| 0ce7176 | Fons Rademakers | 29 August 2013, 14:42:02 UTC | make version v5-34-10. | 29 August 2013, 14:42:02 UTC |
| b0814f8 | Fons Rademakers | 29 August 2013, 14:40:29 UTC | Revert "UseCache()" removal, to avoid API change in patch branch. | 29 August 2013, 14:40:29 UTC |
| dd55a0f | Fons Rademakers | 28 August 2013, 13:19:55 UTC | mark ReAlloc(void*,size_t) as obsolete for v6-02-00 as it is non-safe. | 28 August 2013, 13:19:55 UTC |
| e13d913 | Fons Rademakers | 28 August 2013, 13:19:32 UTC | Finally remove TFile::UseCache() which was marked for removal by v5-32-00. | 28 August 2013, 13:19:32 UTC |
| d914270 | Fons Rademakers | 28 August 2013, 11:24:27 UTC | expand comment of ReAlloc(). Explains ROOT-5115. | 28 August 2013, 11:25:03 UTC |
| 927de44 | Olivier Couet | 28 August 2013, 10:36:19 UTC | Protection added in case the remote file is not accesible. | 28 August 2013, 10:36:19 UTC |
| 3d0f151 | Dario Berzano | 28 August 2013, 00:33:32 UTC | Silence CID 52143: don't deref NULL ptr | 28 August 2013, 09:46:46 UTC |
| 89790c9 | Dario Berzano | 28 August 2013, 00:23:17 UTC | Silence CID 52142: check dynamic_cast NULL retval | 28 August 2013, 09:46:46 UTC |
| e6585ba | Olivier Couet | 28 August 2013, 08:38:53 UTC | The option "A" in the `Draw()` was not cleaning properly the current pad. | 28 August 2013, 08:38:53 UTC |
| 0108a43 | Olivier Couet | 28 August 2013, 06:56:05 UTC | Undo previous fix, it generated an extra page in pdf files. Better fix under investigation. | 28 August 2013, 06:56:05 UTC |
| d994cae | Gerardo Ganis | 27 August 2013, 21:53:17 UTC | Fix warning on Windows | 27 August 2013, 21:53:17 UTC |
| 31b06bc | Gerardo Ganis | 27 August 2013, 21:05:55 UTC | Synchronize TPacketizerUnit with 'master' | 27 August 2013, 21:08:23 UTC |
| bb3894f | Wim Lavrijsen | 27 August 2013, 18:58:58 UTC | fix name completion for instances | 27 August 2013, 18:58:58 UTC |
| d83c1b3 | Gerardo Ganis | 27 August 2013, 15:27:19 UTC | Fix warning | 27 August 2013, 15:27:19 UTC |
| f69bcb5 | Gerardo Ganis | 27 August 2013, 15:18:00 UTC | Add missing forward declaration | 27 August 2013, 15:21:13 UTC |
| 0c16ab2 | Dario Berzano | 26 August 2013, 21:57:33 UTC | No bloating IgProf output and clearer profile type Now the tag to invoke IgProf performance profiling is "igprof-pp" and no longer "igprof" alone. Leaves some space for adding IgProf memory profiling. | 27 August 2013, 13:32:37 UTC |
| a99badd | Dario Berzano | 26 August 2013, 16:41:04 UTC | PROOF: automatic igprof-analyse for IgProf pp logs | 27 August 2013, 13:32:37 UTC |
| dad0eac | Dario Berzano | 26 August 2013, 16:40:13 UTC | Some debug in PROOF log retrieval | 27 August 2013, 13:32:37 UTC |
| 1c50887 | Dario Berzano | 26 August 2013, 12:19:34 UTC | PROOF GUI support for log filtering via pipe cmd | 27 August 2013, 13:32:37 UTC |
| 00e7d78 | Dario Berzano | 23 August 2013, 15:32:15 UTC | Added quotes for igprof cmd in PROOF-Lite | 27 August 2013, 13:32:37 UTC |
| 29f86f4 | Dario Berzano | 22 August 2013, 15:22:23 UTC | Added PROOF logfiles preprocessing before sending | 27 August 2013, 13:32:37 UTC |
| c601e3e | Dario Berzano | 21 August 2013, 22:34:03 UTC | No newline in PROOF logfiles retrieval progress | 27 August 2013, 13:32:36 UTC |
| 37f79c1 | Dario Berzano | 21 August 2013, 22:17:36 UTC | Fixed PROOF pct of completion when getting logs | 27 August 2013, 13:32:36 UTC |
| 12aea14 | Dario Berzano | 21 August 2013, 21:47:56 UTC | PROOF: tag identifying additional logfiles Tags are now used for the Valgrind and IgProf wrappers, generating additional logfiles. They are appended to the "ordinal" number both for the master line and the workers lines. Appending tags to ordinals also incidentally fixes a bug leading to having the additional logfile concatenated to the ordinary logfile while getting the first. | 27 August 2013, 13:32:36 UTC |
| ce6fa9c | Dario Berzano | 16 August 2013, 20:09:40 UTC | Started working on integrating IgProf in PROOF | 27 August 2013, 13:32:36 UTC |
