############################################################################### # # # Trilinos Release 7.0 Changelog # # # ############################################################################### Release Update 7.0.7 April 18, 2007 Change number. Package(s) affected - CVS log message 1. EpetraExt - Bug 3231. Merging changes to remove code from inside assert statements to the 7.0 branch. This was causing problems when using -NDEBUG. 2. EpetraExt - Bug 3232. Fixed the behavior of the utility that writes out an Epetra_MultiVector to a Matrix Market file so that the entries are written out column-wise instead of row-wise. The utility that reads a Matrix Market file into an Epetra_MultiVector was modified so that the rest of the lines of the MultiVector that aren't necessary for the particular processor are read in and discarded. This ensures that each processor ends up at the correct place in the file where the next MultiVector begins. Testing for reading/writing Epetra_MultiVectors to Matrix Market files was added to the inout test which illustrates the bug before the fixes were applied. 3. Amesos - Bug 3230. Updated the Amesos_Time class to be more efficient, using direct accessing of the internal time data instead of a Teuchos::ParameterList. All the solvers that use the Amesos_Time class have been updated. I am only able to check that this update worked for: DSCPACK, KLU, PARAKLETE, SuperLU, SuperLU_DIST, and UMFPACK. The others will have to be checked on beowulf, or any platform that has those 3PLs available. 4. Rythmos - Bug #3234 I removed parallel testing for basicExample after we discovered that this example (and test) is not formulated correctly in parallel. The algorithms are still being tested through the 1DfemTransient example which is correctly formulated in parallel. 5. Amesos - Bug 3230. Fixed the usage of the Amesos_Time class in Test_Basic. 6. Amesos - Bug 3230. Additional methods have been added to the Amesos_BaseSolver class to allow users to obtain timing and status information from the solvers. There is now a GetTiming() method that takes in a Teuchos::ParameterList as an argument and populates the parameter list with the total timings collected by the Amesos solver object. There are also methods for to user to get the number of symbolic and numeric factorizations as well as solves performed by the solver object. All together, this information allows users of Amesos to obtain performance data without having to print out using PrintTiming(). 7. EpetraExt - Bug#3236. The new column map will now use the same index base as the old column map. Previously, it was hardwired to 0. 8. LOCA - Adding a test that demonstrates bug #3250, where copies of LOCA groups are leaving the DerivUtils pointer null, causing an exception to be thrown. 9. LOCA - Changed the LOCA::Abstract::Group copy constructor to use copy() instead of copy constructors for its base classes to fix bug #3250. The LOCA::DerivUtils pointer was not being properly copied due to multiple virtual inheritance. 10. Pliris - In the function perm1_ a new temp array was defined my_rhs_temp and malloc'd to an appropriate size to avoid writing data out of bounds. This variable is in the scope of the function perm1_. 11. Amesos - Changing access to parameters in the solver specific parameter list so that they are optional parameters as stated in the documentation, which have default values. The current retrieval of these parameters assumes that they exist if the solver specific parameter list exists, which is not necessarily true. This is evidenced by the test failures seen on beowulf. 12. Amesos - Bug 3230. Insert examples of how to extract timing information from an Amesos solver. 13. Amesos - Bug 3230. Added a test for retrieving the timing information via a parameter list and retrieving the particular information from the parameter list. This passes in serial and parallel with the limited number of solvers I have on bandito. 14. ML - Bug#3264. The NewGS smoother was causing Hiptmair smoothing to break. The problem was using the diagonal without checking for zeros. The fix is just to replace any zero with a one. Also added one other guard against zeros on the diagonal, and changed some code that I think some compilers might complain about: incrementing a left hand side value. I seem to recall that some compilers might do the increment before assignment. 