pfunit-announce Mailing List for pFUnit

I was hoping to give this transition a bit more fanfare, but a few recent issues have caused me to rush my plans to start using GitHub to distribute pFUnit.

The new home is   https://github.com/Goddard-Fortran-Ecosystem/pFUnit

 Users should consider the sourceforge repository as read only going forward.   All bug fixes and new features will be delivered through the new home.    At some point the SourceForge mailing lists will be deactivated.  I will be using GitHub’s Issues, Pull Requests, and the site wiki to manage development.      

—————————————

I am also making progress on a 4.0 release of pFUnit.      If you look on the GitHub develop branch you’ll find that the build is generally much cleaner than before (fewer warnings), but it does require relatively recent compilers.           Hopefully soon I’ll start documenting the planned 4.0 features in the wiki. 

This is just a bug-fix release:

3.2.8   September 5, 2016
	- Patch for PGI system() call  (sf patch #9)
	- Patch for deffered length strings (sf patch #10)
	- Increased required version of cmake from 2.8 to 3.0
	  It was alreaddy effectively required - just not noticed.

In case you have colleagues that may be interested, NAG is providing a Fortran modernization workshop that
includes a session about pFUnit:

http://www.nag.co.uk/content/fortran-modernization-workshop

pFUnit 3.2.6 fixes some defects involving pointers and target attributes
that were exposed by the recent upgrade to NAG 6.1.   

Note 1: NAG users may need to update to the latest NAG 6.1 patch as well.
Node 2: The bugs fixed here are generally harmless for most builds under
        most other vendors due to common.

My sincere thanks to NAG support for helping to identify the nature and
location of these longstanding, but subtle, defects.

Over the last week there has been several rushed fixes that were
themselves broken.  

For this I truly apologize.  Most of the problem was simply that
I was rushing the process and not waiting for our regression
tests to run.  (Or not correctly intepreting why they were not
completing.)  But that investigation revealed a merge mistake
that also broke support for some older compilers, so at least
some good came of all this.

I very much hope that this is the final release of pFUnit 3.x.
Beta releases of 4.0 should start appearing over the summer.
However, as I expect to push hard on advanced F2008 features in
that release, backward compatibility will not be established
until a formal 4.0 release some time further down the road.

Thank you for your patience and your support,

- Tom

After the last push we added ifort 16 to our regression suite and discovered
an issue when using OpenMP.  This release is primarily a workaround for that
issue, but contains some minor cleanup that was done while diagnosing the
issue.

More details:
	- Workaround for ifort 16.0.2 bug with openmp
	- Various minor improvements to code:
          . Fixed inconsistent names in self tests.
          . Introduced "-qopenmp" in find_package for OpenMP
          . RemoteProxy now ignores output starting with "DEBUG:" - useful
            for debugging self tests.

I did not read the regression log and did not notice that recent merges broke the examples.
This push fixes a minor (but crucial) bug in driver.F90 that otherwise breaks most use cases for pFUnit.

Sorry about the complication.

I am pleased to announce the release of 3.2.0.   ChangeLog is shown below.


3.2.0	April 21, 2016
	- Extension: support test to run on "all available" pes.
	  . This is primarily aimed to enable testing coarray Fortran procedures
	    CAF does not yet have "teams", so tests must use all images.  But
	    MPI users may find it useful as well.   Just use "*" instead of a number
	    when specifying NPES.   True CAF support should be expected in release 4.0.0.
	- Extension: support test case with custom constructor.   With this extension was
	  able to create external extensions to the framework that allow testing ESMF grideded
          components.  Contact Tom Clune if you want to have those extensions - did not want to
	  induce an ESMF requirement for pFUnit.
	- Improved support for CMake (contributed by Pal Levold)
	  . packaging
	  . add_pfunit_test() macro
	  . Ctests (CMake testing package)
	  . NOTE: now requires a more recent CMake version.
	- Increased max filename length to avoid truncation
	- Disabled --verbose command line option (breaks under gcc)
	- Cleanup to reduce/eliminate compiler warnings when building.
	- There is a regression in NAG 6.1, so NAG users should continue to use 6.0 until a
	  fix or a workaround are found

This release contains various minor fixes and compiler workarounds 
that have been accumulating.  Perhaps the most noteworthy item is
not really a charge in pFUnit itself, but that PGI 15.7 appears to
now robustly work with pFUnit.  (15.4 was able to pass the self tests,
but struggled with even simple user applications.)