| 243899f | Olivier Couet | 27 August 2013, 13:27:17 UTC | Fix https://sft.its.cern.ch/jira/browse/ROOT-4516 The option "A" in the `Draw()` was not cleaning properly the current pad. | 27 August 2013, 13:27:17 UTC |
| 292cf1f | Olivier Couet | 27 August 2013, 11:10:43 UTC | Protection added in UnZoom | 27 August 2013, 11:10:43 UTC |
| f3bbd87 | Olivier Couet | 27 August 2013, 10:55:17 UTC | provide symbolic values for different styles as suggested here: https://sft.its.cern.ch/jira/browse/ROOT-4910 | 27 August 2013, 10:55:17 UTC |
| b371aca | Olivier Couet | 27 August 2013, 08:17:48 UTC | Add mnemonic constants: kHAlignLeft=10, kHAlignCenter=20, kHAlignRight=30, kVAlignBottom=1, kVAlignCenter=2, kVAlignTop=3 They allow to write: object->SetTextAlign(kHAlignLeft+kVAlignTop) | 27 August 2013, 08:17:48 UTC |
| a05d022 | Gerardo Ganis | 25 August 2013, 15:45:52 UTC | Fix typo in dependency | 25 August 2013, 15:46:04 UTC |
| 32c0de1 | Wouter Verkerke | 25 August 2013, 09:42:37 UTC | Update version number | 25 August 2013, 09:42:37 UTC |
| 926e422 | Gerardo Ganis | 24 August 2013, 08:43:49 UTC | Use uniform naming scheme for scope rc vars Conflicts: proof/proof/src/TProofServ.cxx | 24 August 2013, 09:39:07 UTC |
| 5bdfb65 | Gerardo Ganis | 24 August 2013, 08:43:49 UTC | Fix warning found by EC | 24 August 2013, 08:43:49 UTC |
| dfe63b8 | Gerardo Ganis | 14 August 2013, 15:43:10 UTC | Master-side logic for controlled output sending | 23 August 2013, 23:29:12 UTC |
| 2c865b6 | Gerardo Ganis | 14 August 2013, 15:33:56 UTC | Fine-tune patch c56ca1b to control output sending | 23 August 2013, 23:29:07 UTC |
| 1d15023 | Gerardo Ganis | 14 August 2013, 15:31:47 UTC | Add status bit for output request | 23 August 2013, 23:29:02 UTC |
| f99558e | Gerardo Ganis | 14 August 2013, 09:02:07 UTC | Implement server-side logic for controlled output sending | 23 August 2013, 23:28:56 UTC |
| 4b7ab81 | Gerardo Ganis | 14 August 2013, 08:59:46 UTC | Add list for ready-to-send workers | 23 August 2013, 23:28:51 UTC |
| b809989 | Gerardo Ganis | 14 August 2013, 08:59:11 UTC | Add message type for output sendign control | 23 August 2013, 23:28:44 UTC |
| cb26dc6 | Philippe Canal | 23 August 2013, 19:03:41 UTC | Merge fixes related to branch offset and io rules into v5-34-00-patches | 23 August 2013, 19:03:41 UTC |
| 4fe9fae | Philippe Canal | 23 August 2013, 14:05:14 UTC | Fix fOffset calculation when encountering cached elements Update the StreamerElement search in TBranchElement::InitializeOffsets to always try to find a StreamerElement that points to one of the member of the actual class. This allows for properly finding the offset of the local object within the outer object. The previous version was somtimes looking at 'cached' StreamerElement; however their local offset are (often) different that the non-cached version and the main offset is very often different. The previous code was then applying this offset to both the actions for the cached StreamerElement and for the regular StreamerElement. For the cached element the two actions compensated each other. For the regular element if the local object had an object within the main object, this was leading to the 'reading' of the local object overwriting content in the main object. Now the offset is calculated for the local object in regard to the outer object using one the StreamerElement that points to one of the member of the actual class *and* AddToOffset is not applied to the cached object (since they are never part of an outer object the offset is always zero). This fixes ROOT-5437 | 23 August 2013, 18:44:51 UTC |
