WCSLIB version 4.14 (2012/07/13)
--------------------------------

* C library

  - Problems were caused for fitshdr(), wcsbth(), wcspih(), wcsulex(),
    datfix() and wcshdo(), by locales such as "fr_FR" in which commas
    are used to delimit the decimal part of floating point numbers.
    In particular, these locales adversely affect the behaviour of
    sscanf() and sprintf() when reading and writing FITS header
    keyvalues.  Thread-safe fixes provided by Michael Droettboom.

  - Applied astropy patches from Michael Droettboom: attend to compiler
    warnings in prj.c, spc.c, tab.c, and wcsutil.c, including those for
    MS-Windows.


WCSLIB version 4.13.4 (2012/04/02)
----------------------------------

* Installation

  - Relaxed the closure tolerance slightly in tlog.f as for tlog.c
    previously.


WCSLIB version 4.13.3 (2012/03/26)
----------------------------------

* Installation

  - Relaxed the closure tolerance slightly in tlog.c for gcc 4.6.3 on
    Ubuntu 12.04 (reported by Ole Streicher).


WCSLIB version 4.13.2 (2012/03/21)
----------------------------------

* Installation

  - Changes to configure.ac and the GNUmakefiles to put '-lm' strictly
    in the correct sequence when producing the sharable library
    (reported by Ole Streicher).


WCSLIB version 4.13.1 (2012/03/15)
----------------------------------

* C library

  - Workaround in wcserr_set() for an unfortunate compiler segv from
    gcc 4.2.1 in MacOSX 10.7.


WCSLIB version 4.13 (2012/03/14)
--------------------------------

* C library

  - Allow naxis == 0 in wcsini() and linini() for a degenerate
    coordinate system as may be produced by wcssub().  In wcssub(),
    ensure that wcsini() is called for coordinate systems with
    naxis == 0.  In wcssptr(), call spcfree() before spcini() to plug a
    memory leak.


WCSLIB version 4.12 (2012/02/29)
--------------------------------

* C library

  - In spctrne(), guard against restfrq == restwav == 0.0 when
    translating between two velocity-characteristic types, or between
    two wave-characteristic types (the translation is independent of
    restfrq and restwav, but a dummy value is needed for intermediate
    calculations).


WCSLIB version 4.11 (2012/02/21)
--------------------------------

* C library

  - Bug fix in spcset() for handling simple linear spectral axes.

* Fortran wrappers

  - Bug fix in spctrne_() for handling ctypeS2.


WCSLIB version 4.10 (2012/02/06)
--------------------------------

* C library

  - datfix() and spcfix() now return informative messages when dates and
    AIPS-convention spectral axes are translated (changes contributed by
    Michael Droettboom).  spcaips() now returns an error status for
    invalid values of VELREF.

  - wcssub() has been augmented with the ability to add new axes onto a
    wcsprm struct.


WCSLIB version 4.9 (2012/01/24)
-------------------------------

* C library

  - Fixes to wcsfixi() for collecting the messages properly in the info
    array (from Michael Droettboom).

  - Handle certain malformed date strings more gracefully in datfix().

  - Make informative messages printed by wcserr_prt() a bit more
    informative.


WCSLIB version 4.8.4 (2011/12/05)
---------------------------------

* C library

  - Fixed the pseudo-random number generator in twcstab.c - gcc 4.6 with
    '-O2' baulked at testing for signed integer overflow.

* Installation

  - In configure.ac, the Fortran compiler's libraries must be added to
    the link list when checking for the PGPLOT libraries since gcc is
    driving the linker.  Likewise in C/GNUmakefile when linking test
    programs that use PGPLOT.

  - Use 'make CHECK=nopgplot check' to run only the non-graphical tests
    (even if PGPLOT is available).

  - After compiling and running the tests, 'make check' now summarizes
    the non-graphical test results and stops if any failed.


WCSLIB version 4.8.3 (2011/11/17)
---------------------------------

* C library

  - Minor generalization of the wcserr diagnostics to allow the return
    of informative messages which are associated with negative status
    values.  wcserr_prt() will now recognize and print them as such.
    Added wcserr_clear() to reset (clear) a wcserr struct.

  - Modified unitfix() to return an informative message if a units
    alias is applied, and wcsfixi() to allow such messages to be
    propagated through the info array (from Michael Droettboom).

  - Modified twcsfix.c to use wcserr diagnostics, in particular to
    report units alias translations.

* Fortran wrappers

  - In wcsfix_(), interpret *naxis == 0 as meaning naxis == 0x0, thus
    causing cylfix() to be skipped.

  - Modified twcsfix.f to reflect changes made to twcsfix.c.


WCSLIB version 4.8.2 (2011/10/04)
---------------------------------

* Installation

  - Changes for Debian package generation contributed by Ole Streicher:
    - Corrections to 'configure' reported by 'lintian'.
    - Generate man pages for the utility programs and install them.


WCSLIB version 4.8.1 (2011/09/19)
---------------------------------

* Installation

  - Set SONAME in the sharable library in accordance with
    tldp.org/HOWTO/Program-Library-HOWTO/shared-libraries.html
    (reported by Ole Streicher, Debian package maintainer).  The
    sharable library will again be installed with full release number
    and with a symbolic link from SONAME pointing to it.  If defined,
    SHRLN will point to SONAME.


WCSLIB version 4.8 (2011/08/15)
-------------------------------

* C library

  - New error diagnostics mechanism contributed by Michael Droettboom:

    Most functions that return a numeric status (error) code now also
    write a detailed error message to a wcserr struct attached to the
    passed-in *prm struct.  See wcserr.h for more information.

    Functions that didn't have a *prm struct as an argument have no
    direct way to return an error message.  Therefore, these functions
    now have duplicate versions with the suffix "e" that take an
    additional "struct wcserr *" parameter.  These functions are:

        spcspx()    ->  spcspxe()
        spctrn()    ->  spctrne()
        spctyp()    ->  spctype()
        spcxps()    ->  spcxpse()
        wcsulex()   ->  wcsulexe()
        wcsunits()  ->  wcsunitse()
        wcsutrn()   ->  wcsutrne()

    A new function wcsfixi() was added which is identical to wcsfix()
    but in addition stores all of the detailed textual messages about
    the fixes that were performed.

  - In wcssub(), ensure that wcstab->m_tab has been initialized
    before trying to free it on status return 12 or 13 (reported by
    Hans Terlow).

  - Bug fixes:
    - In sphx2s() and sphs2x() for the case where |eul[1]| = 180.0.
    - For parsing AIPS-convention VELREF in wcsbth().
    - In spcaips() for translating AIPS-convention CTYPEia without
      Doppler frame.

  - Non-graphical test programs now simply report "PASS" if they satisfy
    the reporting tolerance for closure residuals, etc.  Their full
    output is reported otherwise.  Run 'make MODE=interactive check' to
    revert to the previous behaviour of reporting the full output for
    tests that succeed.

  - Eliminated compiler warnings about type-punning of pointer-to-
    function.

* Fortran wrappers

    Extensive modifications to track the new error handling mechanism
    in the C library.

* Installation

  - configure now prefers gfortran over g77 if available.

  - Don't rely on "." being in the PATH if config.status needs to be
    run in the pgsbox and utils makefile (reported by Peter Teuben).


WCSLIB version 4.7 (2011/02/07)
-------------------------------

* C library

  - Bug fix in celset() for interpreting LATPOLEa when LONPOLEa = phi0.
    Crept in at version 4.4.

  - Fixed the bounds test on y in hpxx2s() (HEALPix projection) for
    unconventional values of H and K.  In hpxx2s() and hpxs2x(),
    corrected the offset of the southern polar half-facets for
    even K.  In hpxs2x(), put the phi = 180 meridian in the expected
    place.

  - Bug fixes in tabx2s() and tabs2x() for default indexes (reported
    by David Berry).  In tabs2x(), if no solution is found then do
    minor extrapolation past the ends of each row (1-D case only).
    Sped up tabs2x() by about 50%.

  - New functions wcsprintf(), wcsprintf_set(), and wcsprintf_buf(),
    declared in wcsprintf.h, provide control over the disposition of
    output from celprt(), linprt(), prjprt(), spcprt(), tabprt(), and
    wcsprt() routines.  Prompted by Michael Droettboom, with an initial
    implementation.

* Fortran wrappers

  - In the various test programs, used EQUIVALENCEs to ensure that the
    CEL, LIN, PRJ, etc. arrays are aligned on a DOUBLE PRECISION
    boundary.