A Cray port remains imminent and is available on a feature branch, where
a few annoying workarounds are required that we don’t want polluting the
master branch.  We expect that in November pFUnit 3.2 will contain support
on the master branch.

- Tom

We are very happy to announce that pFUnit now works with the latest 
PGI compiler (v. 15.1).    This has been perhaps the most common user
request over the years, and we hope this enables useful work for many
people.

This release also includes a number of minor extensions that are covered
in the change log:

3.1 March 20, 2015
- PGI 15.1 now supported.
- Asserts over integer arrays now supported. INT32 and INT64 support added.
- Consolidated assertAssociated directives to:
   @assertAssociated
   @assertNotAssociated
- Added ifndef option to preprocessor directives.
- Fixes: Name length checking, unbalanced allocate, python 3 basestring.


Enjoy,

- Tom

This release fixes a few recently uncovered bugs and provides a variety of relatively minor feature enhancements.  

Also, we are now optimistic than pFUnit will be usable with the next release of the PGI Fortran compiler.  Expected in January.

Changes:
========
* Bug fix: Corrected lack of PRESENT check on some optional arguments.

* Extended maximum ranks of integer arrays to match that of FP arrays for assertEqual

* Added the following directives - meaning should be obvious from context
   @assertEquivalent(...)
   @assertEqual(a,b)
   @assertAssociated(...), @assertUnAssociated(...)
   @assertAssociatedWith(...), @assertUnAssociatedWith(...)

* Added code to parse brackets in directive arguments,
   allowing @assertEquivalent([...],[...]).
   Needed for directives that must parse arguments to construct other calls.

* Extended assertTrue and assertFalse to cover arrays of logical.

* Removed dependency on CPP stringification in the REFLECT macro simplifying build.

* Bug fix: Improved portability of build, fixing problem with OUTPUT_FLAG, i.e. “-o”.

* Bug fix:  Fixed build problem using latest NAG compiler under OS X (cmake/gmake)

* Replaced explicit invocations of python with $(PYTHON), set in GNUmakefile,
   to aid specification of python version.

* Removed an extraneous allocate (Patch 5).


Cheers,

- Tom

This release has a number of minor bug fixes and feature improvements over 2.1.0.   

As mentioned previously, one design improvement exposed an unfortunate bug in
currently released versions of GFortran (4.8.2 and earlier).   The lack of backward
support for those earlier compilers is why this is a major release rather than a minor
release.  pFUnit will build with the trunk of 4.9.0 and we are confident that it and
4.8.3 will be supported when hey are formally released.  Should be very soon.

Noteworthy changes in this release:

- Improvements to default driver
  . Produces a useful return code under unix in serial (and some MPI variants)
  . Additional command line options: --debug, -o <file>, -xml, --help
- Improvements to examples
  . Now covered under regression testing (were relatively unstable before)
  . Makefiles are more portable
- New assert methods for real variables (<,<=, >,>=)
- Improvements to parser
  . Refinements to parser syntax
  . Several bug fixes for rare cases that had not been covered by tests
  . Regression of examples provides better coverage for parser than previously


We still see occasional issues with robust support of OO features even in those
compilers that we claim to support.  We are happy to refine and submit a bug
report to the vendor(s) if you can send us a reproducer.