15. ML - Bug#3273. Cleaned up some purify UMRs in calls to ML's timing routines. This has already been fixed on the trunk. ------------------------------------------------------------------------------ Release Update 7.0.6 March 26, 2007 Change number. Package(s) affected - CVS log message 1. NOX - Fix for bug 3045. A request was made to fix some warnings coming from a sun compiler. All of the erros were of the form: NOX_Solver_TensorBased.C", line 725: Warning: beta hides NOX::Solver::TensorBased::beta. It was legal syntax, but not the best programming practice. 2. ML (Python) - In response to a bug reported via email, an additional #include was added to ML.i to ensure proper compilation 3. Framework level - Bug 2800. Added --enable-python to the end of list of configure arguments passed to packages when --enable-pytrilinos is used. Formerly --enable-pytrilinos would not enable the python wrappers in the various packages. This is not an issue in the development branch as the python wrappers have all been moved to the PyTrilinos package. I tested on my local machine for proper make behavior and to verify that --enable-python is passed to the package configure scripts when --enable-pytrilinos is used. 4. Anasazi - modified TestMultiVecTraits: resizing the input/output vector to MvNorm will now report a warning instead of an error. this addresses an incompatibility with the NOX::Abstract::MultiVector used in Salinas. 5. Teuchos (Python) - Upgraded to support python 2.5 6. Stratimikos - Bug 3166 - Made linear solver actually get the preconditioner factory. 7. ML - Bug 3104. The value for B_total_Nnz was getting set to a too-large value. This is because Bmatrix->getrow->Nrows is too large. Now, B_total_Nnz is always calculated via getrows, rather than using Bmatrix->getrow->Nrows. 8. ML - Bug#3027. The Gauss-Seidel subsmoothers in Hiptmair were not optimized for Epetra_CrsMatrix type matrices. Merging in some code from the development branch written by Ray Tuminaro to fix this. 9. Galeri - Bug 3198. Fixed include paths in Galeri_FiniteElements.h to be correct during compilation and after installation. 10. ML - Bug#3027. I'm removing the call to OptimizeStorage() in Epetra_ML_GetCrsDataptrs(). If the matrix storage can't be optimized, this prints an error to the screen in every application of the preconditioner. 11. Galeri - Bug 3198. Including the current Makefile.export.galeri file instead of Makefile.export in the galeri/examples Makefile.am. This change was already made to the dev branch. I compiled on my linux machine to verify that it fixes the problem that showed up in the test harness. 12. AztecOO - Addresses bug 2035. Changed "nnz<1" to nnz==0. This is not a permanent fix, but should work well for the foreseeable future. 13. Teuchos (Python) - Moved typedef for Py_ssize_t before #include "Teuchos_PythonParameter.hpp" 14. Teuchos (Python) - Moved backward compatibility typedef for Py_ssize_t from Teuchos.i to Teuchos_PythonParameter.hpp, which fixes python 2.4/2.5 compatibility problem in other packages 15. ML - Bug #2854. Changing the default value for the parameter list option "read XML" to false. If this option is true, then each processor looks for an input XML file, which is bad in parallel. 16. ML - Bug#2854. Added a new parameterlist check in BreakForDebugger(). The user must now specify "debug mode" to be true in the parameter list. Otherwise, BreakForDebugger will return immediately. This avoids another file I/O scalability problem. 17. ML - Bug#2854. Turning off ML's debugging capability. ------------------------------------------------------------------------------ Release Update 7.0.5 February 7, 2007 Change number. Package(s) affected - CVS log message 1. Anasazi - Removing a semicolon from the constructor definition on line 277. This was causing a compiler error on a SunOS platform. Removed other unnecessary semicolons from the class definition. 2. Teuchos - Removing an element of the array to illustrate bug #2983, where the method "remove" is not defined for the Teuchos::Array class but used in the Teuchos::HashTable class. 