glib2's gobject subsystem is essentially a wrapper for dlopen. In
view of comments in PR lib/49791 which can be summarised as
"RTLD_GLOBAL is a bug", make gobject use RTLD_DEFAULT instead.

(prlw1)

Fixed a typo

(adam)

Remove blender

(ryoon)

Extend PYTHON_FOR_BUILD_ONLY to test

(ryoon)

Added graphics/MesaLib18 version 18.2.0

(ryoon)

Add MesaLib18

(ryoon)

graphics/MesaLib18: import MesaLib-18.2.0 from wip/MesaLib

MesaLib is a 3-D graphics library with an API which is very similar to
that of OpenGL*.  To the extent that Mesa utilizes the OpenGL command syntax
or state machine, it is being used with authorization from Silicon Graphics,
Inc.  However, the author makes no claim that Mesa is in any way a
compatible replacement for OpenGL or associated with Silicon Graphics, Inc.

(ryoon)

Updated graphics/blender to 2.79b

(ryoon)

Update to

* Fix startup

Changelog:
2.79
In this release:

    Cycles: Built-in Denoising, Shadow catcher, Principled shader, AMD OpenCL optimizations.
    Grease Pencil: New frame interpolation tools, per-layer onion skinning.
    Alembic: Improvements to compatibility, stability and support.
    User Interface: Initial support for reusable custom configurations, automatic DPI scaling.
    Twenty Three new and several updated add-ons.
    And: 100s of bug fixes and other improvements!

2.78
In this release:

    Cycles: Cycles now supports rendering spherical stereo images for VR.
    Grease Pencil: Reworked to be similar to other 2D drawing softwares.
    Alembic: Initial basic support in Blender.
    Improvements to Bendy Bones for Easier and Simpler Rigging.
    Eleven new add-ons.
    And: 100s of bug fixes and other improvements!

(ryoon)

Updated graphics/openimageio to 1.8.15

(ryoon)

Update to 1.8.15

Changelog:
Release 1.8.15 (1 Oct 2018) -- compared to 1.8.14
-------------------------------------------------
* Change to internal thread pool to fix rare crash when lowering number of
  threads. #2013
* Build: Use pkgconfig in FindOpenEXR.cmake if available. #2008
* ImageBufAlgo::channel_append now will work with mixed input data types.
  #2022

Release 1.8.14 (1 Sep 2018) -- compared to 1.8.13
-------------------------------------------------
* Support compilation by clang 7.0. #1995
* Bug fix to ImageBuf::roi_cointains(). #1997
* Fix ImageSpec constructor from an ROI, display/"full" window did not get
  the right default origin. #1996
* GIF read: fix crash when reading GIF with comment extension but no
  comment data. #2001
* JPEG write: robust to accidentally setting the "density" metadata to
  values larger than JPEG's 16 bit integer field will accommodate. #2002
* ImageSpec::erase_attribute() fix bug where it got case-sensitivity of the
  search backwards when built using std::regex rather than boost::regex.
  #2003

Release 1.8.13 (1 Aug 2018) -- compared to 1.8.12
-------------------------------------------------
* Improved finding LibRaw on Windows #1959
* Ensure FFMPEG (if found) is the minimum supported version 2.6.  #1963
* Developer goodies: `intrusive_ptr<>` given a dangerous `release()`
  method (use with caution). #1986

Release 1.8.12 (1 Jun 2018) -- compared to 1.8.11
-------------------------------------------------
* Build fix for 32 bit Windows -- disable SSE that was making trouble. #1933
* Improved UTF-8 filename support for OpenEXR filenames. #1941
* filesystem.h: Improve UTF-8 handling for unique_path and temp_directory
  functions. #1940
* Fixes to un-break the build when EMBEDPLUGINS=0. #1942
* Fix iv (especially on OSX) for various ways it has been broken since the
  shift to Qt5. #1946
* Improvements in finding the location of OpenJPEG with Macports. #1948
* ImageCache and TextureSystem now have close(filename) and close_all()
  methods, which for one file or all files will close the files and release
  any open file handles (also unlocking write access to those files on
  Windows), but without invalidating anything it knows about the ImageSpec
  or any pixel tiles already read from the files, as would happen with a
  call to the much more drastic invalidate() or invalidate_all(). #1950

Release 1.8.11 (1 May 2018) -- compared to 1.8.10
-------------------------------------------------
* Fix to strtof, strtod for non-C locales. #1918
* Add up-to-date Nuke versions to FindNuke.cmake. #1920
* Allow building against ffmpeg 4.0. #1926

Release 1.8.10 (1 Apr 2018) -- compared to 1.8.9
-------------------------------------------------
* oiiotool frame sequence wildcard improvements: fix handling of negative
  frame numbers and ranges, also the `--frames` command line option is not
  enough to trigger a loop over those frame numbers, even if no other
  arguments appear to have wildcard structure. #1894
* TIFF bug fix: read_tile() and read_tiles() input of un-premultiplied tiles
  botched the "shape" of the tile data array. #1907
* Windows/MSVC build fix: use the `/bigobj` option on some large modules
  that need it. #1900, #1902
* fmath.h, hash.h, color.h: changes to make it friendly to Cuda compilation
  (#1888, #1896, #1905).
* fmath.h avx-512 improvements. #1893
* testsuite is not Python 2/3 agnostic.

Release 1.8.9 (1 Mar 2018) -- compared to 1.8.8
-------------------------------------------------
* Properly find newer openjpeg 2.3. #1871
* Bug fix in IBA::copy where uninitialized dst image botched its ROI. #1876
* RAW: Important bug fix when dealing with rotated (and vertical) images,
  which were not being re-oriented properly and could get strangely
  scrambled. #1854

Release 1.8.8 (1 Feb 2018) -- compared to 1.8.7
-------------------------------------------------
* OpenEXR: gracefully detect and reject files with subsampled channels,
  which is a rarely-to-never-used OpenEXR feature that we don't support
  properly. #1849
* Field3d: Prevent crashes when open fails. #1848
* RAW: Add "raw:HighlightMode" configuration hint to control libraw's
  handling of highlight mode processing. #1851
* zfile: more careful gzopen on Windows that could crash when given bogus
  filename. #1839
* DICOM: Fix dcmtk build errors on some platforms. Also, the minimum dcmtk
  version we suport is 3.6.1. #1843
* simd.h: Minor fixes especially for avx512. #1846
* iv: Drop GLEW and obsolete GL stuff from iv in favor of QOpenGLFunctions,
  and fix broken pixelview text rendering. #1834
* On Unix/Linux, add explicit DL library dependency to libOpenImageIO.so
  itself instead of only to the binaries and test utilities. #1860
* Build fixes for Hurd OS. #1850

Release 1.8.7 (1 Jan 2018) -- compared to 1.8.6
-------------------------------------------------
* All string->numeric parsing and numeric->string formatting is now
  locale-independent and always uses '.' as decimal marker. #1796
* oiiotool outputs are now written to temporary files, then atomically moved
  to the specified filename at the end. This makes it safe for oiiotool
  to "overwrite" a file (i.e. `oiiotool in.tif ... -o out.tif`) without
  problematic situations where the file is truncated or overwritten
  before the reading is complete. #1797
* Python bindings for ImageBuf.get_pixels and set_pixels fixed some bugs
  when passed an ROI without a channel range specified. #1802
* More robust parsing of XMP metadata for unknown metadata names. #1816
* strutil.h now includes a to_string<> utility template. #1814

Release 1.8.6 (1 Nov 2017) -- compared to 1.8.5
-------------------------------------------------
* oiiotool: Improved logic for propagating the pixel data format through
  multiple operations, especially for files with multiple subimages.
  #1769
* ImageCache/TextureSystem: improved stats on how long we wait for
  ImageInput mutexes. #1779
* Build: Fix build bug where if the makefile wrapper got CODECOV=0, it would
  force a "Debug" build (required for code coverage tests) even though code
  coverage is instructed to be off. (It would be fine if you didn't specify
  CODECOV at all.) #1792
* Build: minor fixes for Windows build. #1793, #1794
* Developers: The ParamValue class has added get_int_indexed() and
  get_float_indexed() methods. #1773
* Developers: array_view added begin(), end(), cbegin(), cend() methods,
  and new constructors from pointer pairs and from std::array.
* Developers: Fixed build break in simd.h when AVX512VL is enabled. #1781
* Developers: fmath.h now defined OIIO_FMATH_H so other files can easily
  detect if it has been included.
* Build: Fix broken build when Freetype was not found or disabled. #1800
* Python: fixed missing exposure of RATIONAL enum value. #1799

Release 1.8 (1.8.5 - 1 Oct 2017) -- compared to 1.7.x
----------------------------------------------
New minimum dependencies:
* **C++11** (should also build with C++14 and C++17)
* **Compilers**: gcc 4.8.2 - gcc 7, clang 3.3 - 5.0, or MSVS 2013 - 2017
* **Boost >= 1.53** (tested up through 1.65)
* **CMake >= 3.2.2** (tested up through 3.9)
* **OpenEXR >= 2.0** (recommended: 2.2)
* (optional) **Qt >= 5.6**
* (optional) **Python >= 2.7** (3.x is also ok)

**Changes to install layout**: fonts now get installed to
  `PREFIX/share/fonts/OpenImageIO`, OIIO docs now get installed to
  `PREFIX/share/doc/OpenImageIO`, and the Python module gets installed to
  `PREFIX/lib/pythonMAJ.MIN/site-packages`. #1747 #1760 (1.8.5)

Major new features and improvements:
* New oiiotool features:
  * `--info:format=xml` format option requests what format the info
      is printed. Current choices: `"text"`, `"xml"`. #1504 (1.8.0)
  * `--info:verbose=1` verbose option make file info print full metadata,
    but without needing to make other oiiotool operations verbose as would
    happen with `--info -v`. #1504 (1.8.0)
  * `--colormap` applies a color map based on the input values; the
    map can be one of several named ones, or given explicitly with
    numerical values. #1552 (1.8.1)
  * `-i:type=...` lets you override the internal buffer type that will be
    used for an input image. #1541 (1.8.1)
  * `-i:ch=a,...` lets you restrict the input to only the listed channels.
    This is semantically equivalent to following the input with a `--ch`
    command, but by integrating into the input itself, it can sometimes
    avoid using memory and I/O for unneeded channels. #1541 (1.8.1)
  * `--echo STRING` prints the string to the console. This can contain
    expressions! So you can do something like
        oiiotool file.exr -echo "Size is {TOP.width}x{TOP.height}"
    #1633 (1.8.3)
  * `--eraseattrib REGEX` erases all metadata attributes from the top image
    whose names match the regular expression. #1638 (1.8.3)
  * `--text` takes new optional modifiers: `xalign=` (left, right, center)
    and `yalign=` (baseline, top, bottom, center) to control how the text
    is aligned to the (x,y) position specified, `shadow=` (default = 0)
    than when nonzero controls the width of a "drop shadow" that makes the
    text clearer when rendered on a background image of similar color.
    #1646 (1.8.3)
  * `--deepholdout` culls all samples that are farther away than the
    opaque depth of a second holdout image. #1691 (1.8.4)
* New ImageBufAlgo functionality:
  * `color_map()` applies a color map based on the input values; the
    map can be one of several named ones, or given explicitly with
    numerical values. #1552 (1.8.1)
  * Added implementation of ImageBufAlgo::to_IplImage(). #1461 (1.7.9/1.8.1)
  * `render_text()` has added parameters controlling text alignment and
    drop shadows. #1646 (1.8.3)
  * `deep_holdout()` culls all samples that are farther away than the
    opaque depth of a second holdout image. #1691 (1.8.4)
* DICOM file format support (currently read-only). DICOM is the standard for
  medical imaging data. Will only build if dependency "dcmtk" is found at
  build time. #1534 (1.8.1)
* Experimental: The TextureSystem API has been extended to support batches
  of texture lookups in a SIMD fashion. At present, only the new API was
  added, a full implementation is in progress and may not be forthcoming
  or reliable until 1.9. Even the API is experimental, and may change for
  future releases. #1733 (1.8.5)

Public API changes:
* TypeDesc:
  * Rational support: new 'semantic' hint RATIONAL and TypeDesc::Rational.
    A rational is an int of aggregate VEC2 and hint RATIONAL, and should
    be interpreted as val[0]/val[1]. #1698 (1.8.5)
  * Added OIIO-scoped `static constexpr` versions of preconstructed
    TypeDescs (e.g., `TypeFloat`). We are deprecating the ones that were
    static data members of TypeDesc (e.g., TypeDesc::TypeFloat), they will
    be removed in some future release. (1.8.5)
* ImageSpec:
  * New `ImageSpec::serialize()` returns a string with a serialized version
    of the contents of the ImageSpec. It may be text (human readable, like
    is printed by `oiiotool -info -v`) or XML. #1504 (1.8.0)
  * `ImageSpec` has a new constructor that accepts a `ROI` for image
    dimensions and channels. #1646 (1.8.3)
  * New `ImageSpec::channelspec()` retrieves the index of a named channel.
    #1691 (1.8.4)
  * New `ImageSpec::channelname(int)` safely retrieves the index of a named
    channel. #1706 (1.8.5)
* ColorConig::createLookTransform() and createDisplayTransform() have been
  extended to allow multiple key/value context pairs, by making them
  comma-separated lists. The createColorProcessor() method has also been
  extended to take context key/value pairs. #1542 (1.7.8, 1.8.0)
* `ImageBuf::read()` now has a variety that takes a channel range, allowing
  you to populate an ImageBuf with a subset of channels in a file,
  potentially saving memory and I/O for the unneeded channels. #1541 (1.8.1)
* Python: Fix unimplemented ImageBufAlgo.computePixelStats. #1596
  (1.8.2/1.7.11)
* imageio.h: Fix incorrect declaration of declare_imageio_format().
  #1609 (1.8.2/1.7.11)
* `ImageBuf::wrap_mode_from_string()` converts string wrap mode names
  (such as "black") into `ImageBuf::WrapMode` enum values. #1615 (1.8.3)
* New `OIIO::getattribute()` queries:
  * `"input_format_list"` and `"output_format_list"` return comma-separated
    lists of all formats that support input and output, respectively.
    #1577 (1.8.3)
  * `"oiio:simd"` returns a comma-separated list of SIMD capabilities that
      were enabled at build time, and `"hw:simd"` returns the list of
      capabilities available on the currently running hardware. #1719 (1.8.5)
* `ImageBufAlgo::render_text()` API call has been overhauled, in addition
  to new alignment and shadow aprameters, the color has changed from a raw
  pointer to an `array_view<const float>` for better memory safety. Also,
  it is now valid for the destination image to be uninitialized, in which
  case it will be initialized to be just big enough for the text.
  #1646 (1.8.3)
* DeepData:
  * New `DeepData::opaque_z()` returns the depth value at which a pixel
    becomes fully opaque. #1691 (1.8.4)
  * New `DeepData::initialized()` and `allocated()` return whether the
    DD is initialized and allocated, respectively. #1691 (1.8.4)
  * `DeepData::split()` has been changed to return a `bool` indicating
    whether any split occurred. #1691 (1.8.4)
* Remove some long-deprecated varieties of `ImageBufAlgo::colorconvert()`,
  `ociolook()`, and `ociodisplay()`. #1695 (1.8.4)
* TypeDesc now allows specification of "rational" values, using a vec2i
  (aggregate 2-vector of int) with a semantic hint of RATIONAL, i.e.,
  `TypeDesc(INT, VEC2, RATIONAL)`, which is also aliased as
  `TypeDesc::TypeRational`. The value is understood to be val[0]/val[1].
  #1698 (1.8.5)
* ParamValueList::get_float will automatically convert rational values to
  float. #1698 (1.8.5)
* The standard metadata "FramesPerSecond" has had its definition changed
  from `float` to `rational`. Retrieving it as `float` should still work
  as always. But apps and plugins that wish to treat it as a true rational
  with no loss of precision are able to do so. This is only known to
  directly affect the OpenEXR, GIF, and FFMPEG metadata. #1698 (1.8.5)

