Darshan

=============
* Fix segfault affecting new DARSHAN_MOD_DISABLE/ENABLE environment variables
* Fix divide-by-zero condition that can potentially be triggered by new
heatmap module
* Fix potential MPI errors related to calling MPI_Type_size() on a
user-supplied MPI_DATATYPE_NULL type (reported by Jim Edwards)
* cuserid() is no longer the default method for determining username, and must
be manually enabled at configure time
* Fix backwards compatibility bug affecting darshan-3.0.0 logs in darshan-util
C library functions used by PyDarshan
* Suppress noisy output warnings when using darshan-job-summary.pl
* Clarify units displayed by darshan-job-summary.pl (reported by Jeff Layton)

==================
* Add a new "heatmap" module that records histograms of I/O traffic over time
on each process
* Performance optimizations to runtime wrappers to limit function call
overhead and locking costs
* Replace mutex locks with atomics in core library
* Add optional, limited support for RDTSCP-based timing, enabled via
--enable-rdtscp at configure time
* Remove unecessary explicit mpich binding libraries from link line when static linking
* Graceful warnings when parsing log files with unknown module types
* Remove deprecated performance estimate methods from darshan-parser
* Add two new derived metrics to darshan-parser --perf: agg_time_by_slowest
(total elapsed time performing I/O by the slowest rank) and
slowest_rank_rw_only_time (total elapsed time performing read/write
operations by the slowest rank)
* Add unified Darshan configuration support via env vars and config files
* Adopt automake and libtool for Darshan build, and allow top-level build
of both darshan-runtime and darshan-util (contributed by Wei-Keng Liao)
* Increase default memory limit for record names (NAMEMEM config setting)
to 1 MiB

=============
* fixed github issue 399: undefined llapi_layout_get_by_xattr function in some
scenarios because of incomplete dependency information in Darshan shared
library
* modified Darshan to properly handle for non-MPI applications that call fork()

=============
* added AutoPerf modules for tracking MPI communication usage (APMPI) and
for tracking Cray XC performance counters (APXC)

=============
* fixed bug in HDF5 module resulting in warning messages for instrumented
datasets that use H5S_SELECT_NONE (reported by Marta Garcia Martinez)
* disabled DXT's MPI-IO offset tracing for Darshan builds using OpenMPI
to workaround an OpenMPI bug (reported by Tobias Meisel)
* modified Darshan to query Lustre striping information for all
instrumented files to workaround issue of retrieving striping info for
Lustre files that are symlinked to other file systems

=============
* added first version of PyDarshan, which provides interfaces for
accessing Darshan log file data from Python applications
* modified Lustre module to use 'fgetxattr' rather than 'ioctl' for
querying stripe info to avoid known bug in Lustre that could cause
application crashes
* fixed bug leading to potential deadlock when reducing shared file records
for MPI applications (known to affect mvapich2)
* fixed bug preventing usage of Darshan's non-MPI mode in builds that use
MPI compilers
* boosted Darshan's default internal memory allocation from 2 MiB to 4 MiB
* added wrappers for openat, preadv, preadv2, pwritev, and pwritev2 (improves
profiling of ompio)
* various bug fixes to enable usage of HDF5 modules in non-MPI applications
* fixed HDF5 module bug leading to warnings on stdout for applications using
non-MPIIO VFD for HDF5
* fixed HDF5 module bug leading to potentially incorrect values in H5D module
common access counters
* added new H5D record value to indicate corresponding H5F record ID, to
allow easier mapping of HDF5 datasets to files they belong to
* improved error handling in command line parsers to make incomplete logs more
obvious to users
* fixed bug leading to incorrect internal timing information at Darshan
shutdown time
* fixed bug causing negative Darshan counter values from being plotted when
using the darshan-job-summary tool