3. Teuchos - Bug# 2983. Adding the "remove( int i )" method to the Teuchos::Array class. This method was called by the Teuchos::Hashtable class, but never defined by the Teuchos::Array class. The Container/cxx_main.cpp file was modified to include a call to "remove" an element of the hashtable, which illustrates the bug. Now that the method is declared and defined in the Teuchos::Array class, this test compiles and runs. I will make sure this doesn't affect the test-harness before closing the bug. 4. Epetra - Address bug 2998. This fix was already made to the development branch. It corrects the bug where the Epetra_CrsGraph object's importer and exporter attributes were not being updated when the row- and/or column- map are replaced. 5. Framework level - Bug 3008. Modifying the configure arguments for the nightly serial hrothgar build. 6. Epetra - Finish addressing bug 2998. Yesterday's commit make sure that when ReplaceRowMap is called, that the CrsGraph object updates its Import/Export objects. This commit now makes sure that CrsMatrix also replaces its import/export vectors. Now the test code submitted by the user runs, and all epetra tests still run and pass. 7. NOX - Fixed build error with --enable-nox-debug configure flag for bug #3014. Also interchanged RHS and LHS filenames since they were switched. 8. NOX - Added documentation for the new parameter "Skip df/dp" for bug #3013. 9. NOX - For bug #3013, added a parameter list option to the constrained group to skip the df/dp computation. In principle, this can be used for any computation involving the constrained group, but really should only be used for natural continuation. The parameter name is "Skip df/dp", the default is false, and the parameter should be in the stepper sublist. When set to true, the Trilinos 6 natural continuation results should be obtained. 10. NOX - Added a test the exercises the new "Skip df/dp" parameter for natural continuation. It checks the number of residual fills in both cases to make sure they are appropriate. 11. NOX - One more small change for bug #3013. Forgot to copy skipDfDp in copy(). 12. ML - Bug #3027. For an Epetra_CrsMatrix, ML will now use Ifpack's (symmetric) Gauss-Seidel smoothers, if available. Those smoothers exploit the data layout, whereas ML's generic internal SGS does not. All tests pass on my workstation. 13. ML - Bug#3027. Now that ML is using Ifpack (S)GS smoothing for Epetra_CrsMatrix types, this test had to be modified so that it compares ML's internal smoothers to Ifpacks. I also cleaned up the parallel output and the parallel failure exit mode. Compiles cleanly, passes in serial and np=4 under lam mpi. 14. Amesos - Bug# 3042. Returned "if ( ! TrustMe_ )" flag back into parts of the code where it was in Trilinos 6, but removed. The TrustMe_ flag is necessary for ensuring optimal performance of Amesos_Klu in Xyce when the linear systems being solved for are very small (7x7). There should be a better way to deal with this because the TrustMe_ flag not only turns off timing in some parts of the solver that are called numerous times (numeric factorization and solve), but also keeps the solver from doing sanity checks on the input. These are two functions, and probably should be dealt with separately and in a more consistent manner. 15. ML - Bug#2854. Changing the output level at which operator profiling occurs to 11. This should make the output much more readable. 16. ML - Changing output level at which operator profiling occurs to 11. This should make the screen output much more readable. 17. ML - Bugs #3103 and #3104. This fixes a seg fault if ML is built with detailed timing on. This also addresses potentially poor scalability in the matrix-matrix multiply routine due to hashing problems. This has already been fixed in the development trunk. Compiles cleanly and all tests pass on my workstation. Compiles on sun OS with no new warnings. 18. Framework level - Bug 3106. Updating the serial test harness script. I was able to compile using this script. 19. ML - Updating comments in change log. 20. ML - Bug#3104. Certain systems do not have stdint.h available. I am protecting inclusion of stdint.h and inttypes.h via the autoconf macros HAVE_STDINT_H and HAVE_INTTYPES_H. This compiles on my Linux workstation, on qt (Compaq compilers, where compilation was failing before), and sass9000 (SunOS). Tests pass on my workstation. 