* PGSBOX

  - Bug fix for the case where NG1 == 0 and GRID1(0) < 0, and likewise
    for NG2 and GRID2.

* Utilities

  - In wcsware, added a '-w' option to convert world coordinates
    obtained from stdin to pixel coordinates using wcss2p().  Allow
    multiple sets of input coordinates with the '-x' and '-w' options
    and report the value of the intermediate world coordinates.

* User manual

  - Fixed list formatting for function return values < 0 or > 9.

  - New section for wcsprintf() and related routines.

* Installation

  - Changes prompted by Sbastien Fabbro for the Gentoo Linux package:
      a) autoconf updates,
      b) respect LDFLAGS when building the shared library,
      c) install documentation,
      d) recognise DESTDIR for doing a staged installation.

  - As of this release, the minor WCSLIB version number (second field)
    will be incremented if and only if a change is made that affects the
    library itself, not the documentation or utilities.  The version
    number on the installed libraries and header files will omit the
    patch number (third field).


WCSLIB version 4.6.3 (2010/11/24)
---------------------------------

* C library

  - Bug fix in wcsbth() for handling the inheritance of image header
    keywords (uncovered by valgrind, reported by Jim Lewis).


WCSLIB version 4.6.2 (2010/11/22)
---------------------------------

* C library

  - Fixed a memory leak in wcsbth.l (reported by Jim Lewis).


WCSLIB version 4.6.1 (2010/11/18)
---------------------------------

* Fortran wrappers

  - Fixed typos in cel_f.c, celget[cdi] -> celgt[cdi].


WCSLIB version 4.6 (2010/11/16)
-------------------------------

* C library

  - In wcsulex.l and wcsutrn.l, stdlib.h must be included explicitly
    before the redefinition of exit() - most versions of flex do include
    it upfront but some don't (reported by Peter Williams).

* Fortran wrappers

  - Changes intended to avert nuisance compiler warnings that could
    potentially obscure warnings indicative of a genuine problem:

    - To stop messages about unused variables when the relevant compiler
      option is set, e.g. 'g77 -Wunused', the various *_ERRMSG arrays
      defined in the Fortran include files and (formerly) initialized
      therein via DATA statements, e.g. PRJ_ERRMSG in prj.inc, have now
      been placed into COMMON blocks with names such as PRJ_DATA, and
      are initialized via DATA statements in BLOCK DATA subprograms
      defined in separate files, e.g. prj_data.f.

    - To stop messages about subroutines being invoked with inconsistent
      argument lists if the relevant compiler option is not set, e.g.
      'g77 -Wno-globals', the C wrapper functions that take 'void *'
      arguments now have separate forms for INTEGER, DOUBLE PRECISION,
      and CHARACTER arguments that simply invoke the generic function.
      Application code must be modified to take advantage of this.

* User manual

  - In the section on the Fortran wrappers in the manual, warn about the
    need for the INTEGER array holding a data structure to be aligned on
    a DOUBLE PRECISION boundary.


WCSLIB version 4.5.6 (2010/10/28)
---------------------------------

* Installation

  - Fixed the search for CFITSIO and PGPLOT library and include
    directories.


WCSLIB version 4.5.5 (2010/10/14)
---------------------------------

* Installation

  - Build the PGSBOX sharable library.


WCSLIB version 4.5.4 (2010/09/28)
---------------------------------

* C library

  - In wcshdo(), according to the FITS standard, "Letters in the
    exponential form ('E' or 'D') shall be upper case" (reported by
    Michael Droettboom).


WCSLIB version 4.5.3 (2010/09/23)
---------------------------------

* Utilities

  - Various improvements to wcsgrid: correct the scaling set via
    cpgwnad(); label angles other than RA,Dec in decimal degrees;
    draw the projection boundary for projections other than zenithals.


WCSLIB version 4.5.2 (2010/09/23)
---------------------------------

* C library

  - Fixed the translation of GLS to SFL in wcsset() and celfix() when
    the reference longitude is non-zero - it introduces an offset in
    longitude in the normal way.  (This undoes part of the change
    applied in version 4.4.)


WCSLIB version 4.5.1 (2010/08/12)
---------------------------------

* C library

  - New utility function, sphpad(), computes the coordinates of points
    offset by given angular distances and position angles from a given
    point on the sky (complementary to sphdpa()).

* Fortran wrappers

  - New wrapper function:
    - SPHPAD for sphpad().


WCSLIB version 4.5 (2010/07/16)
-------------------------------

* C library

  - Fixed the interpretation of VELREF when translating AIPS-convention
    spectral types.  Such translation is now handled by a new special-
    purpose function, spcaips().  The wcsprm struct has been augmented
    with an entry for velref which is filled by wcspih() and wcsbth().
    Previously, selection by VELREF of the radio or optical velocity
    convention for type VELO was not properly handled.

* Fortran wrappers

  - New wrapper function:
    - SPCAIPS for spcaips().

  - Changed spc.inc, spc_f.c, wcs.inc and wcs_f.c to track VELREF
    changes.

  - Declared functions external in the include files to avoid compiler
    warnings about unused variables (if the particular option is set).

* Utilities

  - Added a '-q' option to fitshdr to quit after a specified number
    of HDUs.


WCSLIB version 4.4.4 (2009/09/14)
---------------------------------

* Installation

  - Added more configure options for controlling the build:
    --disable-fortran, --disable-utils, --without-cfitsio, and
    --without-pgplot.


WCSLIB version 4.4.3 (2009/09/03)
---------------------------------

* C library

  - Set wave number units to "/m" in spctyp(), was "1/m" which is not
    strictly legal and wasn't handled by wcsulex() (reported by
    Hans Terlow).  Also fixed a number of units specifications in the
    prologue of spx.h to conform with Paper I usage.

  - In wcsulex(), allow unit strings like "1/m" in addition to "/m",
    provided that the superfluous "1" is the first non-blank character
    in the expression, or parenthesised sub-expression.

  - In wcssptr(), ensure that i is always reset if given < 0.

* User manual

  - Augmented the list of FITS WCS and related software in the manual.


WCSLIB version 4.4.2 (2009/08/13)
---------------------------------

* C library

  - In sphx2s() and sphs2x(), handle the case where |eul[1]| = 180.0
    separately for speed and accuracy.  This change also fixes a rare
    and subtle bug in cels2x() that occurs when celprm::isolat is set
    and the magnitude of the first latitude in the lat[] vector exceeds
    90 deg (reported by Hans Terlouw).

* Installation

  - Fix relating to creation of symlinks when installing the libraries.


WCSLIB version 4.4.1 (2009/08/11)
---------------------------------

* Installation

  - Fixes for installation of the CHANGES file and for the creation of a
    symbolic link for the sharable library if one already exists.


WCSLIB version 4.4 (2009/08/10)
-------------------------------

* C library

  - Creation of WCSLIB user manual from the header file prologues using
    a special-purpose parser, doxextr, and sed scripts to generate input
    for doxygen.  This required minor formatting changes to all
    prologues plus miscellaneous changes such as naming of arguments in
    function prototypes.

  - Bug fix in wcsset() that affected handling of PROJPn (deprecated)
    and PVi_ma attached to the longitude (not latitude) axis.  Guard
    against long strings when copying the projection code.  In
    wcs_types(), allow for early Paper IV distortion codes (e.g.
    "RA---TAN-SIP") when parsing CTYPEia.

  - Use sincos() whereever possible for a ~15% speedup (patches for
    cel.c, prj.c and sph.c supplied by Michael Droettboom).  configure
    checks for, and uses it automatically if available.

  - Fixed the translation of GLS to SFL in wcsset() and celfix() when
    the reference longitude and latitude are non-zero.  (In the AIPS
    convention, this simply translates the reference point, i.e. the map
    as a whole, to those coordinates without creating an oblique grid.)

  - Bug fix in prjoff(), a utility function used by the prj routines.
    It forces (x,y) = (0,0) at (phi_0,theta_0) when the latter are set
    by PVi_[012]a attached to the longitude (not latitude) axis.  Rarely
    used in practice.

  - New utility function, sphdpa(), computes the distance and position
    angle from a point on the sphere to a set of field points.

  - In sphx2s() and sphx2s(), handle a simple change in origin of
    longitude using a short-cut calculation for speed and accuracy.
    Similarly in celset(), check whether phip == phi0 when computing
    latp and if so use a short-cut that ensures latp == 90.0 (exactly)
    if lat0 == theta0.  The resulting spherical rotation then becomes a
    simple change in origin of longitude.  In particular, these changes
    should assist PGSBOX in drawing grid lines of +/-180 longitude, to
    prevent flip-flopping between one and the other.

