GitLab

A common record comparision/sort routine is now defined in
darshan-common which can be used by any modules with fixed-length
records and whose records first field is the darshan_base_record.
The comparison was also updated to sort secondarily by darshan
record identifier to resolve a potential bug in the way we have
traditionally sorted the records.

Conflicts:
	darshan-runtime/Makefile.in
	darshan-runtime/configure
	darshan-runtime/configure.in
	darshan-runtime/darshan.h
	darshan-runtime/lib/darshan-core.c
	darshan-runtime/lib/darshan-hdf5.c
	darshan-runtime/lib/darshan-mpiio.c
	darshan-runtime/lib/darshan-pnetcdf.c
	darshan-runtime/lib/darshan-posix.c
	darshan-util/Makefile.in
	darshan-util/darshan-convert.c
	darshan-util/darshan-logutils.c
	darshan-util/darshan-logutils.h
	darshan-util/darshan-parser.c

rather than using 'begin_shutdown', 'get_output_data', and
'shutdown', we combine all three into one shutdown routine.

Cleanup the structure and documenation of the code to be in sync
with the new mmap version of Darshan. In particular, I made sure
all extra mmap logic was properly ifdef wrapped so it's more clear
which portions of the code are responsible for implementing that
feature.

This interface provides some convenience functions for mapping
opaque record identifiers (could be a filename, file descriptor,
or any other type of file handle) to a corresponding runtime
structure referencing the actual file record. This runtime
structure should at least contain a pointer to the actual file
record, and can also persist arbitrary state for this record.

- memory management is now entirely handled by darshan-core for
  module records
    * when registering new records, darshan-core returns the
      corresponding buffer to write it to, rather than allowing
      module to determine this
- modules now decide whether darshan-core should track record
  names by passing either a valid name or NULL to the register
  function
    * this allows modules to register records that aren't really
      tracked by darshan-core
- only the POSIX module has been ported to new changes for this
  commit

also some code clean up to change references of record hash
to name hash to be more clear about what it is actually storing.

this calculation was inadvertently accounting for metadata time
twice. also, clarified timing descriptions for slowest unique
file perf calculations.

this functionality is helpful especially on mira/cetus where
the mpich libraries are named depending on the fortran version
and the type of compiler being used (gcc, xl, etc.).

- read and write start timestamps should record start timestamp of first
  read/write to *begin*, not start timestamp of first read/write to
  complete
- mirrors earlier fix to open timestamps

- be careful to record start timestamp of first open to start rather
  than start timestamp of first open to complete

mount information is now sorted in order of descending mount
point lengths to fix a bug in matching file names to mount
points. the longest mount point string which matches the file
name is the corresponding mount. this change requires slight
changes to the darshan-logutils API, but does not modify the
log file format

Otherwise, compile errors are generated when building an external
tool that uses the darshan-logutils api/library.

Darshan now provides debugging tips for cases where log files
are not generated.