21. Framework level - Bug 3106. Updated the beowulf mpi test harness script to reflect the changes necessary since the beowulf upgrade. I made these changes on the dev branch yesterday and the test harness results from today looked good. 22. ML - Bug #3104. Changing how ML_matmat_mult() handles empty rows. Instead of global id 0 for the column id in a (previously) empty row, now a global ID of a processor already involved in the matrix-matrix multiply is used. Fix suggested by Ray Tuminaro. Thanks, Ray! ------------------------------------------------------------------------------ Release Update 7.0.4 December 20, 2006 Change number. Package(s) affected - CVS log message 1. MOOCHO/EpetraExt - Bug 2577 - Finished editing the MOOCHO overview document. I also am checking in an update to the documentation for the EpetraModelEval4DOpt example. This finally finishes the basic documentation for MOOCHO. 2. MOOCHO - Bug 2577 - Added a top-level link to Trilinos. (Documentation) 3. NOX - Bug 2849. Added a large number of files that are necessary for some NOX serial tests to the distribution tarball. I generated a new tarball and ran the serial test harness to make sure the problem I was seeing is fixed. 4. ML - Bug #2555. Allen Robinson pointed out the unused parameters in both SetDefaultsSA() and SetDefaultsMaxwell(): params, options. This commit removes them. Also setting the default output level to zero (i.e., no output) in all of the SetDefault* methods. 5. ML - Bug #2854. Changing the "run-tests" target to support overriding the TRILINOS_TEST_CATEGORY variable on the command line. The result is now a developer can run a particular test category of her choosing, e.g., >cd ml >make run-tests TRILINOS_TEST_CATEGORY=scalability This change was already made to the developer head. 6. ML - Partially addresses bug#2854. I'd added a profiling check just after the edge matrix is repartitioned. There was already a profiling check before the repartitioning, so now we should be able to measure the benefit of load balancing. 7. Framework level - Added template_args.py utility for listing the arguments of a template found in a list of files 8. Framework level - Making runconfigure blow away config.cache on configure failure. 9. AztecOO - Address bug 2866, a memory leak associated with AztecOO performing scalings on Epetra objects. The test program in test/AztecOO was altered to produce the memory leak, (due to successive calls to the scaling function with action == AZ_SCALE_MAT_RHS_SOL) and then the scaling code was fixed, and the test re-run with purify to verify that the memory-leak no longer occurs. Other aztecoo tests were verified to still run and pass. 10. Anasazi - Disabling BlockKrylovSchur_test.exe, BlockKrylovSchurThyra_test.exe, and LOBPCGThyra_test.exe. See bug 2875. 11. Anasazi - changed file headers: they mentioned Belos instead of Anasazi, an artifact from being copied from Belos Status Test 12. ML - Bug #2863. This commit fixes two problems with how ML generates random numbers. There was a bug in the algorithm implementation (lifted from Park and Miller's paper, "Random number generators: good ones are hard to find"). The implementation of the seeding could also lead to an integer overflow situation very easily in parallel. ML's code was lifted verbatim from AztecOO, so this bug exists there, too. I'll open a separate bug and make the necessary commits. Bug #2854. Modifying functions StartTimer() and StopTimer. Added ReportTimer(). These are useful for fine grained timing within ML. They do nothing if ML's print level is zero. StartTimer and StopTimer do what their names imply. ReportTimer prints average time over all processors & a standard deviation. 13. ML - Bug#2863. The iteration count for this test changed by 1 because how the initial random guess is calculated has changed. I also changed exit(1) and exit(0) to used EXIT_SUCCESS/EXIT_FAILURE. 14. AztecOO - Address bug 2598. These fixes were already made in the development branch. They address a bug in the way AztecOO was handling the options AZ_conv = AZ_Anorm and AZ_scaling = AZ_sym_diag. The issue was that the norm of A (used in the convergence test) was taken before the scaling was applied. This is now fixed and the test program is beefed up. 15. AztecOO - Bug#2890. This commit adds a missing "else" statement in AZ_srandom1(), which implements Park & Miller's serial RNG. It also changes how seeing is initially done in AZ_random_vector() to avoid a potential integer overflow condition. I have put in a new test, test_bug2890, in aztecoo/test/AztecOO that checks this patch. This is closely related to ML bug#2863. 