  - wcsbth() & wcspih(): resolved an inconsistency between the
    documentation and code by renamimg WCSHDR_VSOURCEa as
    WCSHDR_VSOURCE.

  - Flex code: moved declaration of helper functions out of global
    scope.

  - Fixed the call to wcss2p() in twcshdr (in a section of code not
    usually exercised).

* Fortran wrappers

  - New wrapper functions:
    - WCSBTH for wcsbth(),
    - WCSBDX for wcsbdx(),
    - CDFIX  for cdfix(),
    - SPHDPA for sphdpa().

  - Updated WCSLEN (in wcs.inc) and added WCS_COLAX and WCS_VELANGL to
    match changes to wcsprm made in v4.3 with corresponding changes to
    the wrapper functions.  Likewise updated TABLEN (in tab.inc) for
    changes to tabprm, and added CEL_LATPREQ for celprm.

  - Struct lengths (WCSLEN, PRJLEN, etc.) are now long enough to
    accomodate 64-bit machines.

  - Updated the flag bits for the RELAX argument in wcshdr.inc to
    reflect changes to wcshdr.h made in v4.3.  Renamed WCSHDR_VSOURCEa
    to WCSHDR_VSOURCE for consistency with the C library.

* PGSBOX

  - Improved grid labelling, particularly in minimizing the number of
    fields required in sexagesimal labels.

* Utilities

  - New utility program:

    - wcsware extracts the WCS keywords for an image from the specified
      FITS file, constructs wcsprm structs for each coordinate
      representation found, and performs a variety of operations using
      them.

  - Old utility programs (first appeared in 4.3 but were not recorded):

    - HPXcvt reorganises HEALPix data into a 2-D FITS image with HPX
      coordinate system.

    - wcsgrid extracts the WCS keywords for an image from the specified
      FITS file and uses pgsbox() to plot a 2-D coordinate graticule for
      each alternate representation found.

    - fitshdr lists headers from a FITS file specified on the command
      line, or else on stdin, printing them as 80-character keyrecords
      without trailing blanks.

* Installation

  - New configure options, --with-pgplotinc, --with-pgplotlib,
    --with-cfitsioinc and --with-cfitsiolib allow additional directories
    to be added to the library and include file search path.

  - Miscellaneous fixes and improvements to the installation process.

  - Generate a metadata file for pkg-config.

  - Added 'make MODE=interactive check' to run the test programs in
    interactive mode rather than batch.

  - Merged the separate CHANGES files for C, Fortran and PGSBOX into
    one (this), with a new section for utilities.


WCSLIB version 4.3.3 (2009/04/30)
---------------------------------

* C library

  - fitshdr.l, wcsbth.l, and wcspih.l: use setjmp/longjmp to preempt
    the call to exit() which is hard-coded in function yy_fatal_error()
    supplied by flex.

  - wcspih.l: if NAXIS is non-zero but there were no WCS keywords at
    all in the header then create a default WCS with blank alternate
    version.


WCSLIB version 4.3.2 (2009/03/16)
---------------------------------

* C library

  - utils/GNUmakefile: create BINDIR if necessary prior to installing
    utilities.


WCSLIB version 4.3.1 (2008/09/08)
---------------------------------

* Installation

  - Top-level GNUmakefile: install header files.


WCSLIB version 4.3 (2007/12/27)
-------------------------------

* C library

  - A new general WCS header parser wcsbth() handles binary table image
    arrays and pixel lists as well as image array headers.  Added
    "colax" to the wcsprm struct to record the column numbers for each
    axis in a pixel list.

  - New function wcsbdx() is the analog of wcsidx() for the array of
    wcsprm structs returned by wcsbth().

  - New function wcshdo() writes out a wcsprm struct as a FITS header.

  - Changes to wcspih():
    - Bug fix, check for a == 0 (indication of a keyword that applies
      to all alternates) in internal helper function wcspih_naxes()
      (reported by Craig Markwardt).
    - Added a new ctrl option to remove valid WCS keyrecords except for
      those with a more general role, namely {DATE,MJD}-{OBS,AVG} and
      OBSGEO-{X,Y,Z} (suggested by Jim Lewis).
    - Added a rule for VELANGLa.  Also added "velangl" to the wcsprm
      struct.
    - Do checks on the i, k & m keyword parameters in <VALUE>.
    - Fixed the test for repeated blanks in the NAXIS and WCSAXES
      patterns.
    - Fixed three <CCi_ma> rules to allow m == 0.
    - Reworked the implementation notes in the prologue.

  - The flex scanners, fitshdr.l, wcsbth.l, wcspih.l, wcsulex.l, and
    wcsutrn.l, invoke yylex_destroy() before returning to avoid a 16kiB
    memory leak.  This was reported by several people, however it may be
    problematic depending on the version of flex used - version 2.5.9 or
    later is required.  If this is not available, C sources pre-
    generated by flex 2.5.33 will be used.

  - In wcs.c, don't define the signbit macro if already defined (for
    MacOSX).

  - In wcs.h, documented wtbarr namespace issues in C++.

  - In wcsset(), always set wcsprm.cunit[i], if possible (primarily for
    use by wcshdo()).

  - In wcsfix.c, parenthesised a boolean expression that was otherwise
    incorrect.

  - Fixed an obscure floating point rounding error in celset() that
    appeared with -O2 optimization in gcc v3.3.5 (Linux).

  - prjset() now correctly propagates the status value returned by the
    specific projection-setting functions (reported by Bill Pence).

  - Bug fix in hpxx2s(), also added bounds checking.  Minor efficiencies
    in carx2s() and merx2s().

  - In the various functions that print the contents of the structs, use
    the "%p" printf conversion specifier to print addresses rather than
    casting the pointer to int and using "#x".  The latter does not work
    on 64-bit machines where sizeof(int) != sizeof(void*).

  - Reorganized the various structs to get alignment on 64-bit machines.

  - All header file prologues now reference the README file for an
    overview of the library.

  - Miscellaneous portability fixes for 64-bit, MacOSX, OSF compiler,
    etc.

  - Elimination of compiler warnings, e.g. parenthesised assignments
    used as truth values (a favourite gcc gripe!), etc.

  - Process flex descriptions using a newer version of flex, primarily
    for MacOSX.  However, the processed files are now only used when
    flex 2.5.9 or later is not available.

  - Removed WCSLIB 2.x backwards-compatibility measures from lin.h,
    prj.h, prj.c, and sph.h.

* Fortran wrappers

  - (No substantive changes.)

* PGSBOX

  - Miscellaneous improvements to PGSBOX.

* General

  - Switched licensing to LGPL 3.0.

  - In comment text, replaced use of the obsolete term "card" with
    "keyrecord" which consists of a "keyword", "keyvalue", and
    "keycomment".

* Installation

  - General improvements to the installation process: autoconf-related
    portability improvements, particularly relating to Fortran name
    mangling; makefile rules for building the shared library, for
    processing flex descriptions; don't rely on "." being in the PATH
    when running tests.


WCSLIB version 4.2 (2005/09/23)
-------------------------------

* C library

  - Brought the installation process under control of GNU autoconf,
    the top-level makefile now builds and tests everything, and the C
    library has a config.h in which WCS_INT64 is set.  Added an INSTALL
    file.

  - Merged the FORTRAN, C and PGSBOX READMEs into one top-level README.

  - Extensions for -TAB coordinate handling: in tabx2s() and tabs2x(),
    allow extrapolation by half a cell at either end of the index and
    coordinate tables; fits_read_wcstab() (in getwcstab.{h,c}) allows
    TDIMn to be omitted for 1-D lookup tables for which it has the form
    '(1,K)', i.e. describing a degenerate 2-D array; wcsprt() now prints
    the wtbarr structs in wcsprm.

  - Bug fixes for -TAB coordinate handling: in tabx2s() and tabs2x()
    the incorrect indexing variable, m instead of i, was used for
    tab->crval[]; wcsp2s() and wcss2p() returned prematurely in the
    tabular coordinate loop; in wcstab(), removed an extraneous
    assignment to wtbp->kind for index arrays.

  - In wcsp2s() and wcss2p(), elements of the stat[] vector that had
    been set were being reset incorrectly to zero.  The stat[] values
    are now set as flag bits for each coordinate element.

