Users can now pass --enable-mpit-pvars to configure to statically
enable/disable MPIT performance variables so as to limit performance
impacts only to those variables they need.

Current valid arguments are as follows:
    none     - No performance info recorded
    recvq    - All message queue-related variables
    nem      - All nemesis-related variables
    all      - All variables above

A macro framework is added to facilitate the addition of new variables
(see MPIR_T_GATE on mpid_common_statistics.h).

Also add a check to the MPIT nemesis test to ensure that the variables
being tested have been registered with the runtime.

The following MPIT performance variables are added:

 - nem_fbox_fall_back_to_queue_count;
 - time_failed_matching_postedq;
 - time_matching_unexpectedq;
 - unexpected_recvq_buffer_size.

Some common MPIT-related routines, handle creation functions and
enable/disable switches are gathered in a common file.

Two test cases are furnished, but not yet integrated with 'make check'.

Reviewed-by: rgunter

These five tests are currently failing on a regular basis with MPICH.
All need to be dealt with at some point, but we cannot fix them right
now and they make it hard to spot new failures in the tests.

references #1767, #1788

Reviewed-by: dinan

This is used to indicate tests that are correct, but functionality that
is expected to fail in the current version of the MPI implementation.
This is expressed in `testlist` files with an argument string like
`xfail=ticketXYZ` or `xfail=REASON` after the program name.
Unfortunately, spaces, quotes, and `#` characters are not permitted in
the format at this time, due to the primitive nature of the testfile
parser code.

If strict MPI testing is requested (by configuring the test suite with
`--enable-strictmpi`) then the `xfail=BLAH` arguments are ignored under
the assumption that they are only accurate for the current version of
the MPICH implementation.

No xfail support has been added to the XML output at this time.

Reviewed-by: dinan

At Jim's request...

Reviewed-by: dinan

This way tests that were skipped because of strictness or version
constraints will show up in the summary.tap output file.

Reviewed-by: dinan

fixup for 51b72604

No reviewer.

fixes #1786

No reviewer.

Should allow PETSc to decide whether or not they can safely use
nonblocking collectives with ch3:sock.

Reviewed-by: buntinas

We made progress for MPI_Wait and friends, but not
MPI_Test/MPI_Iprobe/etc.  Thanks to Jed Brown for letting us know about
this issue.

fixes #1785

Reviewed-by: buntinas

Reviewer: goodell

Reported and largely debugged by Chris Daley (thanks!).

fixes #1784

No reviewer.

Print a format that will display better in most TAP renderers,
particularly Jenkins.

No reviewer.

No reviewer.

Prevents error messages in the crons like:
```
cannot remove path when cwd is /tmp/yS2BmJPSIF for /tmp/yS2BmJPSIF:  at
/usr/share/perl/5.14/File/Temp.pm line 902
```

No reviewer.

No reviewer.

0e247915.

No reviewer.

No reviewer.

Some TAP parsers are OK with omitted test numbers, but the Jenkins TAP
plugin is not.  Consider this a fixup for bbecd7e9.

No reviewer.

`runtests` now also emits a `summary.tap` file that is formatted using
the Test Anything Protocol (TAP).  I do not love this first version,
since it's comment-heavy and doesn't take advantage of the YAMLish
support within the TAP format.

This is a big first step towards allowing the MCS Jenkins CI server to
run some continuous integration testing for us.  Other enhancements in
this vein should ideally include "XFAIL" (and possibly "SKIP") support.

No reviewer.

This reverts commit ba2e4444.

This was causing `accfence1` (and other) failures.  Reverting for now,
since the apparently easy fix is not quite as simple as it first
appears.

We weren't chomping the newline, and we didn't want `--tags` since this
includes non-annotated tags like `last-svn-trunk-commit`.

No reviewer.

The interaction with git was unnecessarily conservative and caused
problems for both the release.pl script and our homebrew formula.

Reviewed-by: balaji

Partially reviewed by balaji@.

When no less than 4 tasks in the same group call MPI_Scatter with
recvcount less than sendcount, non-root tasks will pass a negative count
to MPIC_Send_ft, which won't always be caught by the device.  This adds
a bit of extra error checking to help catch this case.

Based on a patch written by Qi QC Zhang <keirazhang@cn.ibm.com>,
augmented by Dave Goodell <goodell@mcs.anl.gov>.

This commit only improves error checking in the ACC/GACC paths.  It does
not add or remove any actual functionality available to non-erroneous
programs.

Suggested by Jed Brown:
http://lists.mpich.org/pipermail/discuss/2012-December/000214.html

Reviewed by dinan@.

A few small changes to tighten up the error checks in the interprocess mutex
code.

Replaced SETANDJUMP with CHKANDJUMP, so we actually check for a pthread error
before signalling an MPI error.

Reviewer: none

On niagara1, `cp -a` does not work.  But the equivalent `cp -pPR` does.

No reviewer.

Updated the MCS mutex test to work with both regular and shared memory windows.

Reviewer: buntinas

A new test case that uses the Mellor-Crummey and Scott mutual exclusion
algorithm.

Reviewer: buntinas

Reviewer: buntinas