16. ML - Adding new test that's a catch all for checks that don't fit in anywhere else. Includes a check for bug#2863. 17. ML - Adding in a test for bug#2863. 18. Anasazi - Disabled LOBPCG_auxtest.exe. See bug 2875. 19. AztecOO - Bug#2890. Minor change to test output. For the very curious, the bug number is now printed in the output. 20. Anasazi - Re-enabled all tests. Hrothgar has been unscheduled until it can be made to run reliably. 21. Epetra - In response to bug 2895, several Epetra python test scripts have been updated to be compatible with the recently released numpy 1.0, while maintaining backward compatibility with earlier versions of numpy. The problem occurs because we are testing that exceptions get raised in certain situations, and the exceptions have been changed in numpy from TypeError to ValueError. The scripts have been changed so that either exception is acceptable. 22. Epetra - Address bug 2906. Fixes a memory leak in Epetra_Util::Create_OneToOne_Map. This fix was already committed to the development branch. 23. ML - Bug#2854. These changes are almost exclusively improvements to internal ML timing and operator profiling. I cleaned up the output in ML_Gen_MGHierarchy_UsingReitzinger(), especially the operator profiling. Each operator should be profiled just once. In the case that it could be redistributed across processors, it is profiled before and after. Static profiling is now possible, even when ML is not built with detailed timing enabled. Time spent in ML_Gen_MGHierarchy_UsingAggregation() is now reported at output level 10. Moved an annoying global debugging print to output level 15. 24. ML - Bug#2863. The seed for the Park & Miller method was not right. It must be in the interval [1,2^31-2]. This was causing a very non-random behavior within ML setup for particular Alegra simulations. 25. ML - Bug#2863. Changing the error message if the seed is bad and adding a code comment. Removing bad seeds from nightly test. 26. ML - Bug#2863. Making doubly sure that the seed can't be equal to 2^31-1 by multiplying by 2^31-2 rather than 2^31-1 in the seed routine. 27. AztecOO - Bug#2890. There was still a problem in the RNG seed routine. Only integers in [1,2^31-2] are valid seeds. I've updated the nightly tests, as well. This matches what is done in ML. 28. ML - Adding some code to project coordinates to the next level using Ptent. 29. ML - Bug#2854. Adding some profiling hooks around repartitioning call in ML_Gen_MultiLevelHierarchy(). 30. ML - Bug#2917. Fixing a compile error under Windows. 31. ML - Bug #2854. Adding a pid check to an informational repartitioning message to prevent swamping stdout. 32. ML - Bug#2920. Fixing this example for the case when the user does not provide edge and node maps. Thanks to Ray for catching this. 33. ML - Bug#2917. Backing out patch. I didn't have the correct configure options to actually test it. 34. ML - Bug#2917. Recommiting Alan William's patch. I initially missed the 'class' keyword. Compiles fine on my Linux workstation with strict compiler options. 35. ML - Bug#2854. Defining a macro that should always be on. It controls ownership tie breaking in parallel. 36. ML - Bug#2854. In merging changes, a vector initialization (to random) was inadvertently deleted. The effect doesn't show up in serial, but is critical in parallel. 37. Multiple (Python) - No bug was submitted for this problem, but if libraries are specified in the configure invocation as "/dir/path/libname.a" instead of "-L/dir/path -lname", the python wrapper build system was not set up to handle these libraries properly. Extra code was added to all of the setup.py scripts in order to account for this possibility. 38. Multiple (Python) - Bug 1848 is the disabled python wrappers for NOX. In attempting to re-enable this interface, I have run into issues with Teuchos::ParameterList. As a precursor to addressing these issues, I am getting rid of the PyDictParameter class. For a variety of reasons, ParameterList is not suitable for subclassing, so PyDictParameterList was never able to achieve the functionality I had hoped for. I have transferred much of what *did* work for PyDictParameterList to the python version of ParameterList: a constructor that takes a python dictionary, and methods and operators that mimic python dictionary behavior. I have also changed the typemaps so that dictionaries are converted to ParameterLists directly rather than PyDictParameterLists. This affected two other package wrappers: AztecOO and ML. The swig interface files for those packages used the PyDictParameterList class, and they have been altered to use the new ParameterList functionality. 