| c7e5849 | Philippe Canal | 23 August 2013, 16:27:57 UTC | Add proper support for writing in the presence of io rules. In the case a read rule applies to the current StreamerInfo, which is used for writing, there was no good StreamerElement describing the location and means to write the data member (unless there was a repating element). Now, in this case we introduce yet another StreamerElement and mark it as a 'Write' element. We also ignore it in the 'right' places and similarly ignore the caching and artificial element when writing. For example with writeTree.C in roottest/root/io/datamodelevolution/misc we get: StreamerInfo for class: MyClass, version=1, checksum=0x36b5f1f8 MyClass@@1 @@alloc offset= 0 type=1001 TObject BASE offset= 0 type=66 Basic ROOT object TObjArray fArray offset= 24 type=61 (cached) TObjArray fArray offset= 0 type=61 (write) TObjArray fArray offset= 16 type=1000 MyClass@@1 @@dealloc offset= 0 type=1002 i= 0, @@alloc type=1001, offset= 0, len=1, method=0 i= 1, TObject type= 66, offset= 0, len=1, method=0 i= 2, fArray type= 61, offset= 24, len=1, method=0 [cached] i= 3, fArray type= 61, offset= 0, len=1, method=0 [write] i= 4, fArray type=1000, offset= 16, len=1, method=0 i= 5, @@dealloc type=1002, offset= 0, len=1, method=0 | 23 August 2013, 18:44:06 UTC |
| f078d99 | Philippe Canal | 23 August 2013, 14:23:08 UTC | Extent StreamerElement printing. Add the information about whether a StreamerElement is designed for writing. Factor the collection of this information into a single routine TStreamerElement::GetSequenceType to centralize and normalize the generation of this string (instead of being partially duplicated in 4 places). | 23 August 2013, 18:43:35 UTC |
| d56754b | Olivier Couet | 23 August 2013, 14:06:33 UTC | Improve the spacing between lines. Due to the way the vertical text centring is done (bounding based) the spacing between lines may appeared irregular in some cases.. | 23 August 2013, 14:06:33 UTC |
| ae3b744 | Bertrand Bellenot | 23 August 2013, 13:21:47 UTC | Remove duplicate fTotal++; Remove duplicate fTotal++; which is already incremented in the base class (TGLVContainer) Solving a bug reported on the ROOT forum (http://root.cern.ch/phpBB3/viewtopic.php?f=3&t=16971&sid=25ac93dfe11c8ea903e260b159f49519) | 23 August 2013, 13:22:30 UTC |
| af5eeb4 | Philippe Canal | 23 August 2013, 11:50:54 UTC | Fix off by one error in out-of-bound check. Avoid reading pass the end of an array in TTreeFormula. This fixes http://root.cern.ch/phpBB3/viewtopic.php?t=16969. | 23 August 2013, 11:51:13 UTC |
| 763c322 | Olivier Couet | 23 August 2013, 11:36:44 UTC | In TMultiGraph::Add(TMultiGraph *multigraph, Option_t *chopt) If "chopt" is defined all the graphs in "multigraph" will be added with the "chopt" option. If "chopt" is undefined each graph will be added with the option it had in "multigraph". | 23 August 2013, 11:36:44 UTC |
| 6d55f86 | Olivier Couet | 22 August 2013, 14:36:52 UTC | fix https://sft.its.cern.ch/jira/browse/ROOT-5431 | 22 August 2013, 14:36:52 UTC |
| 70b6d47 | Olivier Couet | 22 August 2013, 13:52:36 UTC | Fix: https://sft.its.cern.ch/jira/browse/ROOT-4999 | 22 August 2013, 13:52:36 UTC |
| d62c92e | Pere Mato | 22 August 2013, 08:53:03 UTC | Added protection for ROOT-5428 | 22 August 2013, 08:53:03 UTC |
| f5e0a2a | Olivier Couet | 22 August 2013, 07:00:23 UTC | doc update | 22 August 2013, 07:00:23 UTC |
| ca133fb | Olivier Couet | 22 August 2013, 05:27:52 UTC | Add a protection to fix the problem described here: https://sft.its.cern.ch/jira/browse/ROOT-5429 | 22 August 2013, 05:28:30 UTC |
| 4325ad1 | Lorenzo Moneta | 21 August 2013, 20:43:16 UTC | from M. Schiller: add win32 fix to RooChebychev.cxx, clean up fix in RooMath.h | 21 August 2013, 20:43:16 UTC |
| f7e7c3a | Lorenzo Moneta | 21 August 2013, 15:02:40 UTC | Add fix for Windows (definition of __func__ is in RooMath.h) | 21 August 2013, 15:06:52 UTC |
| db2dde7 | Olivier Couet | 21 August 2013, 13:42:49 UTC | Display the bin error for 2D histograms in the status bar. | 21 August 2013, 13:42:49 UTC |
| 6a569b7 | Pere Mato | 21 August 2013, 11:52:18 UTC | Added dependency to Smatrix (header files). Removed the dependency to Smatrix (library) since it is only a dictionary and Smatrix is all inlined. | 21 August 2013, 11:52:18 UTC |