Fixes, minor enhancements, and performance improvements:
* oiiotool:
  * `--chappend` resolves redundant channel names by using the subimage
    name, if available. #1498 (1.8.0/1.7.8)
  * `--mosaic` now gracefully handles the case of not having enough
    images to completely fill the MxN matrix, with "left over" slots
    black. #1501 (1.8.0/1.7.8)
  * When command line arguments fail to parse properly, `oiiotool` will
    exit with a non-zero shell status. #1540 (1.8.0)
  * `--colorconvert`, `--ociodisplay`, and `--ociolook` can now take
    multiple context key/value pairs (by allowing `key=` and `value=`
    optional paramters be comma-separated lists). #1504 (1.8.0)
  * Handle 'oiiotool --colorconvert X X' (transform from and to spaces that
    are the same) without considering it an error. #1550 (1.8.0/1.7.8)
  * Expression substitution now recognizes the following new metadata
    names: MINCOLOR, MAXCOLOR, AVGCOLOR. An example use is to stretch
    the value range of an image to fill the full [0-1] range:
        oiiotool in.exr -subc {TOP.MINCOLOR} -divc {TOP.MAXCOLOR} -o out.exr
    #1553 (1.8.1)
  * `--fit:exact=1` use of the new `exact=1' option will perform the resize
    to preserve exact aspect ratio and centering of the fit image to true
    sub-pixel precision (at the possible risk of slight blurring,
    especially of the edges), whereas the default (`exact=0`) will keep
    the image sharper but round the size and offset to the nearest whole
    pixel values. (1.8.1).
  * `-i:type=...` optional modifier to `-i` forces a read of the input
    file into a particular data type. #1541 (1.8.1)
  * `-i:ch=...` optional modifier to `-i` specifies that only certain named
    channels should be read from the file. Under extreme circumstances
    when only a small subset of channels is needed from an image file with
    many channels, this may improve speed and I/O. #1541 (1.8.1)
  * Improved logic governing output data formats: non-cached inputs didn't
    set default output data format correctly, and per-channel output
    formats updated defaults when they shouldn't have. #1541 (1.8.1)
  * `--diff` : in addition to the pixel coordinates and differences of the
      biggest differing pixel, it now also prints the full values of all
      channels of that pixel for both images. #1570 (1.8.1)
  * `oiiotool -d` giving per-channel formats is no longer confused by
    channel renaming with `--chnames`. #1563 (1.8.1/1.7.9)
  * `--debug` nor prints the total runtime and peak memory use after each
    individual op. #1583 (1.8.1)
  * `-iconfig` not in all cases correctly propagate the input config
    attribute to the file read. #1605 (1.8.2/1.7.11)
  * Fixed `--crop`: it did not honor the `-a` flag to apply the crop to
    all subimages. #1613 (1.8.2/1.7.11)
  * In the case of runtime errors, `oiiotool` now echoes the entire command
    line. This is helpful for debugging mangled oiiotool command lines
    assembled by scripts. #1614 (1.8.3)
  * Improved error reporting of file open errors when -iconfig is used.
    #1626 (1.8.3/1.7.13)
  * Expression evaluation now substitutes `FRAME_NUMBER` with the numeric
    frame number for frame sequence wildcards, and `FRAME_NUMBER_PAD`
    with the frame number 0-padded to the number of total digits specified
    by the command line `--framepadding` argument. #1648 (1.8.3)
  * `--resize` and `--resample` now have more intuitive behavior for images
    where the display and pixel data windows are not the same, especially if
    the data window had a nonzero origin (such as with crop or overscan).
    #1667 (1.8.4/1.7.14)
  * `--resample` has been extended to work for "deep" images. #1668
    (1.8.4/1.7.14)
  * `--deepmerge` now will give a useful error message when the image do
    not have the same number of channels. #1675 (1.8.4/1.7.14)
  * `--autocc` more gracefully handles unknown color spaces with a warning,
    rather than a full error and termination. #1681 (1.8.4)
  * `--resample` now takes an optional modifier `interp=0` to control
    whether bilinear sample is used (default) or true closest-pixel point
    sampling. #1694 (1.8.4)
  * You can set rational metadata on the command line like this:
      `oiiotool foo.exr --attrib:type=rational onehalf "50/100" -o rat.exr`
      #1698 (1.8.5)
  * `--fillholes` fixed a bug where, if asked to operate on an image with
      nonzero origin (crop, overscan, shrink-wrap), it would incorrectly
      move the pixels to the origin. #1768 (1.8.5)
* ImageBufAlgo:
  * `channel_append()` resolves redundant channel names by using the
    subimage name, if available. #1498 (1.8.0/1.7.8)
  * `colorconvert()`, `ociodisplay()`, and `ociolook()` can now take
    multiple context key/value pairs (by allowing they `context_key` and
    `context_value` paramters be comma-separated lists). #1504 (1.8.0)
  * `draw_rectangle()` (and `oiiotool --box`) wasn't drawing properly for
    the degenerate case of a rectangle that takes up just one
    pixel. #1601 (1.8.2)
  * `resample()` has been extended to work for "deep" images.
      #1668 (1.8.4/1.7.14)
  * `deep_merge()` now will give a useful error message when the images do
    not have the same number of channels. #1675 (1.8.4/1.7.14)
  * `deep_merge()` performance has been greatly improved. #1739 (1.8.5)
  * `resample()` fixed a subtle 1/2 pixel shift, now it more closely
      aligns with `resize()`. #1694 (1.8.4)
  * `fillholes_pushpull()` fixed a bug where, if asked to operate on an
      image with nonzero origin (crop, overscan, shrink-wrap), it would
      incorrectly move the pixels to the origin. #1768 (1.8.5)
* ImageBuf:
  * Fix broken threads(n) method, which didn't correctly pass the right
    number of threads along. #1622. (1.8.3/1.7.12)
  * Copy constructor from another ImageBuf was previously broken for
    IB's that wrap application buffers. #1665 (1.8.4/1.7.13)
* TextureSystem / ImageCache / maketx:
  * `IC::get_image_info` (or `TS::get_texture_info`) queries for "channels"
    on UDIM file patterns now succeed, returning the value for the first
    matching file it finds. (N.B.: Relies on all textures within the same
    UDIM set having the same nchannels.) #1502, #1519, #1530 (1.8.0/1.7.8)
  * maketx: multiple simultaneous maketx process trying to create the same
    texture will no longer clobber each other's output. #1525 (1.8.0/1.7.8)
  * ImageCache: make robust to changes in autotile after opening and reading
    from untiled files. #1566 (1.8.1/1.7.9)
  * ImageCache: fix initialization bug that made the reported stats output
    nonsensical in the numbers it gave for "redundant reads". #1567
    (1.8.1/1.7.9)
  * get_image_info queries of "displaywindow" and "datawindow" did not
    correctly return a 'true' value when the data was found.
    #1574 (1.8.1/1.7.9)
  * maketx fix: two textures that had identical source pixels but differed
    in whether they resized with "highlight compensation" incorrectly
    ended up with identical hashes, and thus could be confused by the
    TextureSystem at runtime into thinking they were duplicates. The hash
    is now fixed. #1599 (1.8.2/1.7.11)
  * Statistics no longer list as "BROKEN" files which were invalidated, or
    files that were initialized with an ImageCacheFile but never opened.
    #1655 (1.8.4)
  * ImageCache::get_image_info() will now return a proper error (and not
    hig an assertion) if asked for information about a subimage or MIP
    level that does not exist in the file. #1672 (1.8.4/1.7.14)
  * TextureSystem::get_texels fixes crashing behavior. #1669 (1.8.4/1.7.14)
  * Big performance improvement on Windows with MSVC by removing certain
    empty destructors in simd.h that was preventing MSVC from fully inlining
    those classes. #1685 (1.8.4/1.7.15)
  * maketx now supports `--colorconfig` option to explicitly point it to
    an OpenColorIO config file, just like `oiiotool `--colorconfig`.
    #1692 (1.8.4)
  * Fix rare edge case crash in ImageCache. #1696 (1.8.4/1.7.15)
  * Improved error messages for broken files. Specifically, it's much more
    clear now when a file is broken because it's being rejected as a
    texture because it's untiled or not MIP-mapped. #1751 (1.8.5)
  * The maketx-generated metadata "oiio:SHA-1", "oiio:ConstantColor" and
    "oiio:AverageColor" are ignored if the file has signs that it was not
    directly generated by `maketx` or `oiiotool -otex` (specifically, if
    it's not tiled, has no "textureformat" tag, or if its "software" tag
    doesn't mention maketx or oiiotool). This helps for the case where a
    maketx-generated file is loaded into PhotoShop (or otherwise altered),
    saved with different pixel values but the old SHA-1, which would no
    longer be valid and therefore cause the new file to be misidentified
    as a duplicate texture even though it's not. #1762 (1.8.5)
* Bug fix to possible crashes when adding dither to tiled file output
  (buffer size miscalculation). #1518 (1.8.0/1.7.8)
* Make sure that sRGB<->linear color transform still work (in the obvious
  way) even when OpenColorIO is present but that its configuration for some
  reason doesn't know about "sRGB" space. #1554 (1.8.1)
* Improved performance of input of many-channel files with differing
  per-channel data formats. #1541 (1.8.1)
* `idiff` : in addition to the pixel coordinates and differences of the
  biggest differing pixel, it now also prints the full values of all
  channels of that pixel for both images. #1570 (1.8.1)
* ImageInput::read_tiles when only a channel subset is read, fixed case
  with certain data sizes where the copy to user buffer got mangled.
  #1595 (1.8.2/1.7.11)
* BMP:
  * Add support for version 5 of the BMP format header. $1616 (1.8.3/1.7.12)
* FFMpeg/movies:
  * "FramesPerSecond" metadata has had its type changed to rational.
    #1709 (1.8.5)
* GIF:
  * "FramesPerSecond" metadata has had its type changed to rational.
    #1709 (1.8.5)
* IFF:
  * Fix IFF output that didn't correctly save the "Author" and "Date"
    metadata. #1549 (1.8.1/1.7.8)
* JPEG:
  * Be more reslient to malformed Exif data blocks with bogus offsets
    #1585 (1.8.1/1.7.10) and #1639 (1.8.3/1.7.13).
  * When you ask the JPEG writer to output files with unsupported channel
    counts (JFIF files only allow 1 or 3 channels), it will just silently
    drop extra channels, rather than having a hard error. #1643 (1.8.3)
* OpenEXR:
  * Fix global attribute "exr_threads" value, -1 should disable IlmImf's
    thread pools as advertised. #1582 (1.8.1)
  * Allow compression "none" for deep exr files. (1.8.2/1.7.11)
  * Fixed input problem with sorting order of spectral alpha channels (RA,
    GA, BA, or AR, AG, AB). #1674 (1.8./1.7.14)
  * Can handle true rational metadata, including FramesPerSecond and
    captureRate. #1698 (1.8.5)
  * Fix problem with 2-channel images putting the channels in the wrong
    order. #1717 (1.8.5/1.7.16)
* PNG: Better extraction of XMP from PNG files. #1689 (1.8.4)
* PSD:
  * Support has been added for "cmyk", "multichannel", and "grayscale"
    color modes. And support was fixed for rgb and grayscale 32 bit per
    sample bit depth. #1641 (1.8.3/1.7.13)
  * Fix issue for layer mask channels. #1714 (1.8.5)
* RAW:
  * Fix possible crash when reading certain raw metadata. #1547 (1.7.8/1.8.0)
  * The default value for missing "raw:use_camera_matrix" has been changed
    to 1 (not 0) to better match default behavior of libraw/dcraw.
    #1629 (1.8.3/1.7.13)
  * Add support for selecting new demosaicing algorithms: "DMT" (mode 11)
    and "AAHD" (mode 12). Renamed the "Modified AHD" to "AHD-Mod" to
    simplify and match libraw terminology. Made matching of demosaicing
    algorithms case-insensitive. #1629 (1.8.3/1.7.13)
  * Support "ACES" color space for direct conversion while reading RAW
    images (supported in libraw 0.18 or newer). #1626 (1.8.3/1.7.13)
  * Add "raw:user_sat" configuration attribute to the reader.
    #1666 (1.7.15/1.8.4)
  * The pixels are now decoded (expensive) only when they are read, not
    when the file is first opened. This makes raw reading much faster for
    apps that are only interested in the metadata and not the pixel data.
    #1741 (1.8.5)
  * Unpack pixels when they are needed, not when the file is opened. This
    makes it much faster to read RAW file only to extract the metadata,
    if you don't need the pixel values. #1741 (1.8.5)
* RLA:
  * Fix RLA reading and writing with certain channel orders and mixded data
    formats. #1499 (1.8.0/1.7.8)
* TIFF:
  * Fix to TIFF handling of certain unusual tags. #1547 (1.7.8/1.8.0)
  * Now has a way to read raw pixel values from CMYK files, without
    the automatic conversion to RGB (pass configuration attribute
    "oiio:RawColor" set to nonzero). #1605 (1.8.2/1.7.11)
  * Improved I/O of color separation images, particularly those with
    custom InkSet attributes. #1658 (1.8.4/1.7.15)
  * Fix typo that prevented correct reading of some Exif fields. #1625
    (1.8.3/1.7.12)
  * TIFF output omitted setting the "Make" and "Model" metadata tags.
    #1642 (1.8.3/1.7.13)
  * Images with fewer than 4 channels, but one of those channels was alpha,
    were not correctly marking spec.alpha_channel. #1718 (1.8.5/1.7.16)
  * The XPOSITION and YPOSITION tags are now interpreted as relative to
    the RESOLUTIONUNIT, whereas before it was assumed to be measured in
    pixels. We are confident that the new way is more in line with the
    intent of the TIFF spec. #1631 (1.8.5)
* webp:
  * Several new sanity checks prevent the webp reader from spending too
    much I/O time and memory reading bogus files (malformed, corrupted,
    or not a webp after all). #1640 (1.8.3/1.7.13)
* Nuke plugin: Fix txReader to properly restore saved MIP level knob value.
  #1531 (1.8.0)
* Fixed several (so far unnoticed) buffer overruns and memory leaks.
  #1591 (1.8.2)
* TIFF, JPEG, others: Improved reading Exif meatdata from XMP blocks, now it
  does a better job of discerning the proper data types. #1627 (1.8.3/1.7.13)
* In several places, the formatting of date metadata has been changed to
  have a leading zero rather than leading space for hours < 10, i.e.,
  "02:00:00" rather than " 2:00:00". #1630 (1.8.3)
* Improve XMP parsing (for any format): for malformed XMP, still honor the
  parts that could be parsed properly; recognize additinoal tags for
  GPano (Google's Photo Sphere metadata schema) and camera raw (crs: prefix)
  metadata; improve speed of XMP parsing.  #1679 (1.8.4)
* Improved handling and color conversion of gamma-corrected images (DPX,
  HDR, PNG, RLA, Targa) by supporting linearization correctly even in the
  presence of OCIO configs that don't know about it. #1684 (1.8.4)
* Fixed static initialization order fiasco error involving interaction
  between ColorConfig and Strutil. #1757 (1.8.5)

Build/test system improvements:
* **Changes to install layout**: fonts now get installed to
  `prefix/share/fonts/OpenImageIO`, OIIO docs now get installed to
  `prefix/share/doc/OpenImageIO`, and the Python module gets installed to
  `prefix/lib/pythonMAJ.MIN/site-packages`. #1747 #1760 (1.8.5)
* Support for building against ffmpeg 3.1 (their API has changed).
  #1515 (1.8.0/1.7.8)
* Build no longer gets confused about include files from older installations
  of OIIO elsewhere on the system. #1524 (1.8.0/1.7.8)
* Improvements in finding OpenJPEG. #1520 (1.8.0/1.7.8)
* Improved finding of OCIO headers. #1528 (1.8.0/1.7.8)
* Fix compile warnings for Clang 3.9. #1529 (1.8.0/1.7.8)
* Minimum C++ standard of C++11 is expected and enforced. #1513 (1.8.0)
* Minimum Boost is now 1.53. #1526 (1.8.0)
* Fix compiler warning on Fedora aarch64. #1592 (1.8.1)
* Tweak OpenJPEG include file search rules to prefer newer versions when
  multiple are installed. #1578 (1.8.1)
* Build option `SANITIZE=...` lets us use the sanitizers. #1591 (1.8.2)
* Big refactoring of the cmake build files. #1604 (1.8.2)
* When using a recent enough C++ compiler and standard library, OIIO will
  now use C++11 std::regex rather than boost regex. #1620,#1644 (1.8.3)
* Support for clang-tidy using build time flags CLANG_TIDY=1,
  CLANG_TIDY_ARGS=..., and optionally CLANG_TIDY_FIX=1. #1649 (1.8.4)
* Fix Windows warnings about SIMD types as function arguments
  and about bool vs int. #1659 (1.8.4)
* Changed the way namespaces work. Instead of a 2-level namespace, make a
  one-level namespace, and automatically embed the major and minor version
  in it. (Can stll override the basename.) #1662 (1.8.4)
* Fixes to OSX rpath behavior of linked binaries. #1671
* Upgraded the local PugiXML (when not using an external system version)
  to release 1.8.1.  #1679 (1.8.4)
* Beef up Strutil unit tests. (1.8.5)
* `iv` has been upgraded to use Qt 5.x. Support for Qt 4.x is hereby
  deprecated. #1711 (1.8.5)
* Make the search for boost_python3 more reliable. #1727 (1.8.5)
* Fix python site-packages path for installation. #1722 (1.8.5)
* Fixes for building with gcc 7. (1.8.5)
* Support and fixes for building with clang 5.0. #1746 (1.8.5)
* Support/fixes for Boost 1.65. #1553 (1.8.5)
* Simplify CMake scripts by using GNUInstallDirs to set standard installation
  paths. #1747 (1.8.5)

