richterm [Sun, 17 Aug 2008 20:40:22 +0000 (20:40 +0000)]
enhanced robustness of object extraction from data buffer, buffer which are obviously too small to hold a TObject are ignored to avoid crashes in the ROOT streamer functions
hristov [Fri, 15 Aug 2008 16:38:00 +0000 (16:38 +0000)]
- AliMUONRecoParam.cxx:
Fill in the "event specie" variable according what the set of
parameters is designed for, as defined in AliRecoParam:
- AliMUONESDInterface.h, AliMUONESDInterface.cxx,
AliMUONTrackReconstructor.cxx, AliMUONTrackReconstructorK.cxx,
AliMUONRecoCheck.h, AliMUONRecoCheck.cxx, MUONRefit.C,
MUONAlignment.C, AliEveMUONData.cxx, AliEveMUONTrack.cxx:
Use the Kalman filter + smoother to recompute track parameters at
each clusters when recovering a MUONTrack from an ESDTtrack.
You can use the static method ResetTracker(...) to change it by
providing a new AliMUONRecoParam object to the interface.
cvetan [Fri, 15 Aug 2008 16:12:05 +0000 (16:12 +0000)]
Nasty bug. The directory is not initialized if the event index is -1. In fact it is initialized later in NextEvent(). Thanks to Matthias for spotting it the problem.
cvetan [Fri, 15 Aug 2008 13:36:09 +0000 (13:36 +0000)]
ESD file is added to the list of proof output files. It is then automatically merged on the master and arrives on the client as a single file. For the moment the default location of the resulting ESD file is the local folder from which the proof session was started.
richterm [Fri, 15 Aug 2008 09:29:08 +0000 (09:29 +0000)]
TRD Update by Konstantin:
- Proper geometry loading (TGeoManager => AliGeomManager)
- Consistent HLT Logging scheme is now used.
- Possibility to load CosmicTest recoParam is added AliHLTTRDClusterizer:
- Avoid coping data in DoEvent() AliHLTTRDTrackerV1Component
- Now we PushBack TClonesArray of TRDTrackV1 instead of ESD
cvetan [Thu, 14 Aug 2008 17:31:18 +0000 (17:31 +0000)]
1. Most of the initialization code is put to Begin method. Corresponding changes in the input list of TProof and the selector. 2. Some rearanged sys-info logs. 3. gGeoManager pointer is reset in the default contructor of AliReconstruction. That is avoid the problem on the proof master with the double input list.
richterm [Thu, 14 Aug 2008 15:12:38 +0000 (15:12 +0000)]
AliHLTTask: improve consistency check for generated output data (allow ptr be set to the output buffer); AliHLTDataBuffer: allow data block descriptors of zero length buffers
abercuci [Thu, 14 Aug 2008 13:07:10 +0000 (13:07 +0000)]
continue on fixing the memory leak. Here is the result of valgrind which
appears not to scale with the number of new/delete !
==21155== 2,664 bytes in 9 blocks are possibly lost in loss record 169
of 223
==21155== at 0x4C22F75: operator new[](unsigned long) (in
/usr/lib64/valgrind/amd64-linux/vgpreload_memcheck.so)
==21155== by 0x1A989CD7: TStreamerInfo::Compile()
(TStreamerInfo.cxx:1497)
==21155== by 0x1A98F67B: TStreamerInfo::Build()
(TStreamerInfo.cxx:433)
==21155== by 0x1948B7D6: TClass::GetStreamerInfo(int) const
(TClass.cxx:2933)
==21155== by 0x1A98E500: TStreamerInfo::Build()
(TStreamerInfo.cxx:251)
==21155== by 0x1948B7D6: TClass::GetStreamerInfo(int) const
(TClass.cxx:2933)
==21155== by 0x1A98E500: TStreamerInfo::Build()
(TStreamerInfo.cxx:251)
==21155== by 0x1948B7D6: TClass::GetStreamerInfo(int) const
(TClass.cxx:2933)
==21155== by 0x1A98E500: TStreamerInfo::Build()
(TStreamerInfo.cxx:251)
==21155== by 0x1A93CC1D: TBufferFile::WriteClassBuffer(TClass*,
void*) (TBufferFile.cxx:3355)
==21155== by 0x10070965: AliTRDtrackV1::Streamer(TBuffer&)
(G__TRDrec.cxx:2223)
==21155== by 0x193E4AA4: TDirectory::CloneObject(TObject const*,
bool) (TDirectory.cxx:263)
cvetan [Thu, 14 Aug 2008 07:11:27 +0000 (07:11 +0000)]
Possibility to instantiate CDB manager from prefetched cache. It is used on the proof slaves in order to use CDB entries loaded on the client side and send via the proof input list.
cholm [Wed, 13 Aug 2008 21:47:43 +0000 (21:47 +0000)]
The line
if (edep<=0) continue;
in AliFMDBaseDigitizer::DigitizeHits was introduced - wrongly - by
Peter Hristov. It _will_ break the digitisation and the following
reconstruction.
The behviour of the digitisation models exactly the front-end as it
should (no matter what memory consumption it may entail). The check
should be on zero suppression, since that's what models the front-end -
if zero suppression is turned on in the front-end, then we can suppress
empty digits - otherwise we shoud never do that. Note, that the line
affects _both_ normal digitisation and digitisation for summable digits,
since the condition is on the energy deposition and not on the actual
number of counts. If this line should go anywhere, it should be in
the possible overloaded AliFMDSDigitizer::AddDigit - not here.
Perhaps this change should be propogated to the release ASAP?