- update Inline dep for ILSM-finding bug

2.019106 2019-04-28
 - fix doc-install oneliner

2.019105 2019-04-26
 - Inline::Pdlapp removed, in its own distro/repo now

2.019104 2019-04-26
 - include @d-lamb complex_atan2 improvements

2.019103 2019-04-22
 - got the version number wrong in previous release

2.019102 2019-04-22
 - Inline::Pdlapp can take package param to put XS subs into

2.019101 2019-04-21
 - make t/badvalue_scalar_cmp.t, t/pdl_from_string.t less fragile
 - de-corrupt FITS file production
 - make not blow up on ActivePerl (zap Cwd::abs_path on non-existent dir)

2.019100 2019-04-21
 - full transition to PDLA, up to date with legacy PDL master
 - change docs to remove $a and $b since that clashes with Perl sort vars
 - CI now measures test coverage for XS and C code
 - fix PDL::Lite so it successfully exports working functions
 - range speedups by @d-lamb
 - compatibility with Perl's experimental bitwise feature
 - pdladoc improvements
 - fix borover and bandover with badvalues
 - fix XS stack-handling bug revealed by Perl 5.28 compiled with debugging
 - more portable tilde expansion

2.019000 2019-04-20
 - cat returns a piddle of the highest input type, not
   the first.
 - rfits/wfits now do tilde filename expansion.
 - det in MatrixOps now uses the cached lu_decomp
   value correctly.
 - Multiple minor fixes, docs, cleanup, and changing
   things to point to github away from sf.net.

2.018001 2019-04-20
 - missed PDLA_INVALID as Type

2.018000 2019-04-20
 - SF.net Bugs fixed:
   - 429 Alien::Proj4->load_projection_information parses PJ_sch parameters incorrectly
   - 432 Work around List::MoreUtils bug t/gd_oo_test.t started failing in testers.
   - 433 GSL SF errors kill perl and pdl2/perldl shell
   - 434 Missing indx type in heading list of conversions.
 - Build improvements:
   - Add requirements and better test diags for PDLA::IO::GD.
   - Apply patch so PDLA::Lite can be used more than once
     Thanks to Shawn Laffan for the patch.
   - Fix build on perl <= 5.14 by adding ExtUtils::ParseXS to
   - Make coretarget generate parallelisable make deps Make
     "core" target be generated by separate function
   - Fixes for coming removal of . from @INC for PDLA build.
     Thanks to Todd Rinaldo for the first patch and raising the issue.
 - Various updates to the documentation.

2.017000 2019-04-20
 - Bugs fixed:
    - 379 Passing qsort an extra argument causes a segfault
    - 393 Tests may fail if perl is compiled with -Duselongdouble
    - 409 PDLA demos with PGPLOT display ignore $ENV{PGPLOT_DEV}
    - 413 PDLA::Core::Dev::pdlpp_postamble() cannot handle .pd files in subdirectories
    - 419 t/#pdl_from_string.t fails on long double systems
    - 421 PDLA::IO::FITS can't handle 64-bit integers (longlong, indx)
    - 422 PDLA misc. compiler warnings.
    - 423 wcols FORMAT option always incorrectly gives error
    - 424 Calling PDLA on a list of piddles fails to propagate bad values
    - 425 svd is broken for everything but 2x2 matrices
 - Several patches contributed from the Debian team have been applied
   that fix documentation spelling errors, make PDLA builds more
   reproducible, and will make packaging PDLA easier.
 - More helpful error message when multi-element PDLA is used in
   a boolean expression (feature request #84)
 - Improve argument size handling and documentation for rle
   (feature request #80)
 - One can now use $pdlc = pdl($pdla,$pdlb) when $pdla and/or $pdlb
   have BAD values and the BAD values will propagate to
   $pdlc. Previously this would only work with a copy of a single
   piddle (e.g., $pdlc = pdl($pdla) ) or with 'cat'.
 - Many changes have been made to make PDLA compile more cleanly and
   emit fewer compiler warnings.
 - Quiet printf format warning for sizeof()
 - Deprecate finite in favor of isfinite.
 - Many cleanups and additions to the test suite to use Test::More
   and to add meaningful test descriptions.
 - Added subtests to primitive.t
 - Add PDLA::Doc::add_module to pdlpp_postamble