  - Added cdfix() to the wcsfix() suite to fix erroneously omitted
    CDi_ja cards.

  - PGSBOX is now compiled into a separate object library, and is
    installed alongside WCSLIB.

  - Eliminated several instances of non-ANSI C library functions and
    header files and some residual K&R C usage.  The Sun C compiler
    complained about const int definitions of initializers used in
    variable declarations in some of the test programs; changed these
    to preprocessor macros.

* Fortran wrappers

  - Fixed handling of 64-bit integer keyvalues in keyget_().

  - Fixed output formatting of 64-bit integer keyvalues in tfitshdr.f.

  - Fixed minor syntax errors in twcsfix.f and tpih1.f reported by the
    Sun Fortran compiler.

  - The output of each test program now identifies the source file.

* PGSBOX

  - (No substantive changes.)


WCSLIB version 4.1 (2005/08/31)
-------------------------------

* C library

  Summary of added functionality:

  - -TAB coordinate axes are now fully implemented in the WCSLIB driver
    functions (in wcs.{h,c}); multiple -TAB axes are supported.  A new
    function, wcstab(), which is automatically invoked by wcspih(),
    parses -TAB-related header cards and sets up structs for a separate
    routine that reads the necessary arrays from a FITS binary table
    extension.

    An implementation of this routine in CFITSIO, fits_read_wcstab(),
    is provided.  Note however that the interface of this function is
    experimental, and the code itself must be considered beta-release in
    WCSLIB 4.1.

  - Units specifications, either from CNAMEia or inline comments (with
    brackets), of arbitrary complexity are now fully implemented via a
    parser, wcsulex(), and converter, wcsunits().  This is invoked
    automatically by wcsset().

  - Translators for non-standard WCS constructs are provided.  These
    cover date formats, units specifications, defunct celestial
    projection types, AIPS spectral axis types, and the repair of
    malformed cylindrical coordinate systems.

  - wcspih() now has options to remove the WCS cards it has processed
    from the header and a new generic FITS header parser, fitshdr(), may
    be used to parse the remaining non-WCS cards.  In addition to the
    more basic types, it handles 64-bit and 'very long' (70 digit)
    integer keyvalues, and also continued string keyvalues.  It also
    does keyword matching and extracts units specifications in inline
    comments.

  - -LOG coordinates are now implemented independently of spectral
    coordinate types.  Multiple -LOG axes are supported.

  - New function wcssptr() translates the spectral axis in a wcsprm
    struct to the required type.

  - The README file now gives an introduction to, and complete overview
    of, WCSLIB.  It provides a point of entry to programming with
    WCSLIB.  Complete descriptions and usage notes for all functions are
    contained in the header files.

  - The FORTRAN wrappers and test programs are now completely up-to-date
    with respect to the C implementation.

  - All code, including the FORTRAN wrappers, PGSBOX, and all test
    programs, now pass 'purify' without memory leaks, uninitialized
    memory reads, memory access violations, or other memory faults.

  Change notes:

  - Added options to wcspih() to remove WCS cards from the input header
    leaving only non-WCS cards behind.  Modified test programs tpih1.c
    and tpih2.c to use CFITSIO optionally via preprocessor macro
    DO_CFITSIO.

  - New function wcstab() in wcshdr.{h,c} parses -TAB-related header
    cards and sets up structs for a separate routine that reads the
    necessary arrays from a FITS binary table extension.  New
    test/demo program twcstab.c using header defined in wcstab.cards.

  - CFITSIO implementation, fits_read_wcstab() in getwcstab.{h,c}, of a
    function, independent of WCSLIB, for extracting arrays from a binary
    table as required in constructing -TAB coordinates.

  - New units specification parser, wcsulex() in wcsunits.h and
    wcsulex.l, and converter, wcsunits() in wcsunits.{h,c}.  New
    test/demo program tunits.c.

  - New parser for non-standard units specifications, wcsutrn() in
    wcsunits.h and wcsutrn.l, also tested by tunits.c.

  - New functions datfix(), unitfix() (which applies wcsutrn()),
    celfix(), and spcfix() join cylfix() in wcsfix.{h,c} to translate
    various forms of non-standard or quasi-standard FITS WCS keyvalues
    in a wcsprm struct.  wcsfix() applies all of these in sequence.
    New test/demo program twcsfix.c, with wcsfix() also now invoked by
    tpih1.c.

  - New generic FITS header parser, fitshdr() in fitshdr.{h,l}.  New
    test/demo program tfitshdr.c uses wcs.cards with extra non-WCS
    cards added.

  - -LOG coordinates are now treated as a coordinate type separate from
    spectral coordinates, implemented via log.{h,c} and test program
    tlog.c.  The logarithmic functions were removed from spx.{h,c}, and
    spc.c.

  - Extensive changes to wcs.{h,c} to support multiple -TAB and -LOG
    coordinate axes and units conversion.  Substantially changed the
    test program, twcs.c, to test the more general functionality.

  - New function wcssptr() in wcs.{h,c} translates the spectral axis in
    a wcsprm struct.

  - Added DATE-AVG to wcsprm.  Also ntab, tab, nwtb, and wtb required
    for -TAB implementation.  Define struct wtbarr.

  - Added a types[] member to the wcsprm struct to identify axis
    coordinate types using a four-digit code.

  - Use memset() in wcsini() to null-fill character arrays in the wcsprm
    struct so that they don't appear to be padded with garbage when
    displayed by gdb.

  - Do alias translation for AIPS-convention spectral types in wcsset()
    using spctyp().  If wcsset() finds a CTYPEia in "4-3" form with an
    unrecognized algorithm code it now returns an error rather than
    assume that it's a linear axis.  wcsset() now also resets lonpole
    and latpole to the values actually used.

  - Modified spctyp() to translate AIPS-convention spectral ctypes, and
    modified the argument list to return the parsed spectral type and
    algorithm code.  The return arguments will not be modified if
    CTYPEia is not a valid spectral type; zero-pointers may be specified
    for any that are not of interest.  Removed the external const
    variables, spc_codes and spc_ncode, as their function is now
    fulfilled by spctyp().

  - Fixed a bug in spctrn() in resolving ctypeS2 wildcarding.

  - Added latpreq member to the celprm struct, set by celset() to
    indicate how LATPOLE is used.  Augmented tcel2.c to report it.

  - New function tabmem() in tab.{h,c} takes control of user-allocated
    memory.

  - tabini() allows K == 0 and also K[m] == 0 to initialize partially
    the tabprm struct (for wcstab()).  It now does fine-grained
    bookkeeping of memory allocation for the index arrays and allocates
    each individually.  tabprm.index[] == 0x0 is recognized as default
    indexing in tabset(), tabx2s() and tabs2x().

  - The *prt() functions report parameters to an extra decimal place.

  - tabprt() prints the array index for elements of the coordinate and
    index vectors.

  - Set the 0th element in all *_errmsg arrays to "Success".

  - Extracted string utility functions used by WCSLIB into
    wcsutil.{h,c}.

  - Removed support for K&R C.

* Fortran wrappers

  - The FORTRAN wrappers and test programs are now completely up-to-date
    with respect to the C implementation.

  - New include files, wrappers, and test programs:
    fitshdr.inc, fitshdr_f.c, getwcstab.inc, getwcstab_f.c, log.inc,
    log_f.c, sph.inc, tab.inc, tab_f.c, tfitshdr.f, tlog.f, ttab1.f,
    ttab2.f, ttab3.f, tunits.f, twcsfix.f, twcstab.f, wcsfix.inc,
    wcsfix_f.c, wcsunits.inc, wcsunits_f.c.

  - Updates to reflect changes to the C library and test programs:
    cel.inc, cel_f.c, prj.inc, spc.inc, spc_f.c, spx.inc, spx_f.c,
    tlin.f, tpih1.f, tpih2.f, tprj1.f, tprj2.f, tspc.f, tsph.f, tspx.f,
    twcs.f, twcsmix.f, twcssub.f, wcs.inc, wcs_f.c, wcshdr.inc,
    wcshdr_f.c.

  - Added *_ERRMSG arrays containing status messages to all include
    files.

  - Removed support for K&R C.

* PGSBOX

  - Fixed a subtle though benign memory fault identified by 'purify'.

  - Reset LATPOLE in the COE example in cpgtest.f when drawing the
    second (native) grid because it will have been set to a non-default
    value by wcsset() when the first grid was drawn; set wcs.flag to -1
    before wcsinit() and call wcsfree() at the end.  Similarly for
    pgtest.f.


