v4.4-2 -- 2 Feb 2024
  -change naming of callback functions (which must be declared extern
   "C") to start with "EMFcb_".  Previously, these functions were
   defined inside a C++ namespace called EMFcb; however, the namespace
   isn't included in the exported names, so this strategy was useless
   at preventing global naming conflicts.  Thanks to Tomas Kaliberña
   for identifying a windows header conflict and this easy solution.

v4.4-1 -- 28 June 2023
  -revert a change to the EMF+ page transform made in v4.4.
   Previously used (fake) "pixel" units but changed to "inch".  Turns
   out Word does not export to pdf properly when using an "inch" page
   transform.  Reverting back to "pixel".
  -fix sign bug in AFM descent calculation

v4.4 -- 14 June 2023
  -avoid linux font metric dependency on Xft, which requires a
   connection to an X server, and instead directly use Fontconfig.
  -fixed a typo that prevented multiple font families from being used
   in the same plot.
  -round instead of truncate when converting from logical units to mm
   to calculate dimensions in EMF header
  -fix typo when specifying figure width and height in mm

v4.3-1 -- 14 Apr 2023
  -changes in v4.3 did not consider the possibility that a *compiling*
   Mac might have different libraries available than the *installing*
   Mac (as can be the case when the package is built by CRAN and
   installed by users as a binary).  This version works around this
   problem by explicitly not trying to use xlib or xft if CoreText is
   available.

v4.3 -- 5 Apr 2023
  -modify autoconf configuration to explicitly test for CoreText on
   Apple platforms (since older OSX versions don't have CoreText) and
   allow use of zlib or xft on Apple if CoreText is unavailable.

v4.2 -- 17 Jan 2023
  -When drawing EMF (not EMF+) text strings, the EMF specification
   allows for supplying inter-character spacing, but the specification
   is ambigiuous as to whether this spacing is required.  In the
   absence of this inter-character spacing, all tested versions of
   LibreOffice and Microsoft Office have no problem loading and view
   EMF files.  However, if Office 365 imports an EMF file with
   individual EMF text strings >= 40 characters long that lack this
   spacing AND then the user attempts to export the Office 365
   document to PDF, then it fails with an error message.  This version
   of devEMF works around this issue by adding inter-character spacing
   to EMF text drawing records.
  -clean up and unify code for handling font metrics.  Add support for
   loading kerning from AFM files.
  -simplify autoconf script & #ifdefs to only specify Xft, not both
   Xft and Freetype, because the existence of Xft implies the
   existence of Freetype.

v4.1-2 -- 14 Dec 2022
  -bug fix: when generating an EMF plot, completely transparent fills
   were previously not handled correctly if they followed a solid
   color fill.  (Thanks to Mauricio Vargas for reporting.)
  -moved code from personal svn repository to public github; added
   github URL to DESCRIPTION file

v4.1-1 -- 30 Oct 2022
  -clipping rectangles (both EMF and EMF+) were previously being
   defined starting from the lower-left corner.  However, the EMF/EMF+
   spec says rectangles should be defined from the upper-left corner.
   Most viewing programs handled either way; however, when Word
   exports to PDF, it requires the definition to match the
   specification.  Fixed now, along with a couple other places
   rectangles were used.  (Thanks to Arnaud Atoch for reporting.)

v4.1 -- 12 May 2022
  -if making an EMF+ file but have the possibility of using any EMF
   drawing records (i.e., unless emfPlusFont=TRUE and
   emfPlusRaster=TRUE), use EMF clipping in addition to EMF+ clipping.
   Previously, if the file was EMF+, only did EMF+ clipping.  The old
   strategy worked with some viewing programs but not all.
  -tweak missing font metric warning message under Linux to remind
   user that they might want to install Xft
  -added "units" argument to emf() function to allow specification of
   device width & height in SI units (thanks to Mauricio Vargas for
   suggesting)

v4.0-2 -- 30 Sept 2020
  -add missing preprocessor conditional
  -enable linear gradient fill for circles

v4.0-1 -- 1 Sept 2020
  -fix configure script to handle case where Xft (or FreeType or Zlib)
   is available but not covered by pkg-config.

v4.0 -- 28 Aug 2020
  -new option emfPlusFontToPath that converts characters to EMF+
   graphics paths and thus makes graphics files portable (otherwise
   fonts must be installed on both the creating system and the viewing
   system).  This route is the closest the EMF/EMF+ spec gets to
   embedding fonts.  This option is implemented for Linux (via
   FreeType support) and Windows; OSX support is not yet implemented.

v3.8 -- 29 June 2020
  -accommodate R graphics engine version 13
  -add support for linear gradient fills (feature of GE13) for EMF+;
   other new features from GE13 do not appear possible to implement
   with EMF/EMF+
  -fix couple minor memory leaks

v3.7 -- 30 Sept 2019
  -For EMF+, request PixelOffsetModeHalf during file setup; this
   allows certain nominally adjacent drawn objects to appear as
   actually adjacent (without a thin gap between).
  -No longer create EMF+ record for polygon/path outline if transparent.
  -eliminate R CMD config CPP from configure.ac (deprecated and unnecessary)
  -request closest EMF/EMF+ text alignment (left-aligned /
   center-aligned / right-aligned); previously always requested
   left-aligned.  Since R allows for arbitrary horizontal alignment
   between [0,1], we still need to calculate text width and adjust the
   reference point manually.  This change works around minor slop in
   the calculation of the string width using font metrics (the degree
   of slop depends on the viewing program and can become noticeable
   with very long right-aligned strings).