Developer goodies / internals:
* Sysutil::Term formatting now works properly in Windows (though is only
  functional for Windows 10 or above). (1.8.0/1.7.8) #1527
* C++11 idioms:
  * We now eschew BOOST_FOREACH, in favor of C++11 "range for". #1535
  * We now use std::unique_ptr, no longer use boost::scoped_ptr or
    boost::scoped_array. #1543 (1.8.0) #1586 (1.8.1)
  * Instead of the various boost components, we now use `std::` versions of
    unordered_map, unordered_set, shared_ptr, hash, bind. #1586 (1.8.1)
  * Change deprecated C headers (such as `<ctype.h>`) to C++ (`<cctype>`).
    #1649 (1.8.4)
  * Use `std::vector<>::emplace_back()` where applicable. #1657 (1.8.4)
  * Mark ImageInput/ImageOutput derived classes as 'final'. (1.8.5)
* array_view.h:
  * Add front() and back() methods. #1724 (1.8.5)
  * Simplified array_view template to be 1D only. #1734 (1.8.5)
* atomic.h:
  * Added atomic_min and atomic_max. #1661 (1.8.4)
  * Added atomic_fetch_add for `std::atomic<float>` and double. #1661 (1.8.4)
  * Assume std::atomic is available, remove all code that is only needed
    for pre-C++11. #1661 (1.8.4)
* benchmark.h:
  * New `Benchmarker` class utility for micro-benchmarking. #1577 (1.8.5)
  * Moved time_trial, timed_thead_wedge, DoNotOptimize, clobber_all_memory
    to benchmark.h. #1577 (1.8.5)
* errorhandler.h: Change all ErrorHandler methods to use variadic templates
  rather than varargs. #1653 (1.8.4)
* filesystem.h:
  * Better exception safety for Filesystem::searchpath_find. #1680 (1.8.4/1.7.15)
* fmath.h:
  * Fixed typo in fmath.h that made bitcast_to_float incorrect. #1543 (1.8.0)
  * Templatize round_to_multiple() so it works with types other than `int`.
    #1548 (1.8.0)
  * `interpolate_linear()` utility linearly interpolates from a list of
    evenly-spaced knots. #1552 (1.8.1)
  * Slight reformulation of clamp() ensures sane results even with NaN
    parameters. #1617 (1.8.3)
  * Bug fixes to `ifloor()` and `floorfrac()`, which turned out to give
    incorrect results for exact negative integer values. Also added
    simd vector-based versions of `floorfrac`. #1766 (1.8.5)
* paramlist.h:
  * ParamValueList has been refactored and now inherets from, rather than
    containts, a `std::vector<ParamValue>`. This removes most of the
    additional code from the class. #1677 (1.8.4)
  * ParamValue new methods: `get<>()`, `get_int()`, `get_float()`,
    `get_string()`, `get_ustring()` retrieve and convert. #1686 (1.8.4)
  * ParamValueList new methods: `get_int()`, `get_float()`, `get_string()`,
    `get_ustring()` search, retrieve, and convert (much like the
    equivalent versions in ImageSpec did). #1686 (1.8.4)
  * Remove the pointless typedefs `ImageIOParameter` and
    `ImageIOParameterList` in favor of `ParamValue` and `ParamValueList`,
    respectively. #1690 (1.8.4)
* platform.h:
  * More `cpu_has_...()` tests for newer CPU capabilties. #1719 (1.8.5)
  * Remove deprecated OIIO_NOTHROW macro, which should now simply be
    C++11 `noexcept`. #1736 (1.8.5)
* strutil.h / Strutil:
  * Add `Strutil::printf()` and `Strutil::fprintf()`, typesafe and
    non-thread-jumbled replacements for C versions. #1579, #1656 (1.8.1, 1.8.4)
  * `from_string<>` has been extended to 'unsigned int'. (1.8.3/1.7.13)
  * `Strutil::parse_identifier_if` #1647 (1.8.3)
  * safe_strcpy now takes a string_view, and more closely conforms to the
    behavior of strncpy by filling in the extra space with 0 padding.
    (1.8.5)
* simd.h:
  * Add a matrix44 constructor from 16 floats. #1552 (1.8.1)
  * Renamed files floatN, intN, boolN to vfloatN, vintN, vboolN, to avoid
    confusion with bit lengths of scalars. #1719 (1.8.5)
  * Overhaul to support AVX-512 via vfloat16, vint16, vbool16 classes.
    #1719 (1.8.5)
  * load_mask, store_mask, and scatter/gather added for all types.
    #1732 (1.8.5)
  * Renamed `floori()` to `ifloor()` to match the analogous scalar function
    in fmath.h. #1766 (1.8.5)
* strided_ptr.h: The `strided_ptr<>` template is now also templated on
    the stride units -- the default is to measure strides in units of
    sizeof(T), but it's possible to override and measure in bytes (the old
    behavior). #1758 (1.8.5)
* thread.h:
  * thread_pool class offers true persistent thread pool.
    #1556, #1581 (1.8.1)
  * Lots of C++ modernization of thread_group and spin_mutex. #1556 (1.8.1)
  * Environment variable `OPENIMAGEIO_THREADS` can artificially raise or
    lower the default number of threads (otherwise, it defaults to the
    number of processor cores available). #1581 (1.8.1)
* timer.h: added timed_thread_widge() that benchmark code and prints
  statistics about thread scaling. #1660 (1.8.4)
* typedesc.h:
  * Modernized TypeDesc with C++11 constexpr where applicable. #1684 (1.8.4)
  * New `TypeDesc::basevalues()` method. #1688 (1.8.4)
* unittest.h:
  * Colored error messages and auto error return on completion. #1731 (1.8.5)
* *NEW* parallel.h:
  * parallel_for, parallel_for_chunked, parallel_for_each offer simple
    thread_pool-based parallel looping in 1 and 2 dimensions.
    #1556, #1581 (1.8.1)
* Internal `OIIO::debugmsg()` call has been renamed to `OIIO::debug()`,
  and setting env variable OPENIMAGEIO_DEBUG_FILE can redirect the debug
  messages to a file. #1580 (1.8.1)
* Upgraded tinyformat to the latest master, also changed all the places
  where we used TINYFORMAT_WRAP_FORMAT to use of C++11 variadic templates.
  #1618 (1.8.3)
* *NEW* function_view.h : function_view<> is a very lightweight, non-owning,
  generic callable object view. Cheaper than std::function, but the view
  is not allowed to outlive the callable object it references. #1660 (1.8.4)
* Deprecate the pre-C++11 macros OIIO_CONSTEXPR, OIIO_CONSTEXPR_OR_CONST,
  and OIIO_NOEXCEPT. #1678 (1.8.4)

Docs:
* Improve docs about deep IBA functions. (1.8.1)
* Fix 'Building OIIO on Windows' link. #1590 (1.8.1)

Release 1.7.19 (1 Jun 2018) -- compared to 1.7.18
-------------------------------------------------
* Build fix for 32 bit Windows -- disable SSE that was making trouble. #1933
* Improved UTF-8 filename support for OpenEXR filenames. #1941
* filesystem.h: Improve UTF-8 handling for unique_path and temp_directory
  functions. #1940
* Clean up some warnings on modern Clang versions.
* Improvements in finding the location of OpenJPEG with Macports. #1948
* ImageCache and TextureSystem now have close(filename) and close_all()
  methods, which for one file or all files will close the files and release
  any open file handles (also unlocking write access to those files on
  Windows), but without invalidating anything it knows about the ImageSpec
  or any pixel tiles already read from the files, as would happen with a
  call to the much more drastic invalidate() or invalidate_all(). #1950

Release 1.7.18 (1 May 2018) -- compared to 1.7.17
-------------------------------------------------
* Update to build against ffmpeg 4.0

Release 1.7.17 (1 Sep 2017) -- compared to 1.7.16
-------------------------------------------------
* Repair build breaks against Boost 1.65. #1753
* Fix a subtle static initialization order problem. #1757
* Build: Improved finding LibRaw. #1749

Release 1.7.16 (1 Aug 2017) -- compared to 1.7.15
-------------------------------------------------
* OpenEXR: fix problem with 2-channel images putting the channels in the
  wrong order. #1717
* TIFF: images with fewer than 4 channels, but one of those channels was
  alpha, were not correctly marking their spec.alpha_channel. #1718
* Several minor updates to simd.h backported from mater.

Release 1.7.15 (1 Jun 2017) -- compared to 1.7.14
-------------------------------------------------
* Add "raw:user_sat" configuration attribute to the reader. #1666
* Better exception safety for `Filesystem::searchpath_find()`. #1680
* Improved I/O of color separation images, particularly those with custom
  InkSet attributes. #1658
* Big TextureSystem performance improvement on Windows with MSVC by removing
  certain empty destructors in simd.h that prevented MSVC from fully
  inlining the class. #1685
* Fix rare case TextureSystem crash. #1685

Release 1.7.14 (1 May 2017) -- compared to 1.7.13
-------------------------------------------------
* oiiotool expression substitution now recognizes FRAME_NUMBER and
  FRAME_NUMBER_PAD. #1648
* oiiotool -resize and -resample now have more intuitive behavior for images
  where the display and pixel data windows are not the same, especially if
  the data window had a nonzero origin (such as with crop or overscan).
  #1667
* oiiotool --resample and ImageBufAlgo::resample() have been extended to
  work for "deep" images. #1668
* ImageCache::get_image_info() will now return a proper error (and not hit
  an assertion) if asked for information about a subimage or MIP level that
  does not exist in the file. #1672
* ImageBuf copy constructor from another ImageBuf was previously broken for
  IB's that wrap application buffers. #1665
* TextureSystem::get_texels fixes crashing behavior. #1669
* Fixes to OSX rpath behavior of linked binaries. #1671
* OpenEXR file input: fixed problem with sorting order of spectral alpha
  channels (RA, GA, BA, or AR, AG, AB). #1674
* ImageBufAlgo::deep_merge() (and oiiotool --deepmerge) now will give a
  useful error message if you try to merge two deep images that do not have
  the same number of channels. #1675
* ImageCache/TextureSystem statistics no longer list as "BROKEN" files which
  were invalidated, or files that were initialized with an ImageCacheFile
  but never opened. #1655
* Fix Windows warnings about SIMD types as function args, and about
  int vs bool. #1659

Release 1.7.13 (1 Apr 2017) -- compared to 1.7.12
----------------------------------------------
* TIFF, JPEG, others: Improved reading Exif meatdata from XMP blocks, now it
  does a better job of discerning the proper data types. #1627
* RAW: The default value for missing "raw:use_camera_matrix" has been changed
  to 1 (not 0) to better match default behavior of libraw/dcraw. #1629
* RAW: Add support for selecting new demosaicing algorithms: "DMT" (mode 11)
  and "AAHD" (mode 12). Renamed the "Modified AHD" to "AHD-Mod" to
  simplify and match libraw terminology. Made matching of demosaicing
  algorithms case-insensitive. #1629
* RAW: Support "ACES" color space for direct conversion while reading RAW
  images (supported in libraw 0.18 or newer). #1626
* oiiotool: Improved error reporting of file open errors when -iconfig is
  used. #1626
* oiiotool `--echo STRING` prints the string to the console. This can contain
  expressions! So you can do something like
      oiiotool file.exr -echo "Size is {TOP.width}x{TOP.height}"
  #1633
* JPEG: Be more reslient to malformed Exif data blocks with bogus offsets.
  #1585, #1639
* TIFF output omitted setting the "Make" and "Model" metadata tags. #1642
* webp: Several new sanity checks prevent the webp reader from spending too
  much I/O time and memory reading bogus files (malformed, corrupted,
  or not a webp after all). #1640
* PSD: Support has been added for "cmyk", "multichannel", and "grayscale"
  color modes. And support was fixed for rgb and grayscale 32 bit per
  sample bit depth. #1641

Release 1.7.12 (1 Mar 2017) -- compared to 1.7.11
----------------------------------------------
* BMP: add support for version 5 headers. #1616
* TIFF: Fix typo that prevented correct reading of some Exif fields. #1625
* ImageBuf: Fix broken threads(n) method, which didn't correctly pass the
  right number of threads along. #1622.
* Fix build warnings about undefined OIIO_MSVS_AT_LEAST_2013 symbol.

Release 1.7.11 (1 Feb 2017) -- compared to 1.7.10
----------------------------------------------
* maketx fix: two textures that had identical source pixels but differed
  in whether they resized with "highlight compensation" incorrectly
  ended up with identical hashes, and thus could be confused by the
  TextureSystem at runtime into thinking they were duplicates. The hash
  is now fixed.  (1599)
* OpenEXR: Allow compression "none" for deep exr files.
* Fix unimplemented python ImageBufAlgo.computePixelStats. (1596)
* IBA::draw_rectangle (and oiiotool --box) wasn't drawing properly for
  the degenerate case of a rectangle that takes up just one
  pixel. (1601)
* Fixed several (so far unnoticed) buffer overruns and memory leaks. (1591)
* ImageInput::read_tiles when only a channel subset is read, fixed case
  with certain data sizes where the copy to user buffer got mangled. (1595)
* oiiotool -iconfig fixed, did not in all cases correctly propagate the
  input config attribute to the file read. (1605)
* TIFF: now has a way to read raw pixel values from CMYK files, without
  the automatic conversion to RGB (pass configuration attribute
  "oiio:RawColor" set to nonzero). (1605)
* imageio.h: Fix incorrect declaration of declare_imageio_format(). (1609)
* `oiiotool --crop` did not properly honor the `-a` flag and apply the crop
  to all subimages. (1613)

Release 1.7.10 (1 Jan 2017) -- compared to 1.7.9
----------------------------------------------
* Fix "exr_threads" value, -1 should disable IlmImf's thread pools. #1582
* Be more reslient to malformed Exif data blocks with bogus offsets. #1585
* Build: Fix regarding iv man page.
* Build: Fix compiler warning on Fedora aarch64. #1592
* Docs: improve docs about deep IBA functions.
* Docs: fix 'Building OIIO on Windows' link. #1590

Release 1.7.9 (1 Dec 2016) -- compared to 1.7.8
----------------------------------------------
* Make sure that sRGB<->linear color transform still work (in the obvious
  way) even when OpenColorIO is present but that its configuration for some
  reason doesn't know about "sRGB" space. #1554
* ImageCache: make robust to changes in autotile after opening and reading
  from untiled files. #1566
* ImageCache: fix initialization bug that made the reported stats output
  nonsensical in the numbers it gave for "redundant reads". #1567
* IC/TS get_image_info queries of "displaywindow" and "datawindow" did not
  correctly return a 'true' value when the data was found. #1574
* oiiotool -d giving per-channel formats is no longer confused by channel
  renaming with --chnames. #1563
* Added implementation of ImageBufAlgo::to_IplImage(). #1461

Release 1.7.8 (1 Nov 2016) -- compared to 1.7.7
----------------------------------------------
* Fix gcc warnings when compiling for AVX2. #1511
* Fix a variety of Windows warnings and breaks. #1512, #1523
* Added support for new API introduced with ffmpeg 3.1. #1515
* Improve oiiotool --mosaic, wasn't reliably clearing the blank spaces
  for missing images.
* Smarter channel_append channel renaming: try to resolve redundant
  channel names by using the subimage name, if available. #1498
* Texture: get_image_info queries for "channels" on UDIM file patterns
  now succeeds, returning the value for the first matching file it finds.
  (Relies on all textures within the same UDIM set having the same
  nchannels.) #1502, #1519
* Bug fix to possible crashes when adding dither to tiled file output
  (buffer size miscalculation). #1518
* maketx: multiple simultaneous maketx process trying to create the same
  texture will no longer clobber each other's output. #1525
* Build no longer gets confused about include files from older installations
  of OIIO elsewhere on the system. #1524
* Improvements in finding OpenJPEG. #1520
* Sysutil::Term formatting now works properly in Windows (though is only
  functional for Windows 10 or above). #1527
* Fix RLA reading and writing with certain channel orders and mixded data
  formats. #1499
* Improved finding of OCIO headers. #1528
* Better recognition of C++11 features in MSVS.
* Fix compile warnings with Clang 3.9. #1529
* Texture: Fix UDIM channels query. #1530
* nuke: Fix txReader to properly restore saved mip level knob value (#1531)
* Fix warnings on some 32 bit platforms #1539
* Exit oiiotool with non-zero status when command-line args fail to
  parse properly. #1540
* Fix typo in fmath bitcast_to_float declaration #1543
* Allow multiple key/value pairs in OCIO wrappers. #1542
* colorconvert API extended to take context key/values #1542
* Fix to TIFF handling of certain unusual tags, which also affected raw
  files. #1547
* fmath.h: templatize round_to_multiple so it works with other types
  (like size_t). #1548
* Fix IFF output that didn't correctly save the "Author" and "Date"
  metadata. #1549
* Handle 'oiiotool --colorconvert X X' (transform from and to spaces that
  are the same) without considering it an error. #1550

