OpenImageIO 2.1.9.0

This is the official transition of the supported production release to 2.1. Some very high level highlights you might want to know about 2.1:

• The primary documentation is now hosted on https://openimageio.readthedocs.io and there you can find both beautiful HTML as well as PDF. A PDF book is no longer included in the distribution. I think everything has transitioned properly, but if you notice anything deficient in the new docs, please let me know.
• We have overhauled the build system, modernized and improved. It is much better at finding