| bc9af65 | Lorenzo Moneta | 21 August 2013, 10:39:17 UTC | Add SMatrix dependency for Cholesky decomposition | 21 August 2013, 10:39:17 UTC |
| 01df9fb | Fons Rademakers | 21 August 2013, 10:23:00 UTC | add explicit linking to libpthread as RooFitCore now uses pthread_mutex. | 21 August 2013, 10:23:00 UTC |
| 8ceb78e | Manuel Tobias Schiller | 21 August 2013, 10:04:55 UTC | fix win32 issue Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 10:11:20 UTC |
| 0eda633 | Wouter Verkerke | 21 August 2013, 08:43:25 UTC | o RooComplex - Add ctor from std::complex<Double_t> so that return values from new RooFit code can be automatically converted to RooComplex as to not break existing code o RooMath - Silence a warning message | 21 August 2013, 08:43:25 UTC |
| ee009cb | Manuel Tobias Schiller | 16 August 2013, 07:34:50 UTC | refactor crc32 code to allow hashes to be updated with new data, include FNV1a hashes (32 and 64 bit versions) for applications where speed and/or good diffusion for short strings is important Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:46 UTC |
| 58d1e1f | Manuel Tobias Schiller | 15 August 2013, 07:30:25 UTC | save bin boundaries in sorted vector instead of set - bin lookup now O(log(nbin)) instead of O(nbin) This patch uses O(log(nbin)) cost bin number lookups in a sorted std::vector throughout instead of scanning though a std::set with an O(nbin) algorithm; the old version could incur an additional penalty due to the fact that the elements of the set are saved in dynamically allocated memory that might be scattered all over the address space of the process, thus leading to cache misses. The downside of the new version is that altering the binning is more expensive. Still, with the more efficient bin lookups, we gain around 5% in speed (typically), although the gain is likely more pronounced in applications which have binnings with many bins, and a lot of bin number lookups happening. Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:46 UTC |
| 56e33ac | Manuel Tobias Schiller | 15 August 2013, 07:30:24 UTC | implement analytical integrals and provide maximum for RooKeysPdf Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:46 UTC |
| 117b432 | Manuel Tobias Schiller | 15 August 2013, 07:30:24 UTC | patch from Thomas Bird <Thomas.Bird@cern.ch>: expose the plotOn method which takes a RooLinkedList as argument for callers which have long lists of CmdArgs to pass Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:45 UTC |
| 053162a | Manuel Tobias Schiller | 15 August 2013, 07:30:22 UTC | when using SumW2, make RooNLLVar keep two offsets, one for NLLs evaluated with the weights, one for NLLs evaluated with the weights squared Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:45 UTC |
| cb56eb7 | Manuel Tobias Schiller | 15 August 2013, 07:30:21 UTC | if the correction formula for SumW2 corrected covariances is correct, the result must be positive definite and symmetric, so use tools which ensure that (and refuse to work when the pos. def. requirement is violated) The reasoning here is: If the covariance obtained by the fit where the weights enter linearly is positive definite and called V, and the Hesse matrix evaluated with the weights entering squared is called C, then C and VCV have exactly the same number of negative eigenvalues, since V cannot flip the sign of the eigenvalues of C - V positive definite. The idea of a proof is to consider the bilinear form <v, w>_M = v^T M w. If M is pos. def., then <v,v>_M > 0 for all vectors v != 0, and the bilinear form induces a norm. To see that each negative eigenvalue of C induces a negative eigenvalue in VCV, you consider the corresponsing eigenvectors of C, use that V is invertible, and arrive at non-zero vectors v- for which <v-, v-> < 0. The code that is used to obtain C from the corresponding covariance matrix is tuned for symmetric pos. def. matrices, and has been used for ages in practically all track fits done in LHCb. Therefore, it's robust, well tested, and quite fast. Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:45 UTC |