Because some users may find some upcoming changes to pFUnit to be disruptive to their own ongoing development, I have carved off a frozen branch ('pfunit_2.1.0') in the git repository.    Some details are provided below, but planned changes will require more recent versions of certain compilers and are likely to require syntax changes in the pFUnit preprocessor for user-defined test cases.

Highlights of new features in 2.1.0 include the following:

 * Significant documentation and additional examples
 * Improvements to portability
   - Port to Windows OS (thanks to Evan Lezar)
   - Port to XLF compiler (thanks to Sean Patrick Santos)
 * Option to use CMake to build/install (beta)
 * Basic framework extensions
   - Expansion of assertions
     . Equality of complex numbers
     . Detect NaN/Infinity
     . Equality of multidimensional integer arrays
   - Limited support for threading with OpenMP
 * Improvements to pFUnit parser:
   - @mpiAssert - enforces synchronization among processes
   - Support for conditional compilation of individual tests controlled by CPP flags
   - Mods to support python 3.x

Upcoming mods that break backwards compatibility:

 * Generating a proper unix return code.  Until now pFUnit returns a Unix "success" regardless of test failures.   Unfortunately the mods for this exposed a subtle bug in gfortran.   The developers were able to quickly patch the compiler but the patch will not appear until 4.8.3 and/or 4.9.0.  Users that need to use older versions of gfortran should stick with pFUnit 2.1.0.

 * Major simplification for user-defined TestCases.   The current version of pFUnit requires users to create an annoying amount of nearly identical boiler plate code for each new extension of TestCase:  pointer component, type-bound procedure, and a constructor.   Upcoming changes will automagically handle all of this in the parser.   Backward compatibility might be possible, but users will be strongly encouraged to delete the unused boiler-plate code from their test suites.


I would like to voice my great appreciation for the numerous suggestions, feature requests, and patches provided by users.  Please keep these coming so that we can all benefit.

- Tom

The previous announcement had artifacts that made the message difficult to read in the archive on the sourceforge site. This posting is an attempt to provide the same information in a more accessible format.  Apologies for the duplication for those that are on the distribution list.

----------------------------------------------

This release is a complete rewrite of pFUnit that is intended to be easier to use and far more extensible than its predecessor.    

A quick highlight of new features:

- Extensibility: Heavy use of Fortran 2003 object-oriented features allows users to extend the framework for their own purposes.  We encourage developers to send us useful extensions.

- Customized preprocessor for user tests:  Simplifies creation of test procedures and use of assertion routines.  For those familiar with JUnit, this preprocessor provides things that look like annotations. Specific capabilities include:

	. @test - indicate next procedure is a unit test
	. @mpiTest(npes=[1,2,5,…]) - indicate next procedure is a unit test requiring MPI and to be run on 1,2,5,… processes
	. @assertEqual(a,b) - provides file and line number for failing tests and returns without executing remaining assertions in the test.

- Simplified parameterized tests:  While still not quite as simple as I would prefer, hopefully elegant enough to be useful to others.  See the example in ./Examples/ParameterizedTest

- Robust runner:  This allows the framework to actually detect failing and hanging tests by having one executable monitor another.   Still needs some tweaks for MPI, as the startup time can make the framework think a test is hanging.  (This has been fixed since the release - I think.)

- More assertions:   We still provide assertEqual for real/complex through rank=5.   But we have added a number of other assertions that would have been nice for situations that arise frequently in practice.   assertAll(), AssertAny(), …   Still more will be added soon such as assertIsMember(), AssertContains(), assertIsPermutation(), etc.


--------------

For those that are interested in the details, the overall design follows closely that of JUnit 4.11.

At this moment, there is very little in the way of useful documentation.  However, the examples in ./Examples are being maintained and are hopefully self explanatory to those with familiarity with the earlier pFUnit or other xUnit frameworks.

To obtain the code use git to clone - the code is on the master branch.   If you want stability, you may want to checkout the pfunit_2.0_beta branch which will only update for bug fixes for this release.