WCSLIB version 4.0 (2005/02/07)
-------------------------------

* C library

  - Implemented tabular coordinates (-TAB).  New files: tab.h and tab.c,
    and test programs ttab[123].c.  These have not been incorporated
    into the higher-level (wcs.h) functions at this stage.

  - New spectral functions: spchek() checks a spectral algorithm code
    for legitimacy; from the spectral keywords given, spcspx() derives
    the corresponding CRVALi and CDELTi keywords for the underlying P-,
    and X-type spectral coordinates; spcxps() does the opposite;
    spctrn() combines spcspx() and spcxps() to translate one set of
    spectral keywords into another, e.g.  'FREQ' -> 'ZOPT-F2W'.

  - Implemented the HEALPix (HPX) projection in the prj functions.

  - Added a new function, wcsidx(), to return an array that indexes the
    alternate coordinate descriptions found by wcspih() (suggested by
    Bill Pence, NASA/Goddard).  Modified tpih1.c to exercise it.

  - In wcsp2s() and wcss2p(), check that nelem equals or exceeds
    wcs.naxis; emphasised this in the usage notes for these functions
    in tab.h (suggested by Bill Pence, NASA/Goddard).

  - Moved the macros used for UNDEFINED values and the corresponding
    macro test function, undefined(), to wcsmath.h for general use.
    Previously, UNDEFINED values were only used internally, but they are
    now visible in some of the structs, particularly values of undefined
    auxiliary header cards in the wcsprm struct.

  - Remove const from the double args in the specx() prototype in spx.h
    to match the definition in spx.c (reported by Bryan Irby,
    NASA/Goddard).

  - Fixed the interaction between the FLAVOUR and PGPLOTLIB definitions
    in the C and FORTRAN Makefiles by introducing a separate variable,
    DO_PLOTS, to control whether to exercise test programs that require
    PGPLOT (reported by Bill Pence, NASA/Goddard).

* Fortran wrappers

  - New wrapper defined in wcshdr_f.c: wcsidx_().  Modified test program
    tpih1.f to use it.

* PGSBOX

  - (No substantive changes.)

* General

  - Changed the copyright notice in all library routines from LGPL to
    GPL as recommended by the FSF (http://www.gnu.org/licenses/why-not-
    lgpl.html).

* Installation

  - General improvements to the installation process: fixed the
    interaction between the FLAVOUR and PGPLOTLIB definitions in the
    Makefile by introducing a separate variable, DO_PLOTS, to control
    whether to exercise test programs that require PGPLOT (reported by
    Bill Pence, NASA/Goddard).  Added an "install" target to the
    Makefile.


WCSLIB version 3.6 (2004/08/25)
-------------------------------

* C library

  - New service routine, wcssub() extracts the coordinate description
    for a subimage from a wcsprm struct.  wcscopy() is now implemented
    as a preprocessor macro that invokes wcssub().  New test program,
    twcssub.c, tests wcssub().

  - In wcspih():

    1) Fixed handling of string-valued keywords where the inline comment
       contains a single-quote character (').

    2) Fixed the address arithmetic for EPOCH and VELREF.

    3) Translate VSOURCEa into ZSOURCEa if required.

    4) Recognize SSYSSRCa.

    5) Support free-format string keyvalues, as well as integer and
       floating-point keyvalues; documented this in the prologue of
       wcshdr.h.

    6) Allow header cards without inline comments.

    7) Fixed function prototyping in wcspih.l (i.e. ANSI and non-ANSI
       forms were potentially mixed).

    8) Catch an unhandled newline character on the END card that was
       echoed to stdout.

    9) In error messages, print "ERROR" (uppercase) - POSIX standard.

  - Modified wcs.cards to explain and test free-format keyvalues, and
    also augmented the inline comment on illegal WCS cards that are to
    be rejected, and WCS-like cards to be discarded.  Added a header
    card with no inline comment.

  - Removed vsource from the wcsprm struct and added ssyssrc.

  - In wcsini(), fixed a problem with memory management for wcs.pv when
    NPVMAX is zero; likewise for wcs.ps and NPSMAX.

  - In wcsprt(), don't print auxiliary coordinate system information in
    arrays with zero address.

  - In wcss2p(), status == 9 (one or more invalid world coordinates) was
    not returned appropriately.

  - Renamed twcs1.c to twcs.c, and twcs2.c to twcsmix.c.

  - "Error status/code/number" is now referred to consistently as the
    "status return value".

  - Some vestiges of K&R C were removed: preprocessor definition of
    const, and K&R function prototypes.

* Fortran wrappers

  - New wrapper defined in wcs_f.c: wcssub_().  New test program,
    twcssub.f.

  - Renamed twcs1.f to twcs.f, and twcs2.f to twcsmix.f.

* PGSBOX

  - (No substantive changes.)

* Installation

  - Worked over the C, FORTRAN, and PGSBOX makefiles, in particular to
    make them all consistent.


WCSLIB version 3.5 (2004/06/28)
-------------------------------

* C library

  - WCSLIB now provides a function, wcspih() implemented as a Flex
    description, that parses a FITS image header, either that of a
    primary HDU or an image extension.  Given a character array
    containing the header it identifies and reads all WCS cards for
    the primary coordinate description and up to 26 alternate
    descriptions and returns this information as an array of wcsprm
    structs.  A service routine, wcsvfree(), is provided to free the
    memory allocated by wcspih().  The relevant header file for these
    functions is wcshdr.h.

    Test programs, tpih1 and tpih2, are provided to verify wcspih.  The
    first simply prints the contents of the structs using wcsprt().  The
    second uses cpgsbox() to draw coordinate graticules.  A FITS WCS
    test header has been developed to provide input to these test
    programs.  It is implemented as a list of card images, wcs.cards,
    one card per line, together with a program, tofits, that compiles
    these into a valid FITS file.  tpih1 uses its own code to read this,
    whereas tpih2 uses the fits_hdr2str() function from CFITSIO.

  - Removed twcsprt, tpih exercises wcsprt() much more thoroughly than
    twcsprt ever did.  Modified twcs1 to print the size of the various
    structs as twcsprt used to.

  - Although they are not used in any coordinate calculations, the
    wcsprm struct now provides data members for storing all of the
    auxiliary FITS WCS header cards defined in Papers I, II, and III,
    such as WCSNAMEa, EQUINOXa, and CNAMEia.  Members are also provided
    for storing the alternate descriptor code (the "a" in CTYPEia), and
    the binary table column number.  These are supported by the high
    level WCSLIB routines, wcsini(), wcscopy(), wcsfree(), and wcsprt().
    Refer to wcs.h for details.

  - The number of PVi_ma cards for which wcsini() allocates memory is
    now set by a global variable, NPVMAX (previously a C-preprocessor
    macro).  This defaults to 64 but may be changed by a new function,
    wcsnpv().  The wcsprm struct contains a new member, npvmax, that
    records the value of this number at the time the struct was
    initialized.  This is in addition to npv which records the actual
    number of cards that were encountered.

    Similarly, NPSMAX (default 8) is used for the number of PSi_ma
    cards, and it may be changed via wcsnps().

    The axis number, i, in the pvcard struct used for storing PVi_ma
    cards may now be set to 0 to indicate the latitude axis.

  - calloc() is now used in place of malloc() in allocating memory for
    arrays, and inclusion of malloc.h has been replaced with stdlib.h
    for all platforms.

    wcsfree() checks that wcs.flag != -1 before testing wcs.m_flag when
    freeing memory allocated by wcsini() in case the struct is
    uninitialized.  Similarly for linfree().

  - In prj.h, renamed C-preprocessor macros INI, PRT, SET, X2S and S2X
    to PRJINI, PRJPRT, PRJSET, PRJX2S and PRJS2X to reduce the
    likelihood of namespace clashes.  Similarly in spc.h.

    Also, in prj.c, changed the name of helper routine offset() to
    prjoff() to reduce the likelihood of global namespace conflicts.

  - In line with bonx2s() and bons2x(), bonset() now recognizes the
    equatorial case of Bonne's projection as Sanson-Flamsteed, mainly so
    that the auxiliary information in the prjprm struct more accurately
    reflects the truth.  Modified tcel2 to exercise this by using an
    equatorial Bonne projection in place of the Hammer-Aitov.

  - zpns2x() used prj.w[0] for bounds checking, though this had not been
    set by zpnset() for polynomials of degree N < 3.  Consequently,
    bounds checking for N < 3 was unreliable (reported by David Berry,
    STARLINK).