| 04a6502 | Manuel Tobias Schiller | 15 August 2013, 07:30:20 UTC | Fix bug in which extended term in PDF might not get calculated. Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:44 UTC |
| fe5566d | Manuel Tobias Schiller | 15 August 2013, 07:30:19 UTC | Propagate Kahan summation carries over process boundaries when using more than one core. Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:44 UTC |
| 6cb88f6 | Manuel Tobias Schiller | 15 August 2013, 07:30:17 UTC | make RooRealMPFE and the whole process of forking children a lot less chatty Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:43 UTC |
| 1dc055e | Manuel Tobias Schiller | 15 August 2013, 07:30:16 UTC | rewrite RooRealMPFE to use BidirMMapPipe, take care to make the communications protocol as efficient as possible (as few trips through the pipe as can be achieved) To this end, the only commands in the parent which flush the buffers to the other end are the "calculate" command which causes the child to start the calculation, and the "retrieve result" command which causes the child to send the result of the calculation. All other commands are queued in the parent until the "calculate" command is sent, and the child gets them when all buffers are flushed after the "calculate" command has been sent as one huge string of requests. In the child, the only reply which flushes the internal buffers is the reply to "retrieve result" which sends the result and any evaluation errors that may have occurred in the child in one go. The advantage of this mode of operation is that both ends may block in the OS while trying to read from the pipe until there is a significant amount of work to be done, resulting in less latency because the OS only has the schedule the end that becomes runnable once instead of scheduling it every time the other end sends a tiny request like a parameter change notification, for example. This makes it explicit that these "little" requests are nothing more than options to the one CPU-intensive request which we really want the child process to perform: the evaluation of the PDF over a certain subset of events. [FIXME:] One more thought on persistency: This class should be rewritten somewhat, because persistency is tricky: you cannot persist the child processes along with the class in the ROOT file and recreate them on the machine which reads the RooRealMPFE class. That means any reference to child processes, pids, file descriptors and other communication structures are meaningless once the class is read back on the target machine. I can see two ways out: Either we forbid persisting this class altogether, or we need some special hooks which always write the persisted version of the class in the Initialize state (irrespective of the actual state) which will force creation of the child processes on the target machine when needed. Since the old class had the exact same issue, I didn't fix this just yet... Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:43 UTC |
| 709d44d | Manuel Tobias Schiller | 15 August 2013, 07:30:15 UTC | Have EvalError contain strings, not fixed sized arrays (which are a waste of space) This patch is in preparation for the RooRealMPFE rewrite using BidirMMapPipe. It will simplify issues with fixed size buffers in the IPC there considerably. This has no impact on the persisted format of classes since EvalErrors are not persisted as far as I can tell. Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:43 UTC |