Release 1.7 (1 Oct 2016) -- compared to 1.6.x
----------------------------------------------
Major new features and improvements:
* New oiiotool commands:
    * `-otex` and `-oenv` allow oiiotool to directly output proper texture
      maps, and it can now do everything that maketx can do. #1351 (1.7.2)
    * `--line` can draw polylines into an image. #1319 (1.7.1)
    * `--box` can draw a filled or unfilled box into an image. #1319 (1.7.1)
    * `--laplacian` computes the Laplacian. #1332 (1.7.1)
    * `--deep_merge` does a full merge/composite of deep images. #1388 (1.7.2)
    * `-i` inputs a file. Options `autocc=`, `now=`, `info=` control aspects
      of reading that one file. #1389 (1.7.2)
    * `--dilate` and `--erode` perform the basic morphological operations
      of dilation and erosion. #1486 (1.7.5)
* New ImageBufAlgo functions: render_point(), render_line(), render_box()
  #1319 (1.7.1); laplacian() #1332 (1.7.2); copy() #1388 (1.7.2);
  deep_merge() #1388,1393 (1.7.2); dilate() and erode() (1.7.5).
* UDIM support for textures: filenames like `"tex_<UDIM>.exr"` will
  automatically be resolved to the correct UTIM tile based on the s,t
  coordinates. #1426 (1.7.3)
* Behavior change: When reading files without explicit channel names,
  single channel images now name their channel "Y" (no longer "A",
  which was confusing to algorithms that treat alpha in special ways).
  Similarly, 2-channel images name their channels "R" and "G". #1434 (1.7.3)

Public API changes:
* DeepData internals and API overhaul: struct internals hidden, now you
  must use the API; DeepData declaration is now in deepdata.h, not in
  imageio.h; DD methods now allow insertion and erasure of individual
  samples. #1289 (1.7.0) New DeepData methods: split, sort, merge_overlaps,
  merge_deep_pixels, occlusion_cull. #1388,1393 (1.7.2)
* imageio.h: Removed items deprecated since 1.4: a version of convert_types()
  that took alpha and z channel indices (but never used them). #1291
* fmath.h: Removed safe_sqrtf, safe_acosf, fast_expf, which have been
  deprecated since 1.5. (1.7.0) #1291
* Removed ImageBufAlgo::flipflop(), which was deprecated since 1.5 and
  is now called rotate180. #1291 (1.7.0)
* Several varieties of ImageCache and TextureSystem getattribute methods
  were noticed to not be properly declared 'const'. This was fixed.
  #1300 (1.7.0/1.6.9)
* For API calls that are deprecated but not yet removed, we now mark
  them with deprecated attributes for compilers that support it,
  meaning that you will get compile warnings and explanations when you
  use deprecated OIIO API functions. #1313,#1318 (1.7.1)
* ImageBuf::contains_roi() reveals whether an ROI is completely contained
  in the data region of the IB. #1310 (1.7.1)
* TypeDesc::is_signed() return true of the TypeDesc returns a type that
  can represent negative values. #1320
* Python: improve: reading with request for type UNKNOWN returns native
  data in an unsigned char array. Also, requesting HALF returns the half
  bits in an unsigned short array (since there is no 'half' type in Python).
  #1362 (1.7.2/1.6.11)
* Deprecate ImageCache/TextureSystem clear() methods, which never did
  anything useful. #1347 (1.7.3)
* ImageSpec::set_format() clears any per-channel format information. #1446
  (1.7.4)
* OIIO::getattribute("library_list", ...) can retrieve a list of all the
  dependent libraries used when OIIO was built. #1458 (1.7.5)
* simd::mask4 class has been renamed simd::bool4 (though the old name
  is still typedef'ed to work for now). #1484 (1.7.5)
* Python bindings for ImageBuf.reset() now properly understands the
  argument names and default values. #1492 (1.7.6)

Fixes, minor enhancements, and performance improvements:
* oiiotool:
    * oiiotool --subimage now takes as an argument either the subimage
      numeric index, or a subimage name. #1287 (1.7.0)
    * oiiotool's image cache was smaller than intended because of typo.
      (1.7.0/1.6.9)
    * Allow command-line expression metadata names to contain ':'. #1321
      (1.7.1/1.6.10)
    * --info more clearly prints info about subimage formats. #1320 (1.7.1)
    * --ch: when the channels were only renamed, not reordered, the
      renaming didn't happen properly. #1326 (1.7.1/1.6.10)
    * Improved error message propagation from the underlying IBA functions
      and from errors encounered during --stats. #1338 (1.7.2)
    * --dumpdata:empty=0 now does something for non-deep files: skips
      reporting of pixels where all channels are black. Also fixed errors
      for dumpdata of deep, but non-float, files. #1355 (1.7.2/1.6.11)
    * '--attrib:type=t name value' lets you explicitly name the top of
      the attribute you're seting. This helps for ambiguous cases, and also
      lets you create aggregate types (such as 'matrix' or 'int[2]' --
      the value can be a comma-separated list in those cases). #1351 (1.7.2)
    * --fixnan can now take option "error", meaning that upon finding a
      NaN, the program considers it an error (rather than fixing).
      #1351 (1.7.2)
    * -o now takes optional arguments that control the output of just that
      one file, including :datatype=, :bits=, :dither=, :autocc=,
      :autocrop=, :autotrim=, :separate=, :contig=.  #1351 (1.7.2)
    * --resize and --fit sped up by approximately 2x. #1372 (1.7.2)
    * --runstats not tracks and reports max memory usage. #1385 (1.7.2)
    * New --cache and --autotile options let you set the ImageCache size and
      autotile options. #1385 (1.7.2)
    * --native results in slightly different behavior: does not force float,
      but still uses the ImageCache as backing if the native data type
      is one of the types directly supported by IC. #1385 (1.7.2)
    * --fixnan now works with deep images. #1397 (1.7.3)
    * --stats when used with deep images, if it encounters any nonfinite
      values, will print the location of a representative one. #1397 (1.7.3)
    * --add, --sub, and --absdiff, when its two image operands have differing
      number of channels, now try to do the "correct" thing and have the
      result have the larger number of channels, treating any "missing"
      channels in either input as if they were 0-valued. #1402 (1.7.3)
    * --attrib:type-timecode lets you set timecodes in "human-readable" form:
        oiiotool in.exr -attrib:type=timecode TimeCode 11:34:03:00 -o out.exr
      #1415 (1.7.3)
    * -info -v will print timecodes in human-readable form. #1415 (1.7.3)
    * Bug fix: --fullsize didn't work properly when followed by --attrib.
      #1418 (1.7.3/1.6.14)
    * Easily understandable warnings for common mistakes: no output specified,
      an image that's modified but never output, or when -autocc causes
      output to be a data format that overrides a previous -d directive.
      #1419 (1.7.3)
    * --origin, --croptofull, and --trim all do their thing to all subimages,
      if the image at the top of the stack has multiple subimages.
      #1440 (1.7.3)
    * --crop operates on all subimages if either the -a flag was used, or
      if the specific optinoal override `--crop:allsubimages=1` is used.
      #1440 (1.7.3)
    * --trim will trim all subimages to the same region, containing the
      union of all nonzero pixels in all subimages. #1440 (1.7.3)
    * --help now prints all of the dependent libraries for individual
      formats. #1458 (1.7.5)
    * Bug fix: make sure to propagate per-channel formats from input
      to output. #1491 (1.7.6)
    * -o:all=n will output all images currently on the stack, and the
      filename argument will be assumed to be a pattern containing a %d,
      which will be substituted with the index of the image (beginning with
      n). For example, to take a multi-image TIFF and extract all the
      subimages separately,
          oiiotool multi.tif -sisplit -o:all=1 sub%04d.tif
      will output the subimges as sub0001.tif, sub0002.tif, and so on.
      #1494 (1.7.6)
    * --mosaic MxN would fail mysteriously if the number of images on the
      stack was less then M*N. This has been fixed to handle too-few images
      gracefully and just leave blank spaces as needed. #1501 (1.7.7)
* ImageBuf:
    * ImageBuf::iterator performance is improved -- roughly cutting in half
      the overhead of iterating over pixels. #1308 (1.7.1/1.6.10)
    * ImageBuf reads from disk have been improved substantially (in some
      cases cutting read time in half) in many cases. #1328 (1.7.1)
    * ImageBuf::copy_pixels() has been sped up. #1358 (1.7.2)
* ImageBufAlgo:
    * The varieties of add(), sub(), mul(), and div() that take an
      image operand and a per-channel constant operand have all been
      modified to work properly for "deep" images. #1297 (1.7.0/1.6.10)
    * mad() is sped up significantly (10x) for the common case of float
      images already stored in memory (not cached). #1310 (1.6.1)
    * render_point(), render_line(), render_box() can be used to render
      points, lines, and boxes into an image. #1319 (1.7.1)
    * channels(): when the channels were only renamed, not reordered,
      the renaming didn't happen properly. #1326 (1.7.1/1.6.10)
    * computePixelStats: Improved numerical accuracy to avoid getting
      NaN values from imprecision. #1333 (1.7.2/1.6.11)
    * laplacian() computes the laplacian of an image. #1332 (1.7.2)
    * fixNonFinite() takes a new option: NONFINITE_ERROR, which will
      return an error if nonfinite values are encountered. #1351 (1.7.2)
    * convolve() and unsharp_mask() have been sped up by about 35% for
      common cases. #1357 (1.7.2)
    * IBA::resize sped up by approximately 2x. #1372 (1.7.2)
    * IBA::fixNonFinite() now works with deep images. #1397 (1.7.3)
    * dilate() and erode() perform basic morphological operations.
      #1486 (1.7.5)
* ImageCache / TextureSystem:
    * Less unnecessary pausing after read errors when falure_retries == 0.
      #1336 (1.7.2/1.6.11)
    * Texture: slight improvement in texture sharpness. #1369 (1.7.2/1.6.11)
    * New statistics related to redundant tile reads. #1417 (1.7.3)
    * TextureSystem option "flip_t", if nonzero, will flip the vertical
      direction of all texture lookups. Use this for renderers that adhere
      to the convention that the t=0 texture coordinate is the visible
      "bottom" of the texture. #1428 (1.7.3) #1462 (1.7.5)
    * UDIM support for textures: filenames like `"tex_<UDIM>.exr"` will
      automatically be resolved to the correct UTIM tile based on the
      s,t coordinates. #1426 (1.7.3)
    * Avoid repeated broken texture error messages. #1423 (1.7.3)
    * New IC/TS attribute: "max_errors_per_file" limits how many error
      messages are printed for each file. #1423 (1.7.3)
    * Improved statistics: for the "top 3" stats, print the top 3 that aren't
      broken. Also print a count & list of broken/invalid files. #1433
      (1.7.3/1.6.15)
    * Add ability to retrieve various per-file statistics. #1438 (1.7.3/1.6.15)
    * IC will clamp the max_open_files to the maximum allowed by the
      system, so you can no longer crash a program by incorrectly
      setting this limit too high. #1457 (1.7.5)
    * IC/TS statistics now report separately the total size of images
      referenced, in terms of in-cache data size, as well as on-disk
      size (the latter may be compressed). #1481 (1.7.5)
* maketx:
    * maketx -u now remakes the file if command line arguments or OIIO
      version changes, even if the files' dates appear to match.
      #1281 (1.7.0)
    * Remove long-obsolete and non-functional command line options: --new,
      --old, --hash. #1351 (1.7.2)
* iinfo:
    * More clearly prints info about subimage formats. #1320 (1.7.1)
    * Print timecodes in human-readable form. #1415 (1.7.3)
* ImageOutput: fix cases with native data but non-contiguous strides.
  #1416 (1.7.3/1.6.15)
* Cineon:
    * Improved deduction/setting of color space info. #1466 (1.7.5)
* GIF:
    * GIF reader failed to set spec full_width, full_height. #1348
      (1.7.2/1.6.11)
* JPEG:
    * Fix bad memory access crash when reading specific JPEG files that were
      written without their comment field including a null character to
      end the string. #1365 (1.7.2/1.6.11)
    * Change in behavior writing JPEG files when XResolution & YResolution
      are not provided, but a PixelAspectRatio is requested. Previously, we
      obeyed the JPEG/JFIF spec exactly, but it turns out that popular apps
      including PhotoShop and Nuke use the field differently than the spec
      dictates. So now we conform to how these apps work, rather than to
      the letter of the spec. #1412 (1.7.3/1.6.15)
* OpenEXR:
    * Fix broken multipart output when parts had different pixel data
      types. #1306,#1316 (1.7.1/1.6.10)
    * Improved error reporting for bad tile reads. #1338 (1.7.2/1.6.11)
    * Fix errors reading tiles for mixed-format EXR files. #1352 (1.7.2/1.6.11)
    * The global OIIO::attribute("exr_threads") has been modified so that 0
      means to use full available hardware, -1 disables the OpenEXR
      thread pool and execute in the caller thread. #1381 (1.7.2)
    * When writing EXR, double check that there are no repeated channel
      names, and if so, rename them so the data is not lost (since the
      underlying libIlmImf will silently drop channels with repeated
      names).  #1435 (1.7.3)
    * More robust detected of when OpenEXR is tiled (for weird files).
      #1441 (1.7.3/1.6.15) (and a fix in #1448/1.7.4)
    * Fixed minor bug with OpenEXR output with correctly setting
      PixelAspectRatio based on the "XResolution" and "YResolution"
      attributes. #1453 (Fixes #1214) (1.7.4/1.6.16)
    * Fix setting "chromaticity" metadata in EXR files. #1487 (1.7.5)
    * When writing OpenEXR, accept compression requests with quality numbers
      appended to the compression algorithm name, such as "dwaa:200" to mean
      dwaa compression with a dwaCompressionLevel set to 200. #1493 (1.7.6)
* PNG:
    * Per the PNG spec, name 2-channel images Y,A. #1435 (1.7.3)
    * Enforce that alpha premultiplication on output MUST consider alpha
      to be the last channel of 2 or 4 channel images, no other cases
      (as dictated by the PNG spec). #1435 (1.7.3)
* PNM:
    * Fixed byte swapping when reading 16 but PNM files. #1352 (1.7.2/1.6.11)
* RAW:
    * Changes to how we instruct libraw to process images when reading:
      Now, by default, auto-bright adjustment is off, camera white
      balance is on, and maximum threshoding is set to 0. There are
      "open with config" overrides for all of these, for anybody who
      doesn't like the default. #1490 (1.7.6)
* RLA:
    * Fixes for both reading and writing of RLA images that are cropped
      (i.e., data window is a subset of display window). #1224 (1.7.0/1.6.10)
* TIFF:
    * When outputting a TIFF file, a special attribute "tiff:half", if
      set to nonzero, will enable writing of 16-bit float pixel data
      (obviously, only if the spec.format is HALF). #1283 (1.7.0)
    * TIFF input: erase redundant IPTC:Caption and IPTC:OriginatingProgram
      if they are identical to existing ImageDescription and Software
      metadata, respectively. (1.7.0/1.6.9)
    * Output: "tiff:zipquality" attribute controls time-vs-quality for
      ZIP compression (1-9, defualt 6, higher means more compression).
      #1295 (1.7.1)
    * Fix typo that made TIFF files incorrectly name color space metadata
      "oiio::ColorSpace" instead of "oiio:ColorSpace". #1394 (1.7.2)
    * More robust handling of non-zero origin of full/display window.
      #1414 (1.6.14/1.7.3)
* Video formats:
    * The ffmpeg-based reader had a variety of fixes. #1288 (1.7.0)
    * Support for reading 10-bit and 12-bit movies. #1430 (1.7.5)
* Improved accuracy of "lanczos3" filter; speed up blackman-harris filter.
  #1379 (1.7.2)
* Speed up linear<->sRGB color conversions (as used by any of the IBA color
  conversion functions as well as oiiotool --colorconvert and friends),
  approximately doubling the speed when no OpenColorIO config is found.
  #1383 (1.7.2)
* ImageInput::create() and ImageOutput::create() will now gracefully
  handle unexpected exceptions inside an ImageInput or ImageOutput
  constructor -- return an error rather than crashing.  #1456 (1.7.4/1.6.16)
* Nuke txWriter adds UI to let you choose which type of texture you are
  building (ordinary 2D texture, latlong env map, etc). #1488 (1.7.6)