  - Changed some variable names in tscs2x(), cscx2s(), cscs2x(),
    qscx2s(), and qscs2x() to match Paper II, and likewise changed some
    inequality tests in qscs2x() without changing the results.

  - Minor tidying up of output formatting in prjprt().

  - Added the alternate version code to FITS WCS keywords mentioned in
    comments, e.g. CTYPEi changed to CTYPEia.

* Fortran wrappers

  - New wrappers defined in wcshdr_f.c: wcspih_() and wcsvfree_(), and
    also a new service function, wcsvcopy_().

    New test programs, TPIH1 and TPIH2, being analogues of tpih1 and
    tpih2.  Removed TWCSPRT.

  - In wcs_f.c, new wrappers wcsnpv_() and wcsnps_(); modified wcsput_()
    and wcsget_() to handle new members of the wcsprm struct.  Also
    modified wcsput_() to null-fill all char[] members of the wcsprm
    struct, and likewise wcsget_() to blank-fill them.

  - Modified wcs.inc to support changes to the wcsprm struct.

* PGSBOX

  - In PGSBOX, increased the dimension of the WORLD and XY vectors from
    2 to 9 to accomodate higher-dimensional coordinate representations
    of up to 9 elements.  Similarly for pgwcsl().  The assumption
    (presently) is that the first two world, and pixel, coordinate
    elements are the relevant ones; the others are all set to zero when
    pgwcsl() initializes and otherwise ignored.

    Assigned some variables in DATA to stop compiler messages about
    uninitialized variables.

  - Generalized the Makefile, bringing it into line with the WCSLIB
    Makefile, and adding separate targets for compiling and running the
    test programs.  The default target now simply compiles pgsbox.c and
    cpgsbox.c.  A separate target compiles pgwcsl.c and inserts it into
    ../C/libwcs.a.


WCSLIB version 3.4 (2004/02/11)
-------------------------------

* C library

  - In aitx2s(), apply the boundary condition 0.5 <= Z^2 <= 1 given
    after Eq. (109) in WCS Paper II to detect outlying pixels.

  - Fixed several special-case bugs in celset():

    1) For theta_0 = 90, in substituting the default value for phi_p
       (LONPOLE),

       a) for the special case when delta_0 = 90, celset() provided the
          wrong value (180 instead of 0),

       b) celset() neglected to add phi_0 (normally 0).

    2) For theta_0 != 90,

       a) for the special case when delta_0 = -90, celset() incorrectly
          computed delta_p (as theta_0 instead of -theta_0),

       b) for the special case when delta_p = +90 (or -90), celset()
          neglected to subtract (or add) phi_0 (normally 0).

    3) For |delta_0| = 90, celset() incorrectly allowed the particular,
       invalid, value of phi_p (LONPOLE) that put the other pole at the
       fiducial point.

    4) For theta_0 = 0, delta_0 = 0 LATPOLE determines delta_p
       completely.  For LATPOLE > 90 celset() now sets delta_p to 90,
       and for LATPOLE < -90 it sets it to -90.

  - Additional refinements in celset():

    1) cel->ref[2] is normalized in the range [-180,180].

    2) Account for rounding error in the computation of delta_p.

  - sphx2s() and sphs2x() incorrectly handled the "change in the origin
    of longitude" special case that arises when delta_p = -90, in the
    even more restrictive case where |theta| = 90 also; it applied
    Eq. (3) instead of Eq. (4) of Paper II.

  - Added a new test program, tcel2.c, to exercise celset() more
    thoroughly.  Renamed the original tcel.c to tcel1.c and modified the
    Makefile to suit.

* Fortran wrappers

  - (No changes.)

* PGSBOX

  - (No substantive changes.)


WCSLIB version 3.3 (2003/10/21)
-------------------------------

* C library

  - In celset(), the default value for phi_p (LONPOLE) is

       phi_p = phi_0 + ((delta_0 < theta_0) ?  180.0 : 0.0)

    Previously phi_0 (which is normally zero) was not added (reported by
    David Berry, STARLINK).

  - wcsprt() and linprt() now check that the structs have been
    initialized.

  - In wcsini(), when the wcsprm flag is -1 also set the linprm flag to
    -1 to force initialization of the memory managed by linset().

  - wcsset() now explicitly initializes the celprm and spcprm structs
    via celini() and spcini().

  - Fixed syntax errors in the macro definitions of linrev_errmsg and
    linfwd_errmsg.

  - In Makefile, added the -ansi option to gcc to force it to behave
    like a strict ANSI C compiler, specifically in setting the __STDC__
    preprocessor macro.

* Fortran wrappers

  - (No changes.)

* PGSBOX

  - PGSBOX now recognizes status returns -1, -2, and -3 from NLFUNC for
    opcodes +2 and +1 which cause it to accept the returned (x,y)
    coordinates but not consider them as one end of a crossing segment
    for labelling world coordinate 1, 2, or both.

  - PGSBOX now takes care not to lose vertical tick marks (and hence
    labels) at the left or right edge of the frame.  Likewise for
    horizontal tick marks at the top or bottom edge of the frame.

  - Tightened up the test in PGSBOX for cycles in angle to catch the
    case where the coordinate value spans a full 360 degrees.

  - PGSBOX will no longer accept frame crossings that are too oblique;
    floating point rounding errors may cause grid lines that should
    otherwise track along the frame to weave down it from side-to-side
    resulting in spurious crossing points.

  - Fixed a bug in pgwcsl_() for processing simple linear coordinates.

  - pgwcsl_() now returns error -2 if the latitude is outside -90 to +90
    for opcodes +2 and +1.

  - Amended the translation of status return codes from WCSLIB in
    pgwcsl_().

  - Provided a header file for pgwcsl_() (mainly for C++ usage).

  - Added extra test plots to PGTEST and cpgtest.

  - Added extra functionality to the Makefile.


WCSLIB version 3.2 (2003/09/09)
-------------------------------

* C library

  - Added the facility of setting the flag member of a wcsprm struct to
    -1 before calling wcsini() for the first time in order to initialize
    memory management.  Likewise for linprm and linini().

  - Renamed wcscpy() to wcscopy() to avoid a conflict with the Posix
    "wide character string" function of the same name (wchar.h).  In
    particular, this is used by the GNU C++ string class.

  - The higher level functions (wcs, cel, spc) no longer return
    prematurely if some of the input coordinate values are invalid.

  - All functions now test whether a null pointer for the particular
    struct (wcsprm, celprm, etc.) has been passed to them.

  - Function return codes have been rationalized into a consistent set
    within each of the wcs, cel, lin, prj, spc, and spx suites of
    functions.  Error messages to match these error codes are now
    encoded in a single character array, e.g. wcs_errmsg and prj_errmsg,
    instead of a separate array for each function.  Macro definitions
    for the older character arrays (e.g. wcsini_errmsg) have been
    provided for backward compatibility.

  - Declared prj_stat as extern in prj.h.

* Fortran wrappers

  - (No changes.)

* PGSBOX

  - Added an ENTRY point, PGLBOX, that provides a simplified interface
    to PGSBOX for linear axes without having to specify an NLFUNC or the
    associated parameters.


WCSLIB version 3.1 (2003/04/29)
-------------------------------

* C library

  - Added "global" and "divergent" prjprm struct informational members
    to record whether the projection is capable of mapping the whole
    sphere, and whether it is divergent in latitude.

  - Function cylfix() provided to fix WCS FITS header cards for
    malformed cylindrical projections (c.f. Paper II, Sect. 7.3.4).

  - Added support for CUNITi cards to wcsprm (but not currently
    implemented).

  - Added macro implementations of the trigd functions to wcstrig.h,
    enabled if WCSTRIG_MACRO is defined.

  - Improved printing of the WCSLIB structs.

  - Added macro definitions for the lengths of the WCSLIB structs
    measured in sizeof(int) units (mainly for the FORTRAN wrappers).

* Fortran wrappers

  - FORTRAN is now supported via a set of wrappers on the C library.
    Refer to the README file.

* PGSBOX



WCSLIB version 3.0 beta release (2003/04/01)
--------------------------------------------

* C library

  - Fully vectorized function interfaces (C preprocessor macros are
    available to implement the scalar interfaces of the proj.c, sph.c,
    and lin.c routines from WCSLIB 2.x).

  - Implementation of Paper II, Sect. 2.5: User-specified
    (phi0, theta0).