39. AztecOO - Commit a fix for an issue related to bug 2598. This fix has already been committed to the development branch. AztecOO's support of scaling with Epetra objects wasn't quite right. If a user ran AztecOO with options[AZ_keep_info] = 1 for the first solve, and then did a subsequent solve with options[AZ_pre_calc] != AZ_reuse, the scaling vector from the first solve would be reused. In this case, the scaling vector should be discarded and replaced with a new one. 40. Teuchos (Python) - added some documentation 41. AztecOO (Python) - Cleaned up commented-out code 42. Anasazi - added new options to control kappa parameter of BasicOrthoManager declared in BKSSolMgr. added new timers to BasicOrthoManager, for accounting for reorthogonalization 43. AztecOO - Another tweak for bug 2598, AztecOO scaling with Epetra matrices. In this case the matrix was being scaled a second time if options[AZ_pre_calc]=AZ_reuse. The matrix should only get scaled if options[AZ_pre_calc] <= AZ_recalc. 44. ML - Bug#2961. Changing some default values so that ALEGRA regression tests agree, whether using Aztec 2.1 or Epetra interface. 45. Multiple (Python) - Bug 2965. Fixing a Python wrapper compile error on the release 7.0 branch that was previously fixed on the dev branch. 46. Teuchos (Python) - Bug 1848. Bootstrapping after a change to remove testPyDictParameterList.py. 47. Claps - correct .hpp files 48. ML - Bug#2555. Backing out change to SetDefaultsSA() and SetDefaultsMaxwell(). ------------------------------------------------------------------------------ Release Update 7.0.3 October 25, 2006 Change number. Package(s) affected - CVS log message 1. Framework level - Updaing the path for hrothgar. 2. Claps - changed back to 6-0 release version rather than development branch version 3. ML - Fixes bug#2444 in the 7.0 branch. This was already fixed in 6.0.20 and the developer's trunk. Previously, a user-specified damping parameter for Gauss Seidel was ignored. Now, the damping parameter is properly overridden by one specified in the parameter list. Otherwise, the damping factor is 1.0 in serial, and automatically calculated in parallel. This was originally pointed out by Joseph Young. 4. RTOp - Bug 2839 - I have fixed a problem where the imaginary part of a complex reduction object was getting dropped. I added a bunch of debug-enabled print statements to help track this down. This turned out to be a simple problem to fix in the RTOp support software. Now the Thyra test programs test_product_space.cpp and test_composite_ops.cpp run currectly for me (tested on four processes). There are still test failures for test_handles.cpp and test_linear_combination.cpp but I am going to throw these over for Kevin to look at (or to disable for MPI). 5. RTOp - Making the change suggested by Ross to fix Bug 2843. Replacing, ...::num_values_(...); with ...::num_values_ = ...; in templated call. 6. Teuchos - Bug 2807 - I forgot to comment out a temporary print statement to show that the right timer was being called on a Microsoft Windows Win32 system. This will not impact non Win32 code. 7. Stratimikos - Bug 2772 - Minor documentation updates. 8. Anasazi - Added test to illustrate the failure in bug 2840. This failure is due to the computation of the Ritz residuals being wrong when the eigenproblem is non-Hermitian and the scalar type is complex<>. 9. MOOCHO - Bug 2577 - I have finally finished the edit on the main page (the MOOCHO Reference Guilde) and I am about half of the way through editing the overview document. I am now very very close to having a clean set of documentation. 10. MOOCHO - Bug 2844 - I have replaced all of the lingering assert(0) statements with TEST_FOR_EXCEPT(true). Using the macro is better since it will thrown an exception and will not bring down the problem. 