Build/test system improvements:
* Default build is now C++11! #1344 (1.7.2) You can still (for now) build
  for C++03 using 'make USE_CPP11=0' or 'cmake -DOIIO_BUID_CPP11=0', but
  some time soon we will be C++11 minimum.
* Fix build break against Boost 1.60. #1299,#1300 (1.7.0/1.6.9/1.5.23)
* filesystem_test now much more comprehensively tests the contents of
  Filesystem. #1302 (1.7.0)
* fmath_test adds benchmarks for various data conversions. #1305 (1.7.0)
* Travis: add DEBUG builds to the matrix to fix any warnings or failures
  that only show up for DEBUG builds. #1309 (1.7.1/1.6.10)
* Fix build issues on some platforms for SHA1.h, by adding proper include
  of `<climits>`. #1298,#1311,#1312 (1.7.1/1.6.10)
* Cleanup of include logic in simd.h that fixed build problems for gcc < 4.4.
  #1314 (1.7.1/1.6.10)
* Fix build breaks for certain 32 bit platforms. #1315,#1322 (1.7.1/1.6.10)
* imagespeed_test can not specify the data conversion type for reads,
  can optionally allow skipping the IB iteration tests, and can set the
  IC tile cache size. #1323 (1.7.1)
* Fix build breaks for gcc 6. #1339 (1.7.2/1.6.11) #1436 (1.7.3/1.6.15)
* Fix errors in finding the correct locaiton of pugixml.hpp when using
  USE_EXTERNAL_PUGIXML=1. #1339 (1.7.2/1.6.11)
* Rewrite of FindOpenEXR.cmake. Solves many problems and is simpler.
  No more FindIlmbase.cmake at all. #1346 (1.7.2/1.6.11)
* 'make CODECOV=1; make CODECOV=1 test' can build and test in a way that
  provides a code coverage report. #1356 (1.7.2)
* Fix Filesystem::open() issues with UTF-8 filenames on MinGW.
  #1353,#1357 (1.7.2/1.6.11)
* Allow build against a wider range of ffmpeg versions. #1359 (1.7.2)
* Build correctly against FFMPEG 3.0. #1374 (1.7.2)
* If found, libjpeg-turbo is used rather than libjpeg; this gives about a
  2x speed improvement for reading & writing JPEG files. #1390 (1.7.2/1.6.13)
* USE_CPP11=... and USE_CPP14=... are the build flags for both the Make
  wrapper and the CMake scripts. (Before, it was confusing to have USE_CPP11
  for make but OIIO_BUILD_CPP11 for CMake. Now they are one.) #1391 (1.7.2)
* Improved LINKSTATIC=1, now mostly works. #1395 (1.7.3)
* Big CMake refactor, got rid of redundancies. #1395 (1.7.3)
* Remove old embedded Ptex, now must find Ptex externally. Also modified
  the build scripts to correctly handle newer versions of Ptex. #1400
  (1.7.3/1.6.13)
* Got Appveyor building OIIO. This is a continuous integration service
  much like Travis, but it's for Windows. Hopefully this means that it will
  be much harder for us to make changes that inadvertently break the build
  on Windows. #1399 (1.7.3)
* Make FindOpenEXR.cmake more robust when the version number is embedded
  in the library name. #1401 (1.7.3/1.6.15)
* Clear up some inconsistencies in the CMake files and the Makefile wrapper:
  the flag to compile with libc++ is now always called USE_LIBCPLUSPLUS,
  not sometimes OIIO_BUILD_LIBCPLUSPLUS. #1404 (1.7.3)
* Overhaul OpenCV dependency finding and make it work with OpenCV 3.x.
  #1409 (1.7.3/1.6.13)
* Allow custom JPEG_PATH to hint location of JPEG library. #1411
  (1.7.3/1.6.13)
* Windows UTF-8 filename safety fixes. #1420 (1.7.3/1.6.14)
* Various Windows compilation & warning fixes. #1443 (1.7.3/1.6.15)
* Now builds correctly against OpenJPEG 2.x, it previously only supported
  OpenJPEG 1.x. #1452  (Fixes #957, #1449) (1.7.4/1.6.16)
* Fix Filesystem::searchpath_find on Windows with UTF-8 paths.
  #1469 (1.7.51.6.17)
* Improved the way OpenEXR installations are found. #1464 (1.7.5)

Developer goodies / internals:
* thread.h has had all the atomic operations split into a separate atomic.h.
  #1443 (1.7.3)
* atomic.h: add atomic and, or, and xor. #1417 (1.7.2/1.6.14);
* parallel_image has been improved in several ways: can choose split
  direction; raised minimum chunk size to prevent thread fan-out for
  images too small to benefit; uses the calling thread as part of the
  pool. #1303 (1.7.0)
* timer.h: DoNotOptimize() and clobber_all_memory() help to disable certain
  optimizations that would interfere with micro-benchmarks. #1305 (1.7.0)
* simd.h improvements: select(); round(); float4::store(half*),
  int4::store(unsigned short*), int4::store(unsigned char*). #1305 (1.7.0)
  Define insert, extract, and ^ (xor), and ~ (bit complement) for mask4,
  and add ~ for int4. #1331 (1.7.2); madd, msub, nmadd, nmsub, rint,
  andnot #1377 (1.7.2); exp, log #1384 (1.7.2); simd::float3 is like float4,
  but only loads and stores 3 components, it's a good Vec3f replacement (but
  padded) #1473 (1.7.5); matrix44 4x4 matrix class #1473 (1.7.5);
  mask4 renamed to bool4, and addition of float8, int8, bool8 classes
  for 8-wide AVX/AVX2 SIMD #1484 (1.7.5).
* fmath.h: convert_types has new special cases that vastly speed up
  float <-> uint16, uint8, and half buffer conversions #1305 (1.7.0);
  ifloor (1.7.2); SIMD versions of fast_log2, fast_log, fast_exp2,
  fast_exp, fast_pow_pos #1384 (1.7.2); fix sign of expm1 for small
  arguments #1482 (1.7.5); added fast_log1p #1483 (1.75).
* Fix pesky precision discrepancy in internal convert_type<> that used
  slightly different math when converting one value at a time, versus
  converting whole arrays. #1350 (1.7.2)
* thread.h: add mutex_pool #1425 (1.7.3/1.6.15)
* compute_test: new unit test can be used to benchmark computation
  times. #1310 (1.7.1)
* filesystem.h: Filesystem::file_size() returns file size in bytes;
  Filesystem::read_bytes() reads the first n (or all) bytes from a file
  into a buffer. #1451 (1.7.4/1.6.16)
* strutil.h: Strutil::extract_from_list_string is more flixible by
  allowing the vals list to start empty, in which case it will add as
  many values as it finds rather than only replacing existing
  values #1319 (1.7.1); Strutil::replace #1422 (1.7.3/1.6.15);
  utf_to_unicode now takes a string_view rather than a std::string&
  #1450 (1.7.4); add Strutil::base64_encode() #1450 (1.7.4).
* sysutil.h: Sysutil::getenv() safely gets an env variable as a string_view
  #1451 (1.7.4/1.6.16); terminal_columns() now has a correct implementation
  on Windows #1460 (1.7.5); max_open_files() retrieves the maximum number
  of files the process may open simultaneously #1457 (1.7.5).
* platform.h: better distinguishing beteen Apple and Generic clang,
  separately set OIIO_CLANG_VERSION and OIIO_APPLE_CLANG_VERSION. Also change
  OIIO_GNUC_VERSION to 0 for clang, only nonzero for true gcc. #1380 (1.7.2)
* ImageCache: remove unused shadow matrix fields, save space. #1424 (1.7.3)
* Many documentation files (such as README, CHANGES, LICENSE, CREDITS,
  and INSTALL) have been changed from plain text to MarkDown. #1442 (1.7.3)
* Sysutil::Term class makes it easy to use color output on the terminal.
  #1479 (1.7.5)

Release 1.6.18 (released 1 Nov 2016 -- compared to 1.6.17)
------------------------------------------------
* Fix setting "chromaticity" metadata in EXR files. #1487
* maketx: multiple simultaneous maketx process trying to create the same
  texture will no longer clobber each other's output. #1525
* Fix compile warnings with Clang 3.9. #1529
* Fix IFF output that didn't correctly save the "Author" and "Date"
  metadata. #1549

Release 1.6.17 (released 1 Sep 2016 -- compared to 1.6.16)
------------------------------------------------
* Fix build for newer ffmpeg release that deprecated functions.
* Improved finding of OCIO installations. #1467
* Fixed Sysutil::terminal_columns() for WIndows. #1460
* Fix build break in Windows when roundf function not found. #1468
* Fix Filesystem::searchpath_find on Windows with UTF-8 paths. #1469

Release 1.6.16 (released 1 Aug 2016 -- compared to 1.6.15)
------------------------------------------------
* Fix EXR tile logic for OpenEXR 1.x (fixes a break introduced in 1.6.15,
  is not an issue for exr 2.x). #1448
* Now builds correctly against OpenJPEG 2.x, it previously only supported
  OpenJPEG 1.x. #1452  (Fixes #957, #1449)
* New utility functions: Sysutil::getenv(), Filesystem::file_size(),
  FileSystem::read_bytes(). #1451
* Fixed minor bug with OpenEXR output with correctly setting
  PixelAspectRatio based on the "XResolution" and "YResolution"
  attributes. #1453 (Fixes #1214)
* Gracefully handle unexpected exceptions inside an ImageInput or
  ImageOutput constructor -- return an error rather than crashing.
  #1456

Release 1.6.15 (released 1 Jul 2016 -- compared to 1.6.14)
------------------------------------------------
* Improved statistics: for the "top 3" stats, print the top 3 that aren't
  broken. Also print a count & list of broken/invalid files. #1433
* Change in behavior writing JPEG files when XResolution & YResolution
  are not provided, but a PixelAspectRatio is requested. Previously, we
  obeyed the JPEG/JFIF spec exactly, but it turns out that popular apps
  including PhotoShop and Nuke use the field differently than the spec
  dictates. So now we conform to how these apps work, rather than to
  the letter of the spec. #1412
* IC/TS: add ability to retrieve various per-file statistics. #1438
* Windows UTF-8 filename safety fixes. #1420
* ImageOutput: fix cases with native data but non-contiguous strides.  #1416
* Make FindOpenEXR.cmake more robust when the version number is embedded
  in the library name. #1401
* Fix build breaks for gcc 6. #1339 (1.7.2/1.6.11) #1436
* More robust detected of when OpenEXR is tiled (for weird files).  #1441
* Various Windows compilation and warning fixes.
* strutil.h: added replace(). #1422
* thread.h: added mutex_pool. #1425

Release 1.6.14 (released 1 Jun 2016 -- compared to 1.6.13)
------------------------------------------------
* More robust handling of TIFF I/O when non-zero origin of full/display
  window. (#1414)
* oiiotool --fullsize didn't work properly when followed by --attrib. (#1418)

Release 1.6.13 (released 1 May 2016 -- compared to 1.6.12)
------------------------------------------------
* Use libjpeg-turbo if found. It's a drop-in replacement for libjpeg, but
  is around 2x faster. #1390
* Fix some Windows compiler warnings and errors.
* Remove old embedded Ptex, now must find Ptex externally. Also modified
  the build scripts to correctly handle newer versions of Ptex. #1400
* Overhaul OpenCV dependency finding and make it work with OpenCV 3.x. #1409
* Allow custom JPEG_PATH to hint location of JPEG library. #1411

Release 1.6.12 (released 1 Apr 2016 -- compared to 1.6.11)
------------------------------------------------
* Build correctly against FFMPEG 3.0. #1374
* The global OIIO::attribute("exr_threads") has been modified so that 0
  means to use full available hardware, -1 disables the OpenEXR thread
  pool and execute in the caller thread. #1381
* Thread-pool counts initialized to hardware_concurrency, not
  physical_concurrency (i.e., they will include hyperthread cores by
  default). #1378
* oiiotool --autocc bug fixed.
* Miscellaneous improvements to simd.h ported from master.
* Fix typo that made TIFF files incorrectly name color space metadata
  "oiio::ColorSpace" instead of "oiio:ColorSpace". #1394

Release 1.6.11 (released 1 Mar 2016 -- compared to 1.6.10)
------------------------------------------------
* Fix potential of IBA::computePixelStats (including oiiotool --stats)
  to end up with NaNs due to numerical imprecision. #1333
* Less unnecessary pausing after read errors when falure_retries == 0.
  #1336
* Fix errors in finding the correct locaiton of pugixml.hpp when using
  USE_EXTERNAL_PUGIXML=1. #1339
* Fix build breaks for gcc 6. #1339
* GIF reader failed to set spec full_width, full_height. #1348
* PNM: Fixed byte swapping when reading 16 but PNM files. #1352
* OpenEXR: Improved error reporting for bad tile reads. #1338
* OpenEXR: Fix errors reading tiles for mixed-format EXR files. #1352
* oiiotool --dumpdata:empty=0 now does something for non-deep files: skips
  reporting of pixels where all channels are black. Also fixed errors
  for dumpdata of deep, but non-float, files. #1355
* Fix Filesystem::open() issues with UTF-8 filenames on MinGW. #1353
* Rewrite of FindOpenEXR.cmake. Solves many problems and is simpler.
  No more FindIlmbase.cmake at all. #1346
* Fix build break for older gcc < 4.4 that didn't have immintrin.h.
* Fix bad memory access crash when reading specific JPEG files that were
  written without their comment field including a null character to
  end the string. #1365
* The ffmpeg-based reader had a variety of fixes. #1288
* Python: improve: reading with request for type UNKNOWN returns native
  data in an unsigned char array. Also, requesting HALF returns the half
  bits in an unsigned short array (since there is no 'half' type in Python).
  #1362
* Texture: slight improvement in texture sharpness. #1369
* Update webp testsuite references for new webp version.

Release 1.6.10 (released 1 Feb 2016 -- compared to 1.6.9)
------------------------------------------------
* ImageBufAlgo add, sub, mul, and div, for the varieties that combine
  an image with a (per-channel) constant, now work for "deep" images.
  #1257
* ImageBuf::iterator performance is improved -- roughly cutting in half
  the overhead of iterating over pixels. #1308
* OpenEXR: Fix broken multipart output when parts had different pixel
  data types. #1306,#1316
* Allow oiiotool command-line expression metadata names to contain ':'.
* Fix oiiotool --ch (or IBA::channels) when the channels were only renamed,
  not reordered, the renaming didn't happen properly. #1326
* Fixes for both reading and writing of RLA images that are cropped
  (i.e., data window is a subset of display window). #1224
* Fix build issues on some platforms for SHA1.h, by adding proper include
  of `<climits>`. #1298,#1311,#1312
* Cleanup of include logic in simd.h that fixed build problems for gcc < 4.4.
  #1314
* Fix build breaks for certain 32 bit platforms. #1315,#1322

Release 1.6.9 (released 5 Jan 2016 -- compared to 1.6.8)
------------------------------------------------
* Several varieties of ImageCache and TextureSystem getattribute methods
  were noticed to not be properly declared 'const'. This was fixed.
  #1300 (1.6.9)
* Fix build break against Boost 1.60. #1299,#1300 (1.6.9/1.5.23)
* The Python bindings for ImageCache was overhauled after several
  of the methods were found to be horribly broken. #1300 (1.6.9)
* oiiotool --subimage now allows a subimage name as argument, as well
  as the numeric index. #1271,#1287 (1.6.9)
* TIFF input: erase redundant IPTC:Caption and IPTC:OriginatingProgram
  if they are identical to existing ImageDescription and Software metadata,
  respectively. (1.6.9)
* Fix oiiotool image cache smaller than intended because of typo. (1.6.9)