  - Implementation of Paper III (excluding "-TAB").

  - Memory management is now implemented in the upper-level (wcs.c)
    routines.

  - New extensible design should accomodate Paper IV (and any other)
    without further change to the function interfaces.

* PGSBOX

  - Added a C wrapper function, cpgsbox(), and C test/demo program,
    cpgtest, that duplicates PGTEST and serves as a C coding template.

  - Added calendar date axes.

  - Sped up the max/min search - if only tickmarks are required there is
    no need to search the interior of the image.

  - Return margin widths in CACHE(,NC).

  - Fixed a buglet that caused ticks at the frame edges to be skipped.

  - Return error 3 if CACHE overflows.

  - Adapted PGWCSL for WCSLIB 3.x - it is now a C function (for
    interfacing to WCSLIB) with a FORTRAN-like interface (for PGSBOX).


WCSLIB version 2.9 (2002/04/03)
-------------------------------

* C library

  - Fixed a bug with alias translation in wcsset().

  - Added a conditional compilation directive to lin.c for Apple's
    MacOS X.

* Fortran library

  - Fixed CUBEFACE handling in WCSSET.


WCSLIB version 2.8 (2001/11/16)
-------------------------------

* C library

  - Added support for the SZP projection with szpset(), szpfwd() and
    szprev(), and generalized AZP with support for the tilt parameter,
    gamma.

  - Added phi0 to the prjprm struct, this is set by the projection
    initialization routines along with theta0.

  - Fixed a problem in wcsmix() caused by numerical imprecision that
    could cause it not to recognize when convergence was reached; break
    out of the loop early if a discontinuity is detected.

  - Clarified the usage of vspan in the prologue to wcsmix().

  - Fixed comments relating to LATPOLE in the prologue to cel.c and
    tcel.c, and replaced references to LONGPOLE with LONPOLE.

  - Augmented the error reports in twcs2.

  - Modified projex() in tproj1 and prjplt() in tproj2 to make use of
    the information stored in the prjprm struct.

* Fortran library

  - Added support for the SZP projection with SZPSET, SZPFWD and SZPREV,
    and generalized AZP with support for the tilt parameter, gamma.

  - Changed the call sequence to PRJSET to return PHI0 along with
    THETA0.

  - Fixed a problem in WCSMIX caused by numerical imprecision that could
    cause it not to recognize when convergence was reached; break out of
    the loop early if a discontinuity is detected.

  - Clarified the usage of VSPAN in the prologue to WCSMIX.

  - Fixed comments relating to LATPOLE in the prologue to CEL and TCEL,
    and replaced references to LONGPOLE with LONPOLE.

  - Augmented the error reports in TWCS2.

  - Modified PROJEX in TPROJ1 and PRJPLT in TPROJ2 to use the generic
    driver routines PRJSET, PRJFWD and PRJREV.  PRJPLT also now uses the
    projection type encoded in PRJ(11).


WCSLIB version 2.7 (2001/02/19)
-------------------------------

* C library

  - Added generic driver routines prjset(), prjfwd() and prjrev().
    These invoke specific projection routines via the pointer-to-
    function elements, prjfwd and prjrev, transferred to the prjprm
    struct from celprm.

  - Added code (3-letter projection code) and theta0 (reference
    latitude) elements to prjprm.

  - The projection code for the Sanson-Flamsteed projection is now SFL.
    The upper-level routines, wcsset(), wcsfwd(), and wcsrev(),
    recognize GLS as an alias for this.

  - wcsset() now recognizes 'xyLN/xyLT' axis pairs.

  - Two bugs in the translation from NCP to SIN in wcsfwd() and wcsrev()
    were fixed: (1) the projection parameter was computed incorrectly
    and (2) they did not honour prj->flag set to -1 to disable strict
    bounds checking.

  - A bug in wcsmix() was fixed - it was relying on the wcsprm struct to
    have been initialized beforehand.

  - The test programs now use the cpgplot interface to PGPLOT, the old
    tpgc.c and tpgf.f wrappers have been removed.

* Fortran library

  - Added generic driver routines PRJSET, PRJFWD and PRJREV.  These are
    keyed to specific projection routines via the value of PRJ(11) which
    now differs for each projection.

  - The projection code for the Sanson-Flamsteed projection is now SFL.
    The upper-level routines, WCSSET, WCSFWD, and WCSREV, recognize GLS
    as an alias for this.

  - WCSSET now recognizes 'xyLN/xyLT' axis pairs.

  - A bug in the translation from NCP to SIN in WCSFWD and WCSREV was
    fixed; they did not honour PRJ(11) set to -1 to disable strict
    bounds checking.

  - A bug in WCSMIX was fixed - it was relying on the WCS array to have
    been initialized beforehand.


WCSLIB version 2.6 (2000/05/10)
-------------------------------

* C library

  - Check for invalid (x,y) in zearev().

  - In wcsmath.h, guard against prior definition of PI and other
    preprocessor variables.

* Fortran library

  - Check for invalid (X,Y) in ZEAREV.

  - Declare COSD and SIND in WCSFWD and WCSREV, reported by Clive Page
    (cgp@star.le.ac.uk).


WCSLIB version 2.5 (1999/12/14)
-------------------------------

* C library

  - Added copyright notice to header files and prefixed include guard
    names with "WCSLIB_".

  - Fixed cube face handling in wcsfwd() and wcsrev() (reported by
    Doug Mink, CfA).  Allow more general face layout in the inverse
    quadcube projections.

  - Fixed the problem in wcsmix() where it failed to find a valid
    solution when the solution point lay at a native pole of a
    projection in which the pole is represented as a finite interval.
    However, wcsmix() will only ever return one solution even when two
    or more valid solutions may exist.

  - wcsmix() now accepts viter in the range 5 - 10, the specified value
    will be pushed up or down into this range if necessary.

  - The projection routines for AZP, TAN, SIN, ZPN, and COP now return
    error 2 if (phi,theta) correspond to the overlapped (far) side of
    the projection.  This strict bounds checking can be switched off by
    setting prj->flag to -1 (rather than 0) when the projections are
    initialized.

  - The upper level routines, wcsset(), wcsfwd(), wcsrev(), and
    wcsmix(), now recognize the NCP projection and convert it to the
    equivalent SIN projection.  The lower level routines do not
    recognize NCP.

  - Extracted definitions of mathematical constants (PI etc.) from
    proj.h into wcsmath.h in order to avoid conflicts with their
    definition in math.h in some systems (such as Linux).

  - Describe the two alternate representations of the quadcube
    projections (i.e. faces laid out or stacked) in the prologue of
    wcs.c.

* Fortran library

  - Fixed cube face handling in WCSFWD and WCSREV, reported by Doug Mink
    (dmink@cfa.harvard.edu).  Allow more general face layout in the
    inverse quadcube projections.

  - Fixed the problem in WCSMIX where it failed to find a valid solution
    when the solution point lay at a native pole of a projection in
    which the pole is represented as a finite interval.  However, WCSMIX
    will only ever return one solution even when two or more valid
    solutions may exist.

  - WCSMIX now accepts VITER in the range 5 - 10, the specified value
    will be pushed up or down into this range if necessary.

  - The projection routines for AZP, TAN, SIN, ZPN, and COP now return
    error 2 if (phi,theta) correspond to the overlapped (far) side of
    the projection.  This strict bounds checking can be switched off by
    setting PRJ(11) to -1 (rather than 0) when the projections are
    initialized.

  - The upper level routines, WCSSET, WCSFWD, WCSREV, and WCSMIX, now
    recognize the NCP projection and convert it to the equivalent SIN
    projection.  The lower level routines do not recognize NCP.

  - Describe the two alternate representations of the quadcube
    projections (i.e. faces laid out or stacked) in the prologue of
    wcs.f.


WCSLIB version 2.4 (1996/09/23)
-------------------------------

* C library

  - In sinrev(), cscrev(), qscrev(), and tscrev(), return error 2 if
    (x,y) do not lie within the perimeter of the projection.  In
    sinrev(), stop the computation of phi for the "synthesis" projection
    being applied to the pure "orthographic" case (reported by
    David Berry, STARLINK).

  - (Internal change) Renamed variables l <-> m in the quadcube
    projections to accord with standard usage (and revised WCS draft
    paper).

* Fortran library

  - In SINREV, CSCREV, QSCREV, and TSCREV, return error 2 if (X,Y) do
    not lie within the perimeter of the projection.  In SINREV, stop the
    computation of PHI for the "synthesis" projection being applied to
    the pure "orthographic" case.  Reported by David Berry
    (dsb@ast.man.ac.uk).