11. Thyra - Bug 2827 - I have removed the programs test_handles.cpp and test_linear_combination.cpp from the parallel test suite for the time being. When Kevin Long has time to look at these he will get back to this. This should result in all Thyra tests passing in parallel now. 12. Anasazi - This commit addresses bug 2840. The block Krylov-Schur solver now has another helper function to enable special treatment of the Ritz residual computation when faced with a non-Hermitian eigenproblem. An additional test was created to generate the bug, this test now passes taking the same number of iterations to solve the discretized 1D Laplacian operator irregardless of whether this operator was declared Hermitian or non-Hermitian. ----------------------------------------------------------------------------- Release Update 7.0.2 October 18, 2006 Change number. Package(s) affected - CVS log message 1. ML - Bugfix: Ray caught an error in handling input parameters of zero 2. ML - Ray's second bug catch of the day on float/int handling 3. MOOCHO - Bug 2577 - I have moved some more text around and I think that the basic content is about there. I still need to read through all of this for an initial edit before this is presentable. I have also added some make targets to the external make files in order to show how to run these cases. 4. ML - Fixing bug with matlab build, bootstrapping 5. ML - Fixing MLMEX compilation bugs 6. MOOCHO - Bug 2577 - Added some missing references did a basic spell check. This included a misspelled work 'occured' in the source code for output. 7. MOOCHO - Bug 2577 - Minor fix. 8. WebTrilinos - My installation of WebTrilinos seems to need this image file. The logs indicated that this file was deleted and moved to the web pages on software, which seems a bit odd. 9. WebTrilinos - Fixing minor typos in various text. 10. LOCA - Re-enabled the LOCA::StatusTest::Wrapper for bug #2821. This was missing after the upgrade to the new LOCA framework and was causing problems for Xyce. All tests pass... 11. LOCA - Additions to address bug #2821. The NOX::Solver::Wrapper was missing an implementation of the reset() function. In trilinos 6.0, we had two overloded imeplemenatations of reset. A third one was added in the 7.0 release. It was in the header file, but not implemented in the .C file. 12. WebTrilinos - Addresses bug#2822. In the setup web page, added a input box for the location of a python binary. This way, the installer can use Python from a nonstandard location, e.g., /usr/local/bin/python. The testing page will echo the user's choice and print the Python version. 13. WebTrilinos - Fixes bug#2823. The image link is now in the form "http://your.machine.gov/blahblah". Before, it was relative and not displaying properly. This necessitated passing in an extra argument in the dictionary (I think that's what it's called) for this python script. This argument is the location of html images. 14. WebTrilinos - Addresses bugs#2822 and 2823. Also fixed a minor typo. 15. WebTrilinos - Accidentally checked in the wrong config.py file. 16. WebTrilinos - Accidentally checked in the wrong config.php file. 17. Thyra - Bug 2827 - I am checking in an update to the test harness testing that should show errors in parallel runs that are occuring. ---------------------------------------------------------------------------- Release Update 7.0.1 October 11, 2006 Change number. Package(s) affected - CVS log message 1. Teuchos - Bug 2790 - Here I have fixed some bad behavior where functions were calling MPI functions from static data after main() ended and MPI_Finalize() has been called. To get this to work, I added two new static functions to Teuchos::GlobalMPISession called mpiIsInitialized() and mpiIsFinalized(). I also updated the rest of the functions to behave reasonably after MPI_Finalize() has been called. Another thing that I did was to move the definitions for the functions getRank() and getNProc() into the *.cpp file so that any future changes will not require recompilation. I then updated the logic in Teuchos::MpiReductionOpSetter to only call MPI_Op_free(...) if Teuchos::GlobalMPISession::mpiIsInitialized() returns true. Note that these changes really only affect code that is based on the Teuchos::MpiComm subclass which means RTOp and therefore Thyra-based parallel code. I built from scratch and run the serial and MPI builds and all tests that worked before work now. 2. Epetra (Python) - Updated configure process to handle numpy 1.0 release candidates. Previously, numpy.lib.UserArray was required. Now I check for nunmpy.lib.user_array, and if and only if that does not exist, then numpy.lib.UserArray is required. 