Release 1.6 (1.6.8 released Dec 21, 2015 -- compared to 1.5.x)
----------------------------------------------
Major new features and improvements:
* New oiiotool functionality:
    * Expression evaluation/substitution on the oiiotool command line.
      Anything enclosed in braces { } in a command line argument will be
      substituted by the evaluation of the enclosed expression. Expressions
      may be numbers, simple arithmetic (like 'expr+expr'), or retrieving
      image metadata from named images or images on the stack.
      Please see the PDF documentation, Section 12.1 for details and
      examples.
    * --absdiff, --absdiffc compute the absolute difference (abs(A-B)) of
      two images, or between an image and a constant color. #1029 (1.6.0)
    * --abs computes the absolute value of an image.  #1029 (1.6.0)
    * --div, divc divide one image by another (pixel by pixel), or divides
      the pixels of an image by a constant color. #1029 (1.6.0)
    * --addc, --subc, --mulc, --powc are the new names for --cadd, --csub,
      --cmul, and --cpow. The old ones will continue to work but are
      considered depcrected. #1030 (1.6.0)
    * --pattern supports new patterns: "fill" makes a solid, vertical or
      horizontal gradient, or four-corner interpolated image (just like
      the --fill commmand) (1.6.0); "noise" can generate uniform, gaussian,
      or salt & pepper noise (1.6.2).
    * --fill, in addition to taking optional parameter color=... to give a
      solid color for the fill region, now also takes top=...:bottom=... to
      make a vertical gradient, left=...:right=... to make a horizontal
      gradient, and topleft=...:topright=...:bottomleft=...:bottomright=...
      to make a 4-corner gradient. (1.6.0)
    * --noise adds noise to the current image: additive uniform or gaussian
      noise, or making "salt & pepper" noise. (1.6.2)
    * --trim crops the image to the minimal rectangle containing all
      the non-0 pixels. (1.6.3)
    * --autocc : when turned on, automatic color conversion of input files
      into a scene_linear space, and conversion to an appropriate space
      and pixel type upon output. It infers the color spaces based on
      metadata and filenames (looking for OCIO-recognized color space names
      as substrings of the filenames). #1120 (1.6.3)
    * --mad takes three image arguments, multiplies the first two and then
      adds the third to it. #1125 (1.6.3)
    * --invert computes the color inverse (1-value) for color channels.
      #1125 (1.6.3)
    * --colorconfig allows you to specify a custom OCIO configuration file
      (rather than strictly relying on the $OCIO env variable). #1129 (1.6.3)
    * --deepen converts flat images to "deep". #1130 (1.6.3)
    * -n (no saved output) performs all calculations (including timing and
      stats) but does not write any output files to disk. #1134 (1.6.3)
    * --debug prints debugging information, this is now separate from
      -v which just makes more verbose (non-debugging) output. #1134 (1.6.3)
    * --pixelaspect rescales the image to have the given pixel aspect
      ratio. #1146 (1.6.5)
    * --ociofiletransform() implements OpenColorIO "file" transforms.
      #1213 (1.6.5)
* New ImageBufAlgo functions:
    * absdiff() computes the absolute difference (abs(A-B)) of two images,
      or between an image and a constant color. #1029 (1.6.0)
    * abs() computes the absolute value of an image. #1029 (1.6.0)
    * div() divides one image by another (pixel by pixel), or divides all
      the pixels of an image by a constant color. #1029 (1.6.0)
    * fill() has been extended with new varieties that take 2 colors (making
      a vertical gradient) and 4 colors (one for each ROI corner, for a
      bilinearly interpolated gradient). (1.6.0)
    * noise() injects noise into an image -- uniform, gaussian/normal,
      or salt & pepper noise. (1.6.2)
    * mad() multiplies the first two arguments and then adds the third to
      it. #1125 (1.6.3)
    * invert() computes 1-val. #1125 (1.6.3)
    * deepen() turns a flat RGBA (and optional Z) image into a "deep"
      image. #1130 (1.6.3)
    * ociofiletransform() implements OpenColorIO "file" transforms.
      #1213 (1.6.5)
* Some open source fonts are now distributed with OIIO (DroidSans,
  DroidSans-Bold, DroidSerif, DroidSerif-Bold, DroidSerif-Italic,
  DroidSerif-BoldItalic, and DroidSansMono), and so those are always
  available to ImageBufAlgo::render_text() and oiiotool --text, on all
  platforms and even if you don't have any other installed fonts on
  your system. DroidSans is now the default font. #1132 (1.6.3)
* GIF output support (including writing animated GIF images, just write it
  as a multi-subimage file). For example, this works:
      oiiotool foo*.jpg -siappendall -attrib FramesPerSecond 10.0 -o anim.gif
  #1193 (1.6.4)

Public API changes:
* TypeDesc:
    * New helper methods: is_array(), is_unsized_array(), is_sized_array().
      #1136 (1.6.3)
    * New constructor and fromstring of a string_view, in addition to
      the old versions that took char*. #1159 (1.6.4/1.5.16)
    * New aggregate type: MATRIX33. #1265,#1267 (1.6.6)
* ImageSpec:
    * ImageSpec::metadata_val() is now static, rather than simply const,
      since it doesn't need access to *this at all. #1063 (1.6.1)
    * Added a new variety of find_attribute that takes a temporary
      ImageIOParameter as scratch space. The advantage of this call is
      that it can retrieve items from the named ImageSpec fields, such
      as "width", "full_x", etc. Also, the get_int_attribute,
      get_float_attribute, and get_string_attribute can now retrieve
      these fixed fields as well.  #1063 (1.6.1)
* ImageInput & ImageOutput:
    * New ImageOutput::supports() tags: supports("alpha") should be true
      for image formats that support an alpha channel, supports("nchannels")
      should be true for output formats that support an arbitrary number
      of output channels. (1.6.2/1.5.13)
    * ImageInput and ImageOutput supports() method has been changed to accept
      a string_view (rather than a const std::string&), and return an int
      (rather than a bool). (1.6.2)
    * ImageInput and ImageOutput have added destroy() static
      methods. They are just wrappers around 'delete', but can help you
      to ensure that II and IO objects are deleted on the same side of a
      DLL boundary as where they were created. (Helps with using OIIO
      from DLL-based plugins on Windows.)  (1.6.3)
    * New ImageInput query: "procedural" -- returns 1 if the ImageInput may
      not correspond to an actual file. #1154 (1.6.4/1.5.16)
    * ImageInput and ImageOutput's error() method is changed from protected
      to public, making it easier for an app to set an error on a reader
      or writer. (1.6.4)
    * ImageOutput::copy_to_image_buffer is a helper function that
      generalizes the existing copy_tile_to_image_buffer, but for any
      rectangle.  #1193 (1.6.4)
    * ImageInput::read_image() variant that takes a channel range to
      read just a subset of the channels present. #1222 (1.6.5)
    * ImageInput and ImageOutput now have new method threads(n) that sets
      the thread "fan-out" for the ImageInput or ImageOutput individually,
      overriding any global attribute("threads"). #1259 (1.6.6)
* ImageBuf:
    * Add make_writeable(), which forces ImageCache-backed read-only
      ImageBuf to read into locally allocated pixels so they can be
      subsequently altered. #1087 (1.6.2)
    * ImageBuf::Iterator has added set_deep_samples() and set_deep_value()
      methods. (1.6.3)
    * ImageBuf::set_pixels() now provides a way to set an arbitrary
      rectancle of an ImageBuf from raw values. #1167 (1.6.4)
    * ImageBuf::get_pixels() now has a variety that takes an ROI to
      describe the rectangle of pixels being requested. #1167 (1.6.4)
    * ImageBuf now has new method threads(n) that sets the thread
      "fan-out" for the ImageInput or ImageOutput individually,
      overriding any global attribute("threads"). #1259 (1.6.6)
* ImageCache/TextureSystem:
    * Clarified in the docs that TextureSystem::get_texture_info and
      ImageCache::get_image_info "exists" queries should return true, and
      place in *data the value 1 or 0 depending on whether the image exists
      and can be read. (1.6.0/1.5.10)
    * Added handle-based versions of TextureSystem get_texture_info(),
      get_imagespec(), imagespec(), and get_texels(), in addition to the
      existing name-based versions of those methods. Note that
      texture(), environment(), and texture3d() already had both
      name-based and handle-based varieties. #1057 (1.6.1) #1083 (1.6.2)
    * Add create_thread_info() and destroy_thread_info() methods that
      allow an app to manage the per-thread records needed by the IC.
      #1080 (1.6.2)
    * Added ImageCache get_perthread_info() and get_image_handle() to
      return opaque perthread and file handle pointers, much like
      TextureSystem already had, and added handle-based versions of
      get_image_info(), get_imagespec(), imagespec(), get_pixels(), and
      get_tile(), in addition to the existing name-based versions of
      those methods. #1057 (1.6.1)
    * ImageCache get_tile and get_pixels have new varieties that let you
      request channel begin/end range. This allows you to control which
      channel ranges are in the cache, and thus be much more efficient
      with cache storage when only a few channels are needed from a file
      with many channels. #1226 (1.6.5)
* ImageBufAlgo:
    * New ImageBufAlgo functions: abs, absdiff, div, fill, noise, mad,
      invert, deepen, ociofiletransform.
    * nchannels() now takes an 'nthreads' parameters, just like all the
      other ImageBufAlgo functions. #1261 (1.6.6)
* Python bindings:
    * Added previously-M.I.A. ImageSpec::erase_attribute(). #1063 (1.6.1)
    * ImageSpec.set_channel_formats() now works when the channel
      type lists are either TypeDesc, in addition to the existing support
      for BASETYPE. #1113 (1.6.3/1.5.13)
    * Added Python bindings for DeepData and deep reads (ImageInput) and
      writes (ImageOutput), as well as additional DeepData and ImageBuf
      methods to fully match the C++ API. #1113 #1122 (1.6.3/1.5.13)
    * ImageBuf.set_pixels, and ImageBuf.get_pixels with ROI. #1167,1179 (1.6.4)
    * Change Python ImageOutput bindings to simplify the write_* methods.
      They no longer take both a TypeDesc and an array; it can figure out
      the type from the array itself. Also get rid of the stride parameters,
      which weren't useful in a Python context. #1184 (1.6.4)
    * ImageBufAlgo colorconvert, ociolook, and ociodisplay now take an
      optional string colorconfig argument. #1187 (1.6.4)
    * Fix missing Python bindings for global OIIO::getattribute(). #1290
      (1.6.8)
* The ColorConfig wrapper for OCIO functionality has been extended to
  parse color names from filename strings, and to report the recommended
  pixel data type for a color space. #1129 (1.6.3)
* C++11 definitions: oiioversion.h defines OIIO_BUILD_CPP11 as nonzero
  if OIIO itself was built in C++11 (or later) mode, and platform.h
  defines OIIO_USING_CPP11 as nonzero if at this moment C++11 (or
  later) mode is detected. Note that these can differ if one set of
  compiler flags was used to build OIIO, and a different set is used to
  build a project that uses OIIO headers. #1148 (1.6.4)
* Renamed the "fps" standard metadata to "FramesPerSecond. #1193 (1.6.4)
* Removed deprecated header "string_ref.h" (use string_view.h). (1.6.1)
* oiioversion.h: Renamed the namespace macros OIIO_NAMESPACE_ENTER/EXIT to
  OIIO_NAMESPACE_BEGIN/END, and roll the braces into it. #1196 (1.6.4)
* array_view.h: Refactor array_view to be more in line with what is slated
  for C++17, in particular it is now templated on Rank and so can be a view
  to a multi-dimensional array. Also change array_view_strided to have
  strides measured in units of sizeof(T), not bytes (to keep with C++17).
  This also adds coordinate.h to give definitions for the offset<>,
  bounds<>, and bounds_iterator<> templates used by array_view. #1205
  (1.6.4)
* Add top-level OIIO::get_int_attribute(), get_float_attribute(), and
  get_string_attribute() helpers, similar to how they work in many
  of the classes. #1283 (1.6.7)

Fixes, minor enhancements, and performance improvements:
* oiiotool
    * Bug fix for frame sequences -- could crash in Windows. #1060 (1.6.1)
    * Gracefully handle requests to save an image with more channels than
      the output file format can handle. Instead of being a fatal error,
      now it's just a warning, and extra channels are dropped. It tries to
      to find R, G, B, and A channels, saving them. If those names are
      not found, it just saves the first 3 (or 4) channels. #1058 (1.6.1)
    * Improve error messages when files can't be read. It is now easier
      to to distinguish files that don't exist from those that are an
      unknown format from those that are corrupted or have read
      errors. #1065 (1.6.1)
    * Flag errors properly when -d specifies an unknown data format name.
      #1077 (1.6.2/1.5.13)
    * oiiotool numeric wildcard improvement: allow more digits to match.
      #1082 (1.6.2/1.5.13)
    * Bug fix: input file data format didn't always end up in the output.
      (1.6.3)
    * --channels bugs were fixed when dealing with "deep" images. (1.6.3)
    * All the color space conversion operations run much faster now,
      since the underlying IBA::colorconvert() has been parallelized. (1.6.3)
    * --crop logic bug fixed in cases where the crop region was the same
      size as the original pixel data window. #1128 (1.6.3)
    * oiiotool now gives proper error messages when asked to perform
      unsupported operations on deep images. (1.6.3)
    * Bug fix: --frames incorrectly overrode explicit frame sequence
      wildcards on the command line. #1133 (1.6.3)
    * --crop, --trim, and --autotrim have been extended to work on
      "deep" images. #1137 (1.6.3)
    * For "procedural" ImageInputs, don't give "file doesn't exist"
      errors. (1.6.4)
    * Suppress output/copying of "textureformat" metadata inherited from
      input if it's not plausibly still a valid texture (i.e., if it's
      no longer tiled or MIPmapped). #1206 (1.6.4)
    * oiiotool's full help message lists all supported formats. #1210 (1.6.5)
    * oiiotool --help prints a briefer help screen. Use --help -v for
      the full-detail help. #1214 (1.6.5)
    * Bug fix in --fit when the image didn't need to be resized.
      #1227 (1.6.5/1.5.21)
    * Bug fix in --ch for "deep" files when the channel reordering is
      the same as it already was. #1286 (1.6.7)
* ImageBufAlgo:
    * compare() (and therefore oiiotool -diff and idiff) did not notice
      image differences when the pixels that differed had NaN or NaN or
      Inf values! Now it is right. #1109 (1.6.3/1.5.13)
    * channels() bugs were fixed when dealing with "deep" images. (1.6.3)
    * colorconvert() has been parallelized, and thus on most systems will
      now run much faster. (1.6.3)
    * render_text() handles UTF-8 input. #1121 (1.6.3)
    * colorconvert(), ociodisplay(), and ociolook() have new varities that
      accept an optional ColorConfig, rather than having no choice but to
      construct a new one internally. (1.6.3)
    * nonempty_region() and crop() have been extended to handle "deep"
      images. #1137 (1.6.3)
    * Fix bug in fft() -- was not always zeroing out the imaginary channel.
      #1171 (1.6.4/1.5.17)
    * Fixed uninitialized variable bugs with rangecompress() and
      rangeexpand() when using luma. #1180 (1.6.4)
    * The lanczos3, radial-lanczos, and catrom filters have been change
      from fixed-width to fully scalable. This fixes artifacts that
      occur when using them as upsizing filters. #1228,#1232 (1.6.5/1.5.21)
* maketx, TextureSystem, and ImageCache:
    * TextureSystem/IC now directly stores uint16 and half pixel data in
      the cache rather than converting internally to float for tile storage,
      thus effectively doubling the cache capacity for files of those
      formats. (1.6.3)
    * Fix broken bicubic texture sampling with non-power-of-two sized
      tiles. #1035 (1.6.0/1.5.10)
    * maketx: when the source image was a crop (data window != display
      window), and the sharpening filters were used, it would
      incorrectly issue an "unknown filter name" error. #1059 (1.6.1/1.5.12)
    * maketx: Flag errors properly when -d specifies an unknown data
      format name. #1077 (1.5.13)
    * maketx now writes to a temporary file, then moving it to the final
      requested output filename only when the write completed without
      error.  This prevents situations where maketx crashes or is killed
      and leaves behind a file that looks correct but is actually
      corrupted or truncated. #1072 (1.6.2/1.5.13)
    * TextureSystem bug fix that occasionally resulted in NaN in the
      alpha channel result when looking up from 3-channel images. #1108
      (1.6.3/1.5.13)
    * maketx --runstats prints runtime staticstics (deprecating --stats).
      #1152 (1.6.4)
    * Fixed trilinear MIPmap texture lookups that gave invalid alpha fill.
      #1163 (1.6.4/1.5.16)
    * The lanczos3, radial-lanczos, and catrom filters have been change
      from fixed-width to fully scalable. This fixes artifacts that
      occur when using them as upsizing filters. #1228,#1232 (1.6.5)
    * Texture cache memory efficiency is much better for the special case
      of accessing just a few channels from a texture file with large
      numbers of channels. #1226 (1.6.5)
    * Eliminate spurious ImageCache invalidation just because the shared
      cache is requested again. #1157 (1.6.4/1.5.16)
    * Statistics output also shows all the option setting values. #1226 (1.6.5)
    * Data copy error in ImageCache::get_pixels for partial-channel-set
      copies. #1246 (1.6.5)
    * maketx -u now remakes the file if command line arguments or OIIO
      version changes, even if the files' dates appear to match.
      #1281 (1.6.8)
* GIF:
    * Write support! #1193 (1.6.4)
    * On input, renamed "fps" metadata to "FramesPerSecond". #1193 (1.6.4)
* IFF:
    * Fix botched output of 16 bit uncompressed data. #1234 (1.6.5/1.5.21)
    * Make "rle" compression the default. #1234 (1.6.5/1.5.21)
* JPEG:
    * Now properly read/write xdensity and ydensity (what OIIO and TIFF
      call "XResolution" and "YResolution" and, therefore,
      "PixelAspectRatio". #1042 #1066 (1.6.0, 1.6.1)
    * Support JPEG files encoded as CMYK (by converting to RGB upon read)
      #1044 (1.6.1)
    * Fix misdeclared supports() which would make the JPEG plugin appear
      to not support exif or iptc. #1192 (1.6.4)
* JPEG-2000:
    * Fix handling of un-premultiplied alpha (which is dictated by the
      JPEG-2000 spec). (1.6.3)
    * Fix reading of YUV-encoded files. (1.6.3)
    * Read and write the ICC profile, if present. (1.6.3)
    * Handle all bit depth precisions properly (previously only 8, 10,
      12, and 16 were right). (1.6.3)
    * Set the full/display window correctly. (1.6.3)
    * Deal with differing per-channel data windows and sampling rates. (1.6.3)
* OpenEXR:
    * Improved handling of density and aspect ratio. #1042 (1.6.0)
    * Fix read_deep_tiles() error when not starting at the image origin.
      #1040 (1.6.0/1.5.10)
    * Fix output of multi-part exr file when some parts are tiled and
      others aren't. #1040 (1.6.0/1.5.10)
    * write_tile() with AutoStride calculated the wrong default strides
      for "edge" tiles when the image width or length was not an integer
      multiple of the tile size. Also clarified the PDF and imageio.h
      docs in how they explain strides for this case. #1055 (1.6.1/1.5.12)
    * Fix bugs in reading deep OpenEXR images with mixed channel types.
      #1113 (1.6.3/1.5.13)
    * OpenEXR output supports("deepdata") now correctly returns 'true'.
      #1238 (1.6.5/1.5.21)
    * A separate global OIIO::attribute("exr_threads") sets the thread pool
      size for OpenEXR's libIlmImf, independent of the OIIO thread fan-out
      attribute OIIO::attribute("threads"). #1244 (1.6.5)
    * Correctly read and write Matrix33 and double (scalar, 2d, 3d, m33, m44)
      metadata. #1265,#1267 (1.6.6)
    * Recognize AR/AG/AB channel names in addition to the old RA/RG/RB
      #1277 (1.6.6)
* PNG:
    * Writing PNG files now honors the PixelAspectRatio metadata.
      #1142 (1.6.3)
* PFM:
    * PFM (float extension of PNM) was incorrectly flipped top to bottom.
      Now fixed. #1230 (1.6.5)
* PSD:
    * Better error handling for files lacking "global layer mask info"
      or "additional layer info". #1147 (1.6.4/1.5.18)
    * Additional PSD signatures for global additional layer info.
      #1147 (1.6.4/1.5.18)
    * Better error handling when dealing with an empty layer mask.
      #1147 (1.6.4/1.5.18)
* TIFF:
    * Improved handling of density and aspect ratio. #1042 (1.6.0)
    * Improved proper handling of the interplay between "XResolution",
      "YResolution", and "PixelAspectRatio". #1042 (1.6.0)
    * TIFF output: recognize special "tiff:write_exif" metadata, which when
      present and set to 0, will skip writing the Exif directory into the
      TIFF file. This can be helpful when you expect the resulting TIFF
      file to be read with very old versions of libtiff. #1185 (1.6.4/1.5.18)
    * Correct read and write of JPEG-compressed TIFF. #1207 (1.6.4)
    * Correct support for reading LAB, LOG, YCbCr, subsampled chroma.
      #1207 (1.6.4)
    * Make robust to strange TIFF files that have unexpected MIP
      level-to-MIP level changes in planarconfig, photometric, palette,
      extrasamples, etc. #1220,1221 (1.6.5/1.5.20)
    * Support output of 2, 4, 10, and 12 bit unsigned ints into TIFF files.
      #1216 (1.6.5)
    * Make TIFF reading more robust to certain subimage-to-subimage
      changes that were thought to be invariant. #1221 (1.6.5)
    * CMYK is properly read and written. Upon read, CMYK is auto-converted
      to RGB (and the "tiff:ColorSpace" metadata is set to "CMYK"). For
      output, if "tiff:ColorSpace" metadata is set and nonzero, the RGB
      passed in will be auto-converted to CMYK upon writing.
      #1233 #1245 (1.6.5)
    * Recognize Exif tags in the main directory, not only the special
      Exif directory. #1250 (1.6.5)
    * Fix bug in read_scanlines when reading TIFF files with UNassociated
      alpha and unusual ystride values. #1278 (1.6.6)
* ImageBuf iterator constructors with 0-size ranges or ROIs have been
  fixed to look like they are immediately done(). #1141 (1.6.3)
* Fix bug in internal convert_image() that could corrupt certain image
  copying of non-contiguous data layouts. #1144 (1.6.3)
* Also search for OIIO plugins in [DY]LD_LIBRARY_PATH. #1153 (1.6.4/1.5.16)
* Nuke plugin: don't crash with NULL Knob* in TxReaderFormat::setMipLabels.
  #1212 (1.6.5/1.5.20)
* idiff -q results in quiet mode -- output nothing for success, only
  minimal errors to stderr for failure. #1231 (1.6.5)