2.016000 2019-04-19
 - Bugs fixed:
   - 416   PDLA::PP creates .pm & .xs files before pp_done is called
   - 414   ccNcompt (i.e. cc4compt and cc8compt) breaks with byte data type
 - All collapse operators now have "<name>over" equivalent names.
   This ends an API wart in which most, but not all, of the collapse
   operators had a short form that did full collapse and a long form
   that did 1-D collapse only (e.g. "and" collapses to a point,
   while "andover" collapses by one dimension).  The exceptions are
   left in for legacy code, but now have regularized "-over" forms
   as well:
      average       -> avgover
      daverage      -> davgover
      maximum       -> maxover
      maximum_ind   -> maxover_ind
      maximum_n_ind -> maxover_n_ind
      minmaximum    -> minmaxover
      minimum       -> minover
      minimum_ind   -> minover_ind
      minimum_n_ind -> minover_n_ind
 - PDLAs with dataflow can now be reshaped.  The dataflow
   connection gets severed by the reshape operation.
 - PDLA::IO::FITS now works better with RICE-compressed FITS
   images, such as are produced by NASA's SDO project.
       - The NAXIS- header keywords are now replaced by their
         ZNAXIS- equivalents, so the NAXIS fields in the header
         are correct after the image is read in.
       - The Z-, TFIELDS, TTYPE*, and TFORM* keywords are now
         deleted from the header object, so that the uncompressed,
         loaded image does not have leftover compression headers.
 - The language preprocessor PDLA::PP now does not automatically
   call pp_done for modules that do not call pp_done themselves.
   This new, stricter behavior requires module authors to call
   pp_done at the end of their PDLA::PP file.  This prevents
   partially-complete .xs and .pm files from being written if
   there is a module build error.
 - PDLA::whichND returns PDLAs of Indx type, to avoid overflows.
 - Empty piddles are handled slightly differently now by
   PDLA::info and `help vars'.  Empty piddles are different
   from null piddles, and now generate different info strings.
   (null piddles lack data or dimensions; empty piddles have
   at least one dimension of size 0).
 - There is now a .gitattributes file so GitHub repo language
   stats are more accurate.
 - The PDLA SF/GitHub deveolpment workflow is integrated into
   the DEVELOPMENT docs.

2.015000 2019-04-19
 - Fixes to improve compile and test of F77 modules
   on OS X (i.e. don't build i386 versions)
 - Basic/Ops/ops.pd - make compatible with MSVC++ 6.0
 - Fix missing PDLA license information
 - Fix sf.net bug #403: reshape() failed when applied to a
   piddle with dataflow.  Now, changing the shape of a PDLA
   that already has dataflow is prevented and an error message
 - Fix sf.net bug 406 Added missing logic for clump(-N) and minor
   cleanup of perl wrapper.
 - force new_pdl_from_string to return a piddle with P=physical flag
 - Add $PDLA::indxformat for PDLA_Indx This avoids loss of
   significance when 64bit PDLA_Indx values are printed.
   Make new_pdl_from_string() avoid converting IV values
   to NVs This allows pdl(indx, '[ 9007199254740992 ]')
   to work without rounding due to the 52bit double
   precision mantissa versus the 63bits + sign for PDLA_Indx.
 - Add type support info to pdl() constructor documentation.
   pdl() has always supported this usage but examples were
   missing in the documentation.