This release should work well under ifort 13.1, gfortran 4.8, and nag 5.3 (907) under Linux and OS X.    Because we are pushing the limits of compiler support for F2003, users should expect to stay current with compiler releases from these vendors.

We very much want feedback - please post on sourceforge if you have issues, questions, requests, or even kudos to share.

Cheers,

- Tom



Thomas Clune, Ph. D. 					<Tho...@na...>
Chief, Software Systems Support Office		Code 610.3
NASA GSFC								301-286-4635
MS 610.8 B33-C128						<http://ssso.gsfc.nasa.gov>
Greenbelt, MD 20771

Hi Tom,
maybe it is my web browser (Firefox) or something I am doing wrong.
Your posting on this newsgroup is unreadable when viewed on-line:
http://sourceforge.net/mailarchive/forum.php?thread_name=83D7F160-F87F-465E-94C2-451D3CE74E7B%40nasa.gov&forum_name=pfunit-announce

There are no line breaks and there seems to be no way to scroll right to see the ends of the lines.

The only way I could view the posting was to open it up by selecting:
Attachments: Message as HTML

regards
Mike

Dr. Mike Rezny                    Model Framework developer
Met Office  Fitzroy Road  Exeter  Devon  EX1 3PB  United Kingdom 
Tel:+44 (0)1392 88 6236
Mik...@me...       http://www.metoffice.gov.uk

All,

This release is a complete rewrite of pFUnit that is intended to be easier to use and far more extensible than its predecessor.    

A quick highlight of new features:

Extensibility: Heavy use of Fortran 2003 object-oriented features allows users to extend the framework for their own purposes.  We encourage developers to send us useful extensions.
Customized preprocessor for user tests:  Simplifies creation of test procedures and use of assertion routines. For those familiar with JUnit, this preprocessor provides things that look like annotations. Specific capabilities include:
@test - indicate next procedure is a unit test
@mpiTest(npes=[1,2,5,…]) - indicate next procedure is a unit test requiring MPI and to be run on 1,2,5,… processes.
@assertEqual(a,b) - provides file and line number for failing tests and returns without executing remaining assertions in the test.
Simplified parameterized tests:  While still not quite as simple as I would prefer, hopefully elegant enough to be useful to others.  See the example in ./Examples/ParameterizedTest
Robust runner:  This allows the framework to actually detect failing and hanging tests by having one executable monitor another.   Still needs some tweaks for MPI, as the startup time can make the framework think a test is hanging.  (I know the fix, but have not yet implemented it.)
More assertions:   We still provide assertEqual for real/complex through rank=5.   But we have added a number of other assertions that would have been nice for situations that arise frequently in practice.   assertAll(), AssertAny(), …   Still more will be added soon such as assertIsMember(), AssertContains(), assertIsPermutation(), etc.

For those that are interested in the details, the overall design follows closely that of JUnit 4.11.

For the moment, there is very little in the way of useful documentation.  However, the examples in ./Examples are being maintained and are hopefully self explanatory to those with familiarity with the earlier pFUnit or other xUnit frameworks.

To obtain the code use git to clone - the code is on the master branch.   If you want stability, you may want to checkout the pfunit_2.0_beta branch which will only update for bug fixes for this release.

This release should work well under ifort 13.1, gfortran 4.8, and nag 5.3 (907) under Linux and OS X.    Because we are pushing the limits of compiler support for F2003, users should expect to stay current with compiler releases from these vendors.

We very much want feedback - please post on sourceforge if you have issues, questions, requests, or even kudos to share.

Cheers,

- Tom

Thomas Clune, Ph. D. 					<Tho...@na...>
Chief, Software Systems Support Office		Code 610.3
NASA GSFC								301-286-4635
MS 610.8 B33-C128						<http://ssso.gsfc.nasa.gov>
Greenbelt, MD 20771