3. MOOCHO - Bug 2773 - I added a test to ensure that if you run Stratimikos without any options on more than one process that it will select something other than Amesos as the default solver. 4. MOOCHO - Bug 2793 - I added a new exception class AbstractLinAlgPack::MatrixSymSecant::UpdateFailedException and put in a try/catch to catch the DenseLinAlgLAPack::FactorizationException exception and rethrow it as a AbstractLinAlgPack::MatrixSymSecant::UpdateFailedException object. I also now print the exception type in the final message. While the problem is not totally fixed yet at least now you get a good exception message. 5. ML - Adding additional help on null space option. How to specify the number of null space vectors was missing. 6. ML - Bug #2575. Synchronizing the 7.0 and dev user guide versions. A section on the ML/Matlab interface was missing from the guide that went out with 7.0.0. 7. AztecOO - Address bug 2598, bad solution values coming from AztecOO when using sym_diag scaling with epetra matrices. I found and fixed the bug, which affected both row_sum scaling and sym_diag scaling. I also beefed up the test aztecoo/test/AztecOO/cxx_main.cpp to catch the bug. The test now passes, as do other AztecOO tests. Also verified that the test runs clean with purify. 8. Framework level - Adding missing script for Cray XT3. 9. Various (Python) - The python test and example directories generally contain scripts, python modules and data files that need to be copied from the source directory to the build directory. These copy commands in the Makefiles have been protected by 'if' statements for the case when the source directory is used as the build directory. However, these conditionals didn't actaully work. I have gone back and changed these statements so that these checks behave as intended. In the process of fixing this, I also found some data files in other packages that are copied from source to build that had the same issue, some with no attempt at protection. I have fixed those, too. Sometimes these copy statements used the make "@" syntax to supress output of the command make was running, which made the failures look very mysterious. I have removed those "@" prefixes so that it is more clear what make is doing. 10. ML - Finally got aggregate mode all sorted out. Also added code to deal with the interface change from 7.2 to 7.3 for functions like mxGetIr (which changed return type) 11. Thyra - This commit takes care of bug # 2804, which reported a missing forward declaration prior to the call: virtual Teuchos::RefCountPtr > productSpace() const = 0; 12. ML - Bug 2808. The double array being passed to the Norm2 method was the wrong length. I simply made the number of rhs and lhs equal to 1. This now runs fine on bandito under the Intel 9.0 compilers. 13. Teuchos - Bug 2807 - I rearranged the ifdefs to call the Win32 time monitors (which have accuracy to 1e-5 seconds typically) automatically when using the Intel C++ compiler for Windows XP which I am using under cygwin on my Windows XP laptop. Now I get very accurate timings for even small problems. 14. MOOCHO - Bug 2577 - Updated some more documentation to MOOCHO. I about have the minimal content that I need to make this complete. 15. MOOCHO - Bug 2577 - Made some more updates to the MOOCHO documentation. Somehow the MoochoOverview.tex file would not the most recent but I had a backup from s858392 that I just copied over. Most of the content is now there but I need to let this sit and then do a detailed edit. 16. ML - Fixed a bug which broke providing coordinates for aggregation in 2D 17. Amesos - Bug 2810. Adding a missing file, Amesos_SCALAPACK_wrappers.h, to the distribution tarball. This change was already applied to the 6.0 release branch. I am also committing a change to a Makefile.in that was not modified recently when the corresponding Makefile.am was modified. 18. MOOCHO - Bug 2577 - A little more progress on the documentation. This does not impact any compiled code as you can see from the file list. 19. Didasko - Bug 2465. Changed the Didasko version number to 7.0 after reading a comment from Marzio noting that Didasko and Trilinos should have the same version number because the Didasko tutorial is so closely tied to each release. 20. ML - Fixing a build error on reddish. Bugs #2813 and #2808. 21. MOOCHO - Contributes to bug 2577. Adding a file to the distribution tarball. This file was found to be missing during the initial 7.0.1 tarball test.