  - (Internal change) Renamed variables L <-> M in the quadcube
    projections to accord with standard usage (and revised WCS draft
    paper).

  - (Internal change) Stopped PRJ(11) doing double service in any
    projection.  It is now set and tested for a specific magic value
    rather than simply being non-zero.


WCSLIB version 2.3 (1996/06/24)
-------------------------------

* C library

  - Fixed two bugs in zpnset().  The first led to an incorrect
    determination of the degree of the polynomial and would mainly have
    affected the efficiency of zpnrev().  The second affected the
    determination of the boundary of the projection but would only have
    been significant for projections with a point of inflection between
    9 and 10 degrees of the pole.  Reported by David Berry, STARLINK.

  - Replaced usage of alloca() in lin.c with malloc() and free() for
    portability as suggested by Klaus Banse, ESO (kbanse@eso.org).

  - Allow for C implementations that provide their own versions of
    cosd(), sind(), tand(), acosd(), asind(), atand(), and atan2d().
    From Klaus Banse, ESO (kbanse@eso.org).

  - Implemented the CUBEFACE axis for quadcube projections.

  - Made all function prototypes const-correct.

  - Adapted the header files to C++ usage.

  - Added a new test program, twcs1, to verify closure of wcsfwd() and
    wcsrev().  The old twcs test program is now called twcs2.

  - Added external arrays of error messages indexed by function return
    value.  For example, extern const char *wcsmix_errmsg[] for
    wcsmix().  Messages for the many proj.c functions are in
    prjfwd_errmsg[], etc.

* Fortran library

  - Implemented the CUBEFACE axis for quadcube projections.

  - Added a new test program, TWCS1, to verify closure of WCSFWD and
    WCSREV.  The old TWCS test program is now called TWCS2.


WCSLIB version 2.2 (1996/01/18)
-------------------------------

* C library

  - Amended the projection equations for the conics (COP, COD, COE, COO)
    and Bonne's projection (BON) to correctly handle southern hemisphere
    projections with PROJP1 < 0 (reported by Lindsay Davis, NOAO).
    Revised tproj1 and tproj2 to test such cases.

* Fortran library

  - Amended the projection equations for the conics (COP, COD, COE, COO)
    and Bonne's projection (BON) to correctly handle southern hemisphere
    projections with PROJP1 < 0 (reported by Lindsay Davis, NOAO).
    Revised TPROJ1 and TPROJ2 to test such cases.

  - Increased the dimension of the WCS array from WCS(0:2) to WCS(0:3)
    to allow for future handling of the CUBEFACE keyword - WCS(3) will
    store an index to the CUBEFACE axis.  This affects the call
    sequences of WCSSET, WCSFWD, WCSREV, and WCSMIX.


WCSLIB version 2.1 (1995/11/17)
-------------------------------

* C library

  The main change of interest to programmers is that of changed argument
  lists for wcsfwd() and wcsrev() as described below.

  - The WCS linear transformations are now implemented in WCSLIB,
    complete with matrix inverter.  The new files are lin.c, lin.h, and
    test program tlin.c.

  - Given either the celestial longitude or latitude plus an element of
    the pixel coordinate a new routine, wcsmix(), solves for the
    remaining elements by iterating on the unknown celestial coordinate
    element using wcsfwd().

  - The high level driver routines wcsfwd(), wcsrev(), and wcsmix() now
    apply the full WCS algorithm chain (except for pixel regularization
    table), including parsing the CTYPEn header cards and computing non-
    celestial elements of the world coordinate.  This required a change
    to their argument lists which now more closely reflect the sequence
    of algorithms applied.  A new routine, wcsset(), parses the CTYPEn.

  - The high level driver routines of WCSLIB 1.0 are available as
    intermediate level drivers celset(), celfwd(), and celrev(), but
    note that their argument lists have been extended to return native
    coordinates.  The related struct is now called celprm instead of
    wcsprm.

  - The reference point for conic projections is now at the midpoint of
    the standard parallels.  The FITS header cards PROJP1 and PROJP2 now
    give the half-sum (midpoint) and half-difference of the latitudes of
    the standard parallels; previously they gave the latitudes of the
    standard parallels themselves.  The change is reflected in this
    release of WCSLIB.

  - A bug in celset() (formerly wcsset()) that misapplied WCS draft
    equations 7 has been fixed (thanks to Rick Ebert IPAC/JPL and
    Lindsey Davis, NOAO for reporting this).  This affected the
    computation of Euler angles for the celestial coordinate
    transformation for those projections that have their reference point
    away from the native pole.  In investigating this a deficiency with
    the formalism was discovered that led to the introduction of a
    LATPOLE FITS header card which may be used to disambiguate where
    CRVAL1, CRVAL2, and LONGPOLE do not uniquely determine the latitude
    of the native pole.  The celprm struct (formerly wcsprm) has been
    extended to accomodate LATPOLE.

  - Default values of LONGPOLE and LATPOLE are now supported and their
    use is recommended where appropriate.

  - Numerical precision was being lost near the native poles in the SIN,
    AIR, and QSC projections and this has been recovered (reported by
    Lindsey Davis, NOAO).  Floating underflows in CSC are now avoided.

  - Numerical precision was also lost in certain circumstances in the
    spherical coordinate transformation routines and this has been
    fixed.

  - The test programs have been enhanced in various ways and the library
    has been validated on an SGI machine using both 32-bit and 64-bit
    compilers.

* Fortran library

  The main change of interest to programmers is that of changed call
  sequences for WCSFWD and WCSREV as described below.

  - The WCS linear transformations are now implemented in WCSLIB,
    complete with matrix inverter.  The new files are lin.f and test
    program tlin.f.

  - Given either the celestial longitude or latitude plus an element of
    the pixel coordinate a new routine, WCSMIX, solves for the remaining
    elements by iterating on the unknown celestial coordinate element
    using WCSFWD.

  - The high level driver routines WCSFWD, WCSREV, and WCSMIX now apply
    the full WCS algorithm chain (except for pixel regularization
    table), including parsing the CTYPEn header cards and computing non-
    celestial elements of the world coordinate.  This required a change
    to their call sequences which now more closely reflect the sequence
    of algorithms applied.  A new routine, WCSSET, parses the CTYPEn.

  - The high level driver routines of WCSLIB 1.0 are available as
    intermediate level drivers CELSET, CELFWD, and CELREV, but note
    that their call sequences have been extended to return native
    coordinates.  The related parameter array is now called CEL instead
    of WCS.

  - The reference point for conic projections is now at the midpoint of
    the standard parallels.  The FITS header cards PROJP1 and PROJP2 now
    give the half-sum (midpoint) and half-difference of the latitudes of
    the standard parallels; previously they gave the latitudes of the
    standard parallels themselves.  The change is reflected in this
    release of WCSLIB.

  - A bug in CELSET (formerly WCSSET) that misapplied WCS draft
    equations 7 has been fixed (thanks to Rick Ebert IPAC/JPL and
    Lindsey Davis, NOAO for reporting this).  This affected the
    computation of Euler angles for the celestial coordinate
    transformation for those projections that have their reference point
    away from the native pole.  In investigating this a deficiency with
    the formalism was discovered that led to the introduction of a
    LATPOLE FITS header card which may be used to disambiguate where
    CRVAL1, CRVAL2, and LONGPOLE do not uniquely determine the latitude
    of the native pole.  The CEL parameter array (formerly WCS) has been
    extended to accomodate LATPOLE as CEL(4), and the flag variable is
    now CEL(5) (formerly WCS(4)).

  - Default values of LONGPOLE and LATPOLE are now supported and their
    use is recommended where appropriate.

  - Numerical precision was being lost near the native poles in the SIN,
    AIR, and QSC projections and this has been recovered (reported by
    Lindsey Davis, NOAO).  Floating underflows in CSC are now avoided.

  - Numerical precision was also lost in certain circumstances in the
    spherical coordinate transformation routines and this has been
    fixed.

  - The test programs have been enhanced in various ways and the
    library has been validated on an SGI machine using both 32-bit and
    64-bit compilers.


WCSLIB version 1.0 (1995/01/31)
-------------------------------

* C library

  Initial release.

* Fortran library

  Initial release.

------------------------------------------------------------------------
$Id: CHANGES,v 4.14 2012/07/13 10:02:44 cal103 Exp $