| 45b3a4e | Manuel Tobias Schiller | 15 August 2013, 07:30:14 UTC | new class BidirMMapPipe, a bidirectional communications channel between a parent process and its child, which buffers communications intenally, and tries to minimise the time spent in the OS (except when you want to block there) This class offers an interface very similar to the familiar read/write syscalls, has a flush operation to flush internal buffers to the other end, and a poll method which allows a parent process with several open pipes to block until an event occurs on any of those pipes (events would be pipe becomes readable, writable, is closed by the other end), allowing for fairly efficient I/O multiplexing on those pipes. In addition to that, it overloads the familiar operator<< and operator>> methods to allow binary I/O over the channel for the basic data types (char, short, int, long, long long and their unsigned counterparts, double and float), C-style zero-terminated strings and std::string, and raw pointers (which are sent as addresses). Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:42 UTC |
| 031f2e6 | Manuel Tobias Schiller | 15 August 2013, 07:30:13 UTC | patch modified from solution provided by Sascha Stahl <sascha.stahl@cern.ch>, fixes JIRA ROOT-5229: fix RooFit's treatment of weighted datasets in conjunction with simultaneous PDFs Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:42 UTC |
| 74277fd | Manuel Tobias Schiller | 15 August 2013, 07:30:12 UTC | some code cleanups, mostly to guard against future typos (if (0 == i) vs if (i = 0)) and to have the nice version of for loops (variable declared only inside for loop, ++i vs i++) Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:42 UTC |
| 76fe14c | Manuel Tobias Schiller | 15 August 2013, 07:30:11 UTC | patch by Thomas Bird <thomas.bird@cern.ch> and Rob Lambert <Rob.Lambert@cern.ch>: have "master process" delegate all compute jobs to children in multi-CPU mode (advantage is that the master process can in theory keep children busy without blocking on some long-running computation itself) Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:42 UTC |
| afcb876 | Manuel Tobias Schiller | 15 August 2013, 07:30:10 UTC | patch from Thomas Bird <thomas.bird@cern.ch>, Rob Lambert <Rob.Lambert@cern.ch>, with some slight robustness improvements and better error checking by Manuel Schiller <manuel.schiller@nikhef.nl>: implement analytical integrals for Chebychev polynomials Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:41 UTC |
| 6ac853f | Manuel Tobias Schiller | 15 August 2013, 07:30:09 UTC | patch from Thomas Bird <thomas.bird@cern.ch> and Rob Lambert <Rob.Lambert.cern.ch>: allows to project a region and produce an asymmetry plot of it Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:41 UTC |
| aaf1bcf | Manuel Tobias Schiller | 15 August 2013, 07:30:08 UTC | Use continued fraction approximation in Faddeeva function for large |z|, as it's slightly faster, and just as accurate in that regime Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:41 UTC |
| c1c22ca | Manuel Tobias Schiller | 15 August 2013, 07:30:07 UTC | deprecate RooComplex class, adapt rest of RooFit code base to use std::complex instead, clean up RooMath interface while I'm at it The old RooFit functions all had "error function" somewhere in their name, which was incorrect. They all calculated w(z)=exp(-z^2)*erfc(-i*z). The new functions all have different names. If you used to call one of the old functions in your code, the new functions you want to call are either faddeeva_fast (which correspond to the fast interpolated version in the old RooFit code), or faddeeva. The new code also implements the (complementary) error function erf(z) (erfc(z)) for complex arguments z. Since RooComplex is never persisted anywhere in RooFit, the class can be dropped in a future release. New code should use std::complex and the faddeeva, faddeeva_fast, erf, erf_fast, erfc and erfc_fast routines in RooMath directly. The old routines in RooMath and the RooComplex constructor print deprecation warnings. The deprecation warnings for RooComplex and old style RooMath calls are designed for maximum annoyance (every use of a deprecated function prints a warning to stdout, so users change their code), but there is an upper bound to cause minimum offence (after 4096 warning messages, the printout of warning messages stops, so log files will not fill up disks, and jobs can still be run). Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:40 UTC |
