Re: SegV in 5.18/00 handling long strings with --build=debug

Hi Nick,

I can reproduce the problem when compiling ROOT in debug mode and with gcc3.4.
The fact that it was working before or with the optimized mode is just pure chance. A local stack array
is overwritten in your case in v6_var.cxx. As a workaround when you compile ROOT, I suggest to change line 791 of $ROOTSYS/cint/inc/G__ci.h from
#define G__ONELINE 256 /* Length of subexpression,parameter,argument */
to
#define G__ONELINE 4096 /* Length of subexpression,parameter,argument */

The CINT gurus will add a protection and a patch to 5.18 when one will be produced.

Rene Brun

Nick West wrote:
> Dear Roottalk,
>
> This thread started privately last Thursday when I emailed Rene saying I
> thought I have found a bug in the code that was about to be tagged as
> 5.18/00. Code that work with 5.17/08 SegVed with 5.18/00. I saw the
> problem building directly from the Repository at two separate SL4
> sites. Rene took my standalone macro and ran it perfectly on all
> supported platforms. Once the release was announced I took ROOT's
> Linux SL4 binary build and confirmed that it worked too.
>
> I think I now at least understand part of the problem. It is to do
> with the configuration option:-
>
> --build=debug
>
> If I take the 5.18/00 source tar and configure with:-
>
> ./configure linux --build=debug
>
> then build and install in the usual way, then:-
>
> root bug.C
>
> SegVs. For reference my machine is:-
>
> Scientific Linux SL release 4.5 (Beryllium)
> gcc version 3.4.6 20060404 (Red Hat 3.4.6-8)
>
> The bug.C attached contains the statement:-
>
> jc.Path.Create("Spill",
> "DataQualityReader::Reco "
> "RecordSetupModule::Get "
> "NeardetBeamSelect::Ana "
> "DigitListModule::Get "
> "DigitListModule::Reco "
> "FilterDigitListModule::Reco "
> "StripSRListModule::Reco "
> "SliceSRListModule::Reco "
> "TrackCamListModule::Reco "
> "FitTrackCamListModule::Reco "
> "ClusterSRListModule::Reco "
> "SubShowerSRListModule::Reco "
> "ShowerSRListModule::Reco "
> "EventSRListModule::Reco "
> "RecordSetupModule::Reco "
> "NtpBDLite::Reco "
> );
>
> and the SegV can be fixed by reducing the size of the second string
> argument. CINT problems handling long strings has been discussed on
> this list a number of times before, what makes this different is that
> if the current code is configured without --build=debug, say
>
> ./configure linux
>
> the problem goes away. It also goes away in 5.17/08 even with
> --build=debug
> so something has changed since then that is sensitive to this option.
>
> There is a simple workaround but if people experience SegVs when
> moving forward to 5.18/00 and are using the debug option, it's
> something to bear in mind.
>
> Cheers,
>
> Nick.
>
Received on Sun Jan 20 2008 - 17:50:51 CET