon Kohn Tel.: +377 97 97 41 51
> > > Software Developer
> > > The Maia Institute Fax.: +377 97 97 41 59
> > > Le Patio Palace
> > > ____________________________________________
> > >
> > > Important Notice
> > >
> > > This email and any attachments to it are confidential and intended
> solely
> > > for the individual(s) to whom they are addressed. Any modification or
> > > dissemination of the contents of this e-mail is strictly prohibited
> unless
> > > expressly authorised by the sender. If you receive this e-mail by
> mistake,
> > > please advise the sender immediately by using the reply facility in your
> > > e-mail software. Please also delete the message from your computer, and
> > > destroy any paper copies. Thank you for your co-operation.
> > >
> > > Name: testbigfile.tar.gz
> > > testbigfile.tar.gz Type: Unix Tape Archive (application/x-tar)
> > > Encoding: base64
> >
This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:09 MET