Build/test system improvements:
* Python plugin is now build as a cmake "module" rather than "library",
  which fixes some things on OSX. #1043 (1.6.0/1.5.10)
* Various build fixes for Windows. #1052 #1054 (1.6.1)
* New CMake build-time option to specify the default plugin search path.
  #1056 (1.6.1/1.5.12)
* Fix build breaks for very old versions of Ilmbase (1.6 and earlier)
  that lack a definition of V4f used by our simd.h. #1048 (1.6.1/1.5.11)
* Fix signed/unsigned warning on 32 bit platforms in jpeginput.cpp.
  #1049 (1.6.1/1.5.11)
* New CMake build-time option to specify the default plugin search path.
  #1056 (1.6.1/1.5.12)
* Fix gcc 5.0 compiler warning in PtexHalf.cpp. (1.6.1/1.5.12)
* Remove dependency of OpenSSL by default. #1086 (1.6.2/1.5.13)
* Fix warnings when compiling with C++11. (1.6.3/1.5.13)
* Dont link Python framework on OSX. #1099 (1.6.3/1.5.13)
* Changed the way testtex warps the image to give faux perspective to
  test texture mapping. (1.6.3)
* Build-time USE_SIMD=... has been changed from accepting a single tag to
  a comma-separated list of feature options. So you can, for example, do
  make USE_SIMD=avx,f16c ...  (1.6.3)
* make USE_NINJA=1 causes CMake to build Ninja build files instead of
  Makefiles (they execute much faster, espectially for incremental builds).
  #1158 (1.6.4)
* PSD & JPEG plugins fixes for Win32 compilation. #1150 (1.6.4/1.5.16)
* Fix Nuke plugin build files to not do anything if USE_NUKE=0.
  #1156 (1.6.4/1.5.16)
* Builds now produce much less console output by default (use VERBOSE=1
  to get all the details, most of which is only useful when debugging
  broken builds). #1162 (1.6.4)
* Fix support for older ffmpeg version on Ubuntu 14.04. #1168 (1.6.4/1.5.17)
* Build-time fixes for Nocona CPUs that have SSE3 without SSSE3.
  #1175 (1.6.4/1.5.17)
* ustring internals fixes for gcc 5.x changs to std::string ABI. #1176 (1.6.4)
* Fixes for clean build with clang 3.6. #1182,1183 (1.6.4)
* Fix signed/unsigned comparison error. #1186 (1.6.4)
* Top-level Makefile option USE_OPENCV=0 to turn off even searching for
  OpenCV components. #1194 (1.6.4/1.5.18)
* If a system-installed (external) PTex implementation is found, use
  it.  Only use the "bundled" version if no other is found. Also add a
  top-level USE_PTEX=0 that will skip PTex support, even if the library
  is found.  #1195,1197 (1.6.4)
* Fix compiler warnings about int vs size_t mismatches. 1199 (1.6.4)
* Improve C++11 and C++14 readiness. #1200
* Fix build break with certain new versions of libraw. #1204 (1.6.4/1.5.19)
* Fix build warnings for new Apple tools release that upgrades the standard
  clang release. #1218 (1.6.5/1.5.20)
* When compiling in C++11 mode, std::unordered_map, mutex,
  recursive_mutex, lock_guard, bind, ref, cref, thread, shared_ptr will
  be used rather than boost equivalents, and our own thread_group and
  intrusive_ptr are now used rather than the boost equivalents. We
  believe that this completely removes all Boost headers and types from
  the OIIO public APIs when in C++11 mode. (Though internals still use
  Boost in some cases.) #1262 #1266 (1.6.6)
* We are now set up to use Travis-CI (https://travis-ci.org) for continuous
  integration / automatic builds of all merges and pull requests.
  #1268, #1269, #1273 (1.6.6)
* Don't install fonts if USE_FREETYPE is disabled. #1275 (1.6.6)
* Use ccache for builds when detected and safe (unless USE_CCACHE=0).
  #1274,#1285 (1.6.7)
* Failed tests now print their non-matching text output to the console
  when doing 'make test'. This makes it much easier to spot most errors.
  #1284 (1.6.7)

Developer goodies / internals:
* Strutil additions: parse_until, parse_nested (1.6.1), repeat
  (#1272/1.6.6/1.5.21).
* Give Strutil::parse_string an option to not strip surrounding quotes.
  (1.6.4)
* Made TypeDesc::equivalent accept comparisons of arrays of unspecified
  length with ones of definite length. #1072  (1.6.2/1.5.13)
* Add Filesystem::rename() utility. #1070  (1.6.2/1.5.13)
* New SIMD methods: insert<>, xyz0, vreduce_add, dot, dot3, vdot, vdot3,
  AxBxCxDx, blend0not (1.6.2)
* array_view enhancements that let you initialize an `array_view<const float>`
  from a const `std::vector<float>&`.  #1084 (1.6.2/1.5.14)
* hash.h contains several new hashes in namespaces 'OIIO::xxhash' and
  'OIIO::farmhash'. Also, Strutil::strhash now uses farmhash rather than
  the Jenkins one-at-a-time hash, bringing big speed improvements
  (including ustring creation). Beware that the strhash value returned
  will be different than they were before. #1090 (1.6.3)
* fmath: safe_fast_pow improves the precision of its results for
  special cases of pow(x,1) and pow(x,2). #1094 (1.6.3/1.5.13)
* Added TypeDesc::TypeHalf(). #1113 (1.6.3/1.5.13)
* thread.h: our atomic types have had their API adjusted somewhat to
  more closely conform to C++11's std::atomic. (1.6.3)
* ustring's internals and underlying hash table have been overhauled,
  yielding much higher performance, especially when many threads are
  simultaneously creating ustrings. (1.6.3)
* ROI improvement: make intersection & union robust to uninitialized ROIs
  as arguments. (1.6.3)
* osdep.h is deprecated. Use platform.h instead. (1.6.3)
* The DISPATCH_TYPES utility macros used internally by IBA have been
  improved, and in particular the DISPATCH_COMMON_TYPES now handle ALL
  types ("uncommon" ones are silently converted to float). (1.6.3)
* platform.h moves the endian functions into the OIIO namespace. (1.6.3)
* platform.h adds functions for runtime query of CPU capabilities. (1.6.3)
* simd.h: float4 and int4 can now construct and load from unsigned short*,
  short*, unsigned char*, char*, and 'half'. (1.6.3)
* Strutil::utf8_to_unicode (1.6.3)
* Filesystem::current_path(). #1124 (1.6.3/1.5.21)
* Filesystem enumerate_file_sequence and scan_for_matching_filenames
  have been modified to clear their result vectors rather than simply
  assume they are empty. #1124 (1.6.3)
* oiiotool internals have been refactored to be class-oriented and move
  a lot of boilerplate repeated in each op to be part of the base
  class. #1127 (1.6.3)
* timer.h: Timer and ScopedTimer have changed slightly. This isn't used
  in any public OIIO APIs, but may affect 3rd party programs that like
  to use OIIO's timer.h for convenience. #1201 (1.6.4/1.5.19)
* dassert.h: added OIIO_STATIC_ASSERT macros for static
  assertion. Doesn't affect existing OIIO apps since they are new
  additions, but feel free to use them! #1202 (1.6.4/1.5.19)
* New unit test for imagecache. #1246 (1.6.5)
* Sysutil::hardware_concurrency() and physical_concurrency(). #1263
  (1.6.6/1.5.21)

Release 1.5.24 (1 Mar 2016) -- compared to 1.5.23)
---------------------------------------------------
* Fix oiiotool --dumpdata, didn't work properly for non-float files.
* Fix broken OpenEXR multi-part output when parts have different pixel types.
* Update webp testsuite references for new webp version.

Release 1.5.23 (28 Dec 2015) -- compared to 1.5.22)
---------------------------------------------------
* Fix build break against Boost 1.60. #1299,#1300

Release 1.5.22 (16 Dec 2015) -- compared to 1.5.21)
---------------------------------------------------
* Deep OpenEXR: recognize the newer AR/AG/AB channel name convention. #1277
* Fix ffmpeg plugin compilation in some configurations. #1288
* Bug fix: TIFF read_scanlines of files with unassociated alpha didn't
  honor the 'ystride' parameter and could run off the end of the buffer
  for nonstandard stride arranagements. #1278
* Fix missing Python bindings for global OIIO::getattribute(). #1290

Release 1.5.21 (1 Dec 2015) -- compared to 1.5.20)
---------------------------------------------------
* Bug fix in --fit when the image didn't need to be resized. #1227
* IFF: Fix botched output of 16 bit uncompressed data. #1234
* IFF: Make "rle" compression the default for output. #1234
* OpenEXR output supports("deepdata") now correctly returns 'true'. #1238
* The lanczos3, radial-lanczos, and catrom filters have been changed
  from fixed-width to fully scalable. This fixes artifacts that
  occur when using them as upsizing filters. #1228,#1232
* Filesystem::current_path(). #1124
* Sysutil::hardware_concurrency() and physical_concurrency(). #1263
* Strutil::repeat() #1272

Release 1.5.20 (28 Sep 2015) -- compared to 1.5.19)
---------------------------------------------------
* Nuke plugin: don't crash with NULL Knob* in TxReaderFormat::setMipLabels.
  #1212
* Fix build warnings for new Apple tools release that upgrades the standard
  clang release. #1218
* Make TIFF reader robust to strange TIFF files that have unexpected MIP
  level-to-MIP level changes in planarconfig, photometric, palette,
  extrasamples, etc. We previously assumed these things would never vary
  between MIP levels of the same file, and Murphy called our bluff. #1220,1221

Release 1.5.19 (8 Sep 2015) -- compared to 1.5.18)
--------------------------------------------------
* Fix compile warnings on some platforms/compilers.
* Fix build break with certain new versions of libraw. #1204
* Internals: Timer and ScopedTimer have changed slightly. This isn't used
  in any public OIIO APIs, but may affect 3rd party programs that like
  to use OIIO's timer.h for convenience. #1201
* Internals: dassert.h has added OIIO_STATIC_ASSERT macros for static
  assertion. Doesn't affect existing OIIO apps since they are new
  additions, but feel free to use them! #1202

Release 1.5.18 (4 Aug 2015) -- compared to 1.5.17)
---------------------------------------------------
* PSD input improvements: better error handling for files lacking "global
  layer mask info" or "additional layer info"; additional PSD signatures
  for global additional layer info; better error handling when dealing
  with an empty layer mask. #1147
* TIFF output: recognize special "tiff:write_exif" metadata, which when
  present and set to 0, will skip writing the Exif directory into the TIFF
  file. This can be helpful when you expect the resulting TIFF file to be
  read with very old versions of libtiff. #1185
* Top-level Makefile option USE_OPENCV=0 to turn off even searching for
  OpenCV components. #1194

Release 1.5.17 (13 Jul 2015) -- compared to 1.5.16)
---------------------------------------------------
* Fix support for older ffmpeg version on Ubuntu 14.04. #1168
* Fix bug in fft -- was not always zeroing out the imaginary channel. #1171
* Build-time fixes for Nocona CPUs that have SSE3 without SSSE3. #1175
* ustring fixes for new gcc (5.1+) and new std::string ABI. #1176
* Fixes for unit test timer_test for new OSX versions with timer
  coalescing. #1181
* Fix bugs with rangecompress and rangeexpand when using luma. #1180
* Fixes for clean build when using clang 3.6. #1182

(ryoon)

Updated textproc/link-grammar to 5.5.1

(ryoon)

Update to 5.5.1

Changelog:
[ANNOUNCE] Link-Grammar Version 5.5.0 is now available.

Version 5.5.0 of link-grammar has been released.  It contains several
important bug-fixes for opencog users.

* The previous version accidentally broke the opencog API. This version
  fixes it.

* Linkages generated by the "ANY" random parser were not actually being
  randomized.  This is now fixed.  (Bug reported by Andres.)

* Poorly-formated dictionaries no longer report errors.  (Bug reported
  by Alexei/Anton)

The complete list of changes is:

* Fix accidental API breakage that impacts OpenCog.
* Fix memory leak when parsing with null links.
* Python bindings: Add an optional parse-option argument to parse().
* Add an extended version API and use it in "link-parser --version".
* Fix spurious errors if the last dict line is a comment.
* Fix garbage report if EOF encountered in a quoted dict word.
* Fix garbage report if whitespace encountered in a quoted dict word.
* Add a per-command help in link-parser.
* Add a command line completion in link-parser.
* Enable build of word-graph printing support by default.
* Add idiom lookup in link-parser's dict lookup command (!!idiom_here).
* Improve handling of quoted words (e.g. single words in "scare
* quotes").
* Fix random selection of linkages so that it's actually random.