2.014000 2019-04-19
 - PDLA can now create and use piddles having more
   than 2**32 elements.  Of particular use is that
   you can now use PDLA::IO::FlexRaw to memory map
   piddles from files on disk which can allow for
   simplified processing and IO on extremely large
 - PDLA now detects when your perl installation has been
   built with floating point longer than 8 bytes and gives
   a one time warning that precision will be lost converting
   from perl NV to PDLA_Doubles.  This warning is given on
   "use PDLA;"
 - Skipping t/bigmem.t to avoid OOM CPAN Testers fails.
 - Minor fixes to C code to meet stricter compiler and C99 requirements.
 - A new ipow() method has been added with 64bit integer
   support.  ipow() calculates the integer exponentiation
   of a value by successive multiplications.  This allows
   one to avoid loss of significance in integer exponents.
   pow() converts the value to double and will always have
   <52bits precision.
 - Corrected return types of intover, borover, bandover
 - nbadover and ngoodover now return indx type (PDLA_Indx)
 - Fixed compile problem in clang from using finite() on an integer datatype.
 - Mark some failing tests in t/primitive.t as TODO
   to avoid CPAN Testers failures.
 - Implement PDLA_Anyval data type as union to support 64bit indexing
   This adds a union data type to add 64bit integer support to
   the original PDLA-2.x types which assumed that double was
   capable of holding all the "lesser" types.  With the PDLA_Anyval
   type, we can correctly handle 64bit integer data types and
   avoid errors and loss of precision due to conversions to or
   from PDLA_Double.  Special thanks to kmx and zowie for their
   help to complete and debug this implementation.
 - Due to the new PDLA_Anyval type changes, existing
   PDLA::PP modules may need to be updated to support
   the new PDLA Core version 12 API.  Authors, please
   see PDLA::API for macros to ease the porting.
   Users, be aware that some modules may not work
   until they are updated by their maintainers.
 - Many fixes to the build process to make PDLA more robust
   to build, test, and install.  This takes advantage of new
   automated CI testing via Travis CI on the github site.
   Thanks much to Ed and Zakariyya for their work to get
   this started and maintained.  This CI testing makes
   this the best tested and best testing PDLA release ever!
 - Various documentation clean-ups and work to improve
   on-line viewing at http://metacpan.org and others.
   (Thanks kmx and Derek!)
 - "use PDLA" now includes "use PDLA::IO::Storable"
   This avoids a hard to diagnose crash that can
   occur when a user tries using Storable without the
   necessary "use PDLA::IO::Storable".
 - MANY sf.net bugs fixed:

   400  dataflow slice crash around 2**31 boundary
   399  Small doc fixes
   398  $pdl->hdr items are lost after $pdl->reshape
   396  online docs for modules only work first time in PDLA shells
   395  ipow (integer exponentiation) support for 64bit index support
   394  demo cartography fails
   383  gcc/#gfortran 4.9.2 needs -lquadmath
   378  where on dice of pdl bad results
   376  PDLA Segmentation fault working with Storable restored PDLA
   347  t/#pdl_from_string.t has a failure if BADVAL_NAN=1
   346  ExtUtils::F77 dependency causing problems for CPAN install
   343  longlong constructor and display lose digits due to...
   340  orover of byte data returns long type

2.013009 2019-04-19
 - fix non-PDLA-ed test

2.013008 2019-04-19
 - restore PDLA::IO::Misc, ::FITS, ::Pic

2.013007 2019-04-15
 - rename pdl2 exe to pdla2, pdldoc to pdladoc, perldl exe to perldla,
   pdl exe to pdla
 - move Perldl2 into PDLA::Core
 - . in @INC update

v2.013006 Sun Aug 30 15:33:09 BST 2015
 - further tweak to how $VERSION recorded/used

v2.013005 Sun Aug 30 03:43:01 BST 2015
 - minor tweak to how $VERSION recorded/used

v2.013004 Sun Aug 30 03:00:57 BST 2015
 - PDLA::Core to separate distro and repo

v2.013003 Sat Aug 29 22:28:17 BST 2015
 - fixup $VERSION in PDLA::Core more

v2.013002 Sat Aug 29 21:58:05 BST 2015
 - fixup $VERSION in PDLA::Core
 - "make core[test]" now can do parallel build which will be quicker
 - use Alien::HDF4
 - big test cleanup
 - improve PDLA::Graphics::PGPLOT build robustness

v2.013001 2015-08-29 19:00:00+01:00
 - TODO fragile t/lut.t test
 - stop claiming packages C::StructObj C::StructType C::Type C::Var Inline::MakePdlppInstallable Inline::Pdlpp SymTab Win32::DDE::Netscape XS

v2.013-PDLA 2015-08-29 04:00:00+01:00
 - fork from PDL
 - rename all PDL -> PDLA