v3.6-3 -- 20 June 2019
  -Workaround for recent Microsoft security patch that breaks the EMF
   specification and does not allow viewing of EMF files containing
   certain Font-related records in Microsoft Office.  Files using only
   EMF+ fonts are not affected.  The relevant CVE seems likely to be
   https://www.zerodayinitiative.com/advisories/ZDI-19-557/
  -Error check: return error if user requests negative or NA width/height
  -Bug fix: handle text & font metrics for Unicode characters consuming more
   than 2 bytes

v3.6-2 -- 2 Jan 2019
  -Bug fix: white text color not being set until a non-white text
   color was used first (bug only in EMF).

v3.6-1 -- 22 May 2018
  -emphasize inability to embed fonts in help page
  -attempt to fix compiler error on certain newer OSX systems

v3.6 -- 21 July 2017
  -for rotated text in EMF (as opposed to EMF+), insert rotation into
   font specification rather than using world transform.  This change
   makes for more ungainly files but greater compatibility with
   viewing programs that only support a subset of the EMF specification.
  -fix roundoff error when calculating page size in micrometers (in
   EMF header; unclear if used by any viewing program)

v3.5 -- 8 June 2017
  -implement clipping at EMF/EMF+ level (previously relied on R to
   clip, which is more aggressive, particularly with regard to text,
   and led to difference between EMF output vs. other devices)
  -only issue select object command for EMF font immediately before writing
   text (previously issued when first requested font metrics, which was not
   necessary)
  -change EMF world rotation commands to avoid depending on
   SAVE/RESTORE DC (which is unclear from spec whether it is supposed
   to save world rotation state).  This change avoids a bug triggered
   when printing figures with rotated text in Word 2010 (Thanks to
   Philip Harslett for reporting)

v3.4 -- 7 April 2017
  -add option for user to specify DPI used when setting up device
   coordinate system.  Previously was fixed at one dot per 0.01mm (aka
   2540 DPI; 0.01mm units are used elsewhere by EMF), but some lazy
   EMF clients appear to render the image as a bitmap of size equal to
   inches * DPI.. which can lead to out of memory errors.  Changing
   default to be 300 DPI.
  -fix handling of miter limit for miter-type line joins in EMF+ files.
  -correct R_init_* registration entrypoint name
  -use registered R symbol as entrypoint in .External call
  -fix PROTECT error (thanks to Tomas Kalibera's tool for identifying)

v3.3-1 -- 10 Feb 2017
  -fix edge case on linux where both Xft and zlib are installed but a
   connection cannot be made to the X server (i.e., if running R from
   ssh session without forwarding X connections)
  -add cleanup script

v3.3 -- 9 Feb 2017
  -fix additional Solaris compiler errors & warnings
  -add OSX linker option "-framework CoreText", which allows the
   devEMF shared library to be loaded separately from standard R (and
   hence pass cran checks).
  -fix memory leak in EMF font handling (EMF+ was ok)
  -fix memory leak in Windows GDI font metric code
  -fix minor memory leak in Linux font metric code

v3.2 -- 19 Sept 2016
  -add raster image support
  -embed devEMF version number in EMF output (in a comment field)
  -rework text output to use world transformations
  -bug fixes related to path rendering
  -eliminate compiler errors on Solaris C++98 compiler

v3.1 -- 17 Mar 2016
  -add back ability to use Adobe font metric (AFM) files, in case no
   graphical interface (Xft, Windows, Apple) is available to query.
   Adobe distributes AFMs for the 14 core PDF fonts, and these are now
   included in the devEMF package (note before v2.0, devEMF accessed
   the AFMs included in the grDevices package).  To keep this package
   size managable, the AFMs are gzipped, so using this ability
   requires zlib.
  -add support for path rendering

v3.0 -- 3 Apr 2015
  -add support for EMF+ output.  This format has less complete support
   in client programs but allows for partial transparency.  EMF+
   output is now the default, but the pure EMF output is still
   possible with an option to the "emf()" call that opens the device.

v2.0 -- 29 Jan 2015
  -switch to platform-specific font metric code instead of accessing
   the AFM files supplied with core R via a non-public R API (this
   previous approach violated CRAN policy).  The primary downside is
   that the new code will not operate on non-OSX, non-Windows
   platforms unless X windows is installed along with the Xft
   development headers.  It should work on OSX and Windows out of the
   box; please contact me if you have trouble (I primarily work on
   linux).
  -bug fix: EMF polygon record had potential rounding mismatch between
   vertex coordinates and bounding box coordinates.
  -bug fix: when PS_NULL, set pen BS_NULL (although seems unnecessary
   in practice)

v1.2-1  --  18 Feb 2013
  - Bug fix: EMF custom line types not working properly for solid
    lines. (Thanks to Nis Martensen for reporting and patching)

v1.2  --  22 Feb 2012
  - Bug fix: TextUTF8 used incorrect vertical alignment. (Thanks to
    Jared Decker for reporting)
  - change yLineBias to be 0.2 for compatibility with R >= 2.14

v1.1  --  6 Feb 2012
  - Add support for different line end and join styles. (Thanks to
    Mathias Walter for suggesting)

v1.0-1
  - Amend help to mention that partial transparency is not supported
    by the EMF format

v1.0
  - Been using for 6mo with no problems; bump version to 1.
  - Replace remaining few uses of printf with Rprintf

v0.9-4
  - Placed package within NAMESPACE (in prep for R-2.14.0)
  - modified ps_fonts.h to make compatible with R-2.14.0 revised
    handling of gzip i/o