You can download link-grammar from
http://www.abisource.com/downloads/link-grammar/current/

The website is here:
https://www.abisource.com/projects/link-grammar/

WHAT IS LINK GRAMMER?
The Link Grammar Parser is a syntactic parser of English (and other
languages as well), based on Link Grammar, an original theory of English
syntax. Given a sentence, the system assigns to it a syntactic structure,
which consists of a set of labelled links connecting pairs of words.

=================================================================
=================================================================
=================================================================

[ANNOUNCE] Link-Grammar Version 5.4.4 is now available.

I'm pleased to announce that version 5.4.4 is now available. I don't
normally announce minor versions, but this one was almost named 5.5.0.
Which suggests that there were some important changes. Dictionary
loading is now thread safe. Security vulnerabilities are fixed. Parsing
of Russian is now 2x faster than before. Connectors can be individually
given length limits - handy for morphology and phonetic agreement - and
the root reason for the Russian speedup.  An assortment of fixes to the
English dictionary, including a reversal of some back-sliding in the
test corpus.

You can download link-grammar from
http://www.abisource.com/downloads/link-grammar/current/

The website is here:
https://www.abisource.com/projects/link-grammar/

WHAT IS LINK GRAMMER?
The Link Grammar Parser is a syntactic parser of English (and other
languages as well), based on Link Grammar, an original theory of English
syntax. Given a sentence, the system assigns to it a syntactic structure,
which consists of a set of labelled links connecting pairs of words.

=================================================================
=================================================================
=================================================================

[ANNOUNCE] Link-Grammar Version 5.4.0 is now available.

I'm pleased to announce that version 5.4.0 is now available. Besides
including various bug fixes, this release is notable for completely
restructuring the organization of the source code, grouping files into
directories according to the processing stage that they implement.  See
below for the full ChangeLog.

You can download link-grammar from
http://www.abisource.com/downloads/link-grammar/current/

The website is here:
https://www.abisource.com/projects/link-grammar/

WHAT IS LINK GRAMMER?
The Link Grammar Parser is a syntactic parser of English (and other
languages as well), based on link grammar, an original theory of English
syntax. Given a sentence, the system assigns to it a syntactic
structure, which consists of a set of labelled links connecting pairs of
words. The parser also produces a "constituent" (Penn tree-bank style
phrase tree) representation of a sentence (showing noun phrases, verb
phrases, etc.).

(ryoon)

Try to fix "warning: duplicate script for target "pre-build" ignored"

(ryoon)

miniupnpd: consistently fails with MAKE_JOBS set for me
it doesn't wait until generating config.h is done.

(maya)

doc/TODO: add some

+ MesaLib-18.2.2, gstreamer1-1.14.4, nickle-2.82, ocaml-4.07.1,
  py-django2-2.1.2, py-mercurial-4.7.2.

(wiz)

doc: Updated security/vault to 0.11.2

(fhajny)

## 0.11.2 (October 2nd, 2018)

CHANGES:

- `sys/seal-status` now includes an `initialized` boolean in the
  output. If Vault is not initialized, it will return a `200` with
  this value set `false` instead of a `400`.
- `passthrough_request_headers` will now deny certain headers from
  being provided to backends based on a global denylist.

FEATURES:

- AWS Secret Engine Root Credential Rotation: The credential used by
  the AWS secret engine can now be rotated, to ensure that only Vault
  knows the credentials it is using.
- Storage Backend Migrator: A new `operator migrate` command allows
  offline migration of data between two storage backends.
- AliCloud KMS Auto Unseal and Seal Wrap Support (Enterprise):
  AliCloud KMS can now be used a support seal for  Auto Unseal and
  Seal Wrapping.

BUG FIXES:

- auth/okta: Fix reading deprecated `token` parameter if a token was
  previously set in the configuration
- core: Re-add deprecated capabilities information for now
- core: Fix handling of cyclic token relationships
- storage/mysql: Fix locking on MariaDB
- replication: Fix DR API when using a token
- identity: Ensure old group alias is removed when a new one is
  written
- storage/alicloud: Don't call uname on package init
- secrets/jwt: Fix issue where request context would be canceled too
  early
- ui: fix need to have update for aws iam creds generation
- ui: fix calculation of token expiry

IMPROVEMENTS:

- auth/aws: The identity alias name can now configured to be either
  IAM unique ID of the IAM Principal, or ARN of the caller identity
- auth/cert: Add allowed_organizational_units support
- cli: Format TTLs for non-secret responses
- identity: Support operating on entities and groups by their names
- plugins: Add `env` parameter when registering plugins to the catalog
  to allow operators to include environment variables during plugin
  execution.
- secrets/aws: WAL Rollback improvements
- secrets/aws: Allow specifying STS role-default TTLs
- secrets/pki: Add configuration support for setting NotBefore
- core: Support for passing the Vault token via an Authorization
  Bearer header
- replication: Reindex process now runs in the background and does not
  block other vault operations
- storage/zookeeper: Enable TLS based communication with Zookeeper
- ui: you can now init a cluster with a seal config
- ui: added the option to force promote replication clusters
- replication: Allow promotion of a secondary when data is syncing
  with a "force" flag

(fhajny)

doc: Updated mail/rspamd to 1.8.0

(fhajny)

mail/rspamd: Update to 1.8.0.

- [Feature] Add arguments schemas to processors and extractors
- [Feature] Add functional selectors library
- [Feature] Add generic selector to reputation module
- [Feature] Add more ratelimits: by digest, by attachments data, by
  filenames
- [Feature] Add preliminary stop words detection support
- [Feature] Add pure Lua debugm function
- [Feature] Add schema validation for Redis settings
- [Feature] Add selectors combine function
- [Feature] Add some recursion protection to lua logger
- [Feature] Add support for Lua API tracing
- [Feature] Allow to apply schema to arguments
- [Feature] Allow to get dkim signing data directly from HTTP headers
- [Feature] Allow to reuse existing authentication results
- [Feature] Cache selectors results in re runtime
- [Feature] Implement new text tokenizer based on libicu
- [Feature] Integrate selectors framework to multimap
- [Feature] Relax FORGED_RECIPIENTS
- [Feature] Support (almost) all html entities
- [Feature] Support adding and deletion of recipients in the milter
  block
- [Feature] Support gathering HTTP body from fragments in lua_http
- [Feature] Support multi flag in regexp and glob maps
- [Feature] Support selectors in ratelimit module
- [Feature] Support selectors in settings
- [Feature] Use khash in HTML parser
- [Feature] Use pure Lua debugm function
- [Fix] Add fail-safety for destroying sessions
- [Fix] Allow to add result-less fake DNS records
- [Fix] Another try to fix race conditions on config unload
- [Fix] Call Lua callback on DNS timeouts
- [Fix] Deprecate task:inc_dns_req as it is redundant
- [Fix] Do not allow events deletions on cleanup
- [Fix] Do not try to process skipped messages
- [Fix] Fix HTTP requests with no body
- [Fix] Fix another cleanup race condition
- [Fix] Fix bug in processing of pcre regexps
- [Fix] Fix byte array allocation in the pool
- [Fix] Fix crashes on task cleanup
- [Fix] Fix dynamic buckets in ratelimits
- [Fix] Fix endless loop when waiting for Rspamd to stop
- [Fix] Fix lua_util.str_split in case of delimiters set
- [Fix] Fix more issues with watching of async events
- [Fix] Fix stop words detection and loading logic
- [Fix] Fix various corner cases for language detection
- [Fix] Fix watchers in lua_tcp
- [Fix] Fix words decay algorithm
- [Fix] Implement watchers replacement to handle nested calls
- [Fix] Save faked code into fake dns record
- [Fix] Show the proper frame when using lua_util.debugm
- [Fix] Use fake dns records in tests
- [Fix] Use unicode replacements for HTML entities
- [Fix] fixed "cannot find dependency on symbol 1" issue when using
  replaced symbols in spamassassin rules
- [Fix] partition_id is not available in old versions of CH
- [Project] Add implicit conversion logic to selectors
- [Project] Add initial support for selectors in regexps
- [Project] Add method concept
- [Project] Further changes in unicode operations
- [Project] Implement Clickhouse migrations
- [Project] Implement implicit conversions to userdata
- [Project] Implement insert method
- [Project] Implement selectors registration for regular expressions
- [Project] Implement selectors support in re_cache
- [Project] Improve language detector: cleanup unused files,
  categorize
- [Project] Migrate CH data to a fat table
- [Project] Rework selectors logic
- [Project] Start Clickhouse utilities library
- [Project] Start unicode rework
- [Project] coroutine threaded model for API calls: thread pool
- [Rework] Move phishtank to a DNS based service
- [Rework] Rework Clickhouse plugin to use the new API
- [Rework] Rework language detector
- [Rework] Rework utf content processing in text parts
- [WebUI] Add progress bar for AJAX requests
- [WebUI] Avoid errors table reinitialization
- [WebUI] Avoid history table reinitialization
- [WebUI] Avoid throughput summary table reinitialization
- [WebUI] Destroy summary table on disconnect
- [WebUI] Fix "auth" request URL
- [WebUI] Fix disabling and hiding controls on page reload
- [WebUI] Fix maps loading from neighbours
- [WebUI] Fix symbols sorting by score
- [WebUI] Fix tables destroying
- [WebUI] Fix throughput data consolidation
- [WebUI] Fix upload buttons disabling

(fhajny)

doc: Updated net/ucspi-tcp6 to 1.10

(schmonz)

Update to 1.10. From the changelog:

- complete new build upon fehQlibs.
- tcprules delimiter can be customized via #define DELIMITER in tcprules.c.
- Removed options '-4' and '-6' for shell scripts and fixed bugs.
- Native handling of IPv4 and IPv6 addresses for tcpclient.

(schmonz)

doc: Added net/fehqlibs version 0.9

(schmonz)

Add and enable fehqlibs.

(schmonz)

Initial import of fehqlibs, a basic framework for byte, string, I/O,
socket, and IPv4/IPv6 handling in addition with the constant database
cdb together with dnsstub resolver routines for Internet services.

(schmonz)

doc: Updated devel/libowfat to 0.32

(schmonz)

Update to 0.32. From the changelog:

- remove OpenBSD #warning (obsd maintainer says no longer needed)
- fix fmt_ip6 (Erwin Hoffmann)
- add MSG_ZEROCOPY support (only used for buffers >8k)
- use writev in buffer_put for a slight perf improvement
- add experimental iom API for multithreaded I/O multiplexing (in io.h)

(schmonz)

ticket #5837-#5840

(spz)

Updated net/py-gcs-oauth2-boto-plugin, www/py-google-apitools, net/py-gsutil

(adam)

Pullup ticket #5839 - requested by maya
lang/spidermonkey52: security update
www/firefox-l10n: security update
www/firefox: security update

Revisions pulled up:
- lang/spidermonkey52/Makefile                                  1.10
- lang/spidermonkey52/distinfo                                  1.5
- lang/spidermonkey52/patches/patch-CVE-2018-12387              1.1
- www/firefox-l10n/Makefile                                    1.133
- www/firefox-l10n/distinfo                                    1.123
- www/firefox/Makefile                                          1.344
- www/firefox/distinfo                                          1.326

-------------------------------------------------------------------
  Module Name:    pkgsrc
  Committed By:  maya
  Date:          Wed Oct  3 17:30:30 UTC 2018

  Modified Files:
          pkgsrc/www/firefox: Makefile distinfo

  Log Message:
  firefox: update to 62.0.3

  Fixed hangs on macOS Mojave (10.14) when various dialog windows (upload, download, print, etc) are activated (bug 1489785)
  Fixed playback of some encrypted video streams on macOS (bug 1491940)

  Unvisited bookmarks can once again be autofilled in the address bar (bug 1488879)
  WebGL rendering issues (bug 1489099)
  Updates from unpacked language packs no longer break the browser (bug 1488934)
  Fix fallback on startup when a language pack is missing (bug 1492459)
  Profile refresh from the Windows stub installer restarts the browser (bug 1491999)
  Properly restore window size and position when restarting on Windows (bugs 1489214 and 1489852)
  Avoid crash when sharing a profile with newer (as yet unreleased) versions of Firefox (bug 1490585)
  Do not undo removal of search engines when using a language pack (bug 1489820)
  Fixed rendering of some web sites (bug 1421885)
  Restored compatibility with some sites using deprecated TLS settings (bug 1487517)
  Fix screen share on MacOS when using multiple monitors (bug 1487419)

  CVE-2018-12386: Type confusion in JavaScript
  CVE-2018-12387:
  CVE-2018-12385: Crash in TransportSecurityInfo due to cached data

  To generate a diff of this commit:
  cvs rdiff -u -r1.343 -r1.344 pkgsrc/www/firefox/Makefile
  cvs rdiff -u -r1.325 -r1.326 pkgsrc/www/firefox/distinfo

-------------------------------------------------------------------
  Module Name:    pkgsrc
  Committed By:  maya
  Date:          Wed Oct  3 17:31:07 UTC 2018

  Modified Files:
          pkgsrc/www/firefox-l10n: Makefile distinfo

  Log Message:
  firefox-l10n: catch up to www/firefox update.

  To generate a diff of this commit:
  cvs rdiff -u -r1.132 -r1.133 pkgsrc/www/firefox-l10n/Makefile
  cvs rdiff -u -r1.122 -r1.123 pkgsrc/www/firefox-l10n/distinfo

-------------------------------------------------------------------
  Module Name:    pkgsrc
  Committed By:  maya
  Date:          Wed Oct  3 18:58:23 UTC 2018

  Modified Files:
          pkgsrc/lang/spidermonkey52: Makefile distinfo
  Added Files:
          pkgsrc/lang/spidermonkey52/patches: patch-CVE-2018-12387

  Log Message:
  spidermonkey52: backport patch for CVE-2018-12387

  Don't inline push with more than 1 argument

  A vulnerability where the JavaScript JIT compiler inlines Array.prototype.push with multiple arguments that results in the stack pointer being off by 8 bytes after a bailout. This leaks a memory
  address to the calling function which can be used as part of an exploit inside the sandboxed content process.

  Bump PKGREVISION

  To generate a diff of this commit:
  cvs rdiff -u -r1.9 -r1.10 pkgsrc/lang/spidermonkey52/Makefile
  cvs rdiff -u -r1.4 -r1.5 pkgsrc/lang/spidermonkey52/distinfo
  cvs rdiff -u -r0 -r1.1 \
      pkgsrc/lang/spidermonkey52/patches/patch-CVE-2018-12387

(spz)

py-gsutil: updated to 4.34

Release 4.34:

New features
- Added bucket lock support to gsutil. Currently, your project must be
  whitelisted for use with the new bucket lock functionality. This restriction
  will be lifted in the near future.

Bug Fixes
- Fixed issue where "rsync -P" would fail if run as the root user.
- Fixed an issue with credential caching where the source credentials for an
  entity would change but the old cached credentials would still be used.

Other Changes
- OAuth2 token exchanges now go to https://oauth2.googleapis.com/token instead
  of https://accounts.google.com/o/oauth2/token. Users using gsutil behind a
  firewall may need to adjust their firewall rules.
- If invoked via the Cloud SDK, gsutil's debug output now displays the path to
  gcloud's gsutil wrapper script for "gsutil path", rather than the actual entry
  point for the bundled gsutil component.
- Improved error messages for failed Cloud KMS requests.
- Improved error messages for "iam ch" command to clarify that assigning
  roles to project convenience groups (e.g. "projectEditor") is not allowed.
- Enhanced perfdiag command to include GCE instance details (if applicable)
  and the target bucket's location and storage class.
- Several documentation updates and clarifications.

(adam)

py-google-apitools: updated to 0.5.24

0.5.24:
Bug fixes

(adam)

py-gcs-oauth2-boto-plugin: updated to 2.2

2.2:
Bug fixes.

(adam)

doc: Updated sysutils/etckeeper to 1.18.8

(schmonz)

Update to 1.18.8. From the changelog:

* Work around git commit's lack of robustness, by providing reasonable
  default values for GIT_COMMITTER_EMAIL etc. This was already done as
  part of the su/sudo handling, and is now always done.
* Don't hardcode the master branch when pushing to PUSH_REMOTE.
  Instead, let git push whatever branches it is configured to push to
  that remote.
* Added some unit tests.
  Thanks, Henrik Riomar.
* etckeeper will work on systems that do not have perl installed.
  (perl is still used when available as it's faster)
  Thanks, William Johansson and radhus.
* Prevent LC_ALL overriding the LC_COLLATE used to sort metadata.
* Only show errors (no progress indicators) when pushing Git/Mercurial repos
  to avoid unncessary cron mails.
  Thanks, Nils Steinger.
* Fix regex in 20-warn-problem-files.
* Added support for apk (alpine linux)
  Thanks, Henrik Riomar.

(schmonz)

Updated textproc/py-alabaster, security/py-rsa

(adam)