| a6b75bd | Manuel Tobias Schiller | 15 August 2013, 07:30:05 UTC | follow usual unix return code convention: 0 return from main means all tests good, return value > 0 counts number of failed tests (helpful as part of a git bisect script) Signed-off-by: Wouter Verkerke <verkerke@nikhef.nl> | 21 August 2013, 07:51:40 UTC |
| fa2d8d4 | Lorenzo Moneta | 19 August 2013, 15:20:40 UTC | Fix documentation about use of Minos in RooAbsPdf::fitTo, see #ROOT-5423 | 20 August 2013, 19:23:14 UTC |
| 52da4ff | Pere Mato | 20 August 2013, 14:34:56 UTC | Added ${PYTHIA6_DIR} from command line as a location for Pythia6 | 20 August 2013, 14:34:56 UTC |
| 9f9054a | Christian Gumpert | 19 August 2013, 14:22:53 UTC | use forward declaration instead of include | 19 August 2013, 14:37:03 UTC |
| aa19e0f | Olivier Couet | 19 August 2013, 07:28:30 UTC | Fix https://sft.its.cern.ch/jira/browse/ROOT-5413 | 19 August 2013, 07:28:30 UTC |
| 20a7411 | Olivier Couet | 16 August 2013, 14:11:30 UTC | Fix https://sft.its.cern.ch/jira/browse/ROOT-5053 | 16 August 2013, 14:11:30 UTC |
| 96d2385 | Fons Rademakers | 16 August 2013, 13:33:05 UTC | Add basic implementation for GetColumns. Fixes ROOT-5332. Patch by Oliver Freyermuth. | 16 August 2013, 13:33:05 UTC |
| 6314876 | Olivier Couet | 16 August 2013, 12:08:20 UTC | Apply fix suggested here: http://root.cern.ch/phpBB3/viewtopic.php?f=14&t=16930 | 16 August 2013, 12:08:20 UTC |
| 881fe40 | Christian Gumpert | 16 August 2013, 08:14:40 UTC | fix memory leak in HistFactory::Sample | 16 August 2013, 08:50:59 UTC |
| 074f911 | Philippe Canal | 15 August 2013, 19:16:08 UTC | Add missing protection in OptimizeBasket. Some of the branch (because they have been disabled for example) may not have any entry. So we now protect against this case (actually seen in stress). This fixes ROOT-5410 | 15 August 2013, 19:20:04 UTC |
| 6fd8155 | Lorenzo Moneta | 15 August 2013, 14:10:44 UTC | Fix the cloning of the OneDimMultiFunctionAdapter. The X vector was not copied. See http://root.cern.ch/phpBB3/viewtopic.php?f=15&t=16824 | 15 August 2013, 14:11:55 UTC |
| bf8ca3c | Gerardo Ganis | 15 August 2013, 13:24:36 UTC | Merge branch 'v5-34-00-patches' of github.com:gganis/root into v5-34-00-patches | 15 August 2013, 13:24:36 UTC |
| 3f17c35 | Dario Berzano | 15 August 2013, 11:36:38 UTC | Harmonized a couple of R_ASSERTs | 15 August 2013, 13:24:05 UTC |
| 782625f | Dario Berzano | 15 August 2013, 10:05:21 UTC | Fixed CID 52067 and retval of PollForNewWorkers() | 15 August 2013, 13:24:04 UTC |
| 036acf4 | gganis | 15 August 2013, 13:18:12 UTC | Merge pull request #3 from dberzano/v5-34-00-proof-dynamic-workers Coverity fix and better consistency checks | 15 August 2013, 13:18:12 UTC |
| a5248ff | Dario Berzano | 15 August 2013, 11:36:38 UTC | Harmonized a couple of R_ASSERTs | 15 August 2013, 11:39:53 UTC |
| eb06ab4 | Dario Berzano | 15 August 2013, 10:05:21 UTC | Fixed CID 52067 and retval of PollForNewWorkers() | 15 August 2013, 11:39:53 UTC |
| 649d259 | Olivier Couet | 15 August 2013, 11:08:54 UTC | In TGraph2DPainter::PaintLevels the colour levels used to paint the triangles did not match the minimum and maximum set by the user on the TGraph2D. | 15 August 2013, 11:08:54 UTC |
| b52f175 | Olivier Couet | 14 August 2013, 15:40:57 UTC | Protection added to avoid a Seg Fault on ".q" when SetHistogram is called on a TGraph2D. | 14 August 2013, 15:40:57 UTC |
| 9f2233c | Fons Rademakers | 14 August 2013, 11:06:04 UTC | only run sqlitetest when sqlite plugin has been build. | 14 August 2013, 11:06:04 UTC |
| 57b52e7 | Fons Rademakers | 14 August 2013, 10:56:46 UTC | fix typo. | 14 August 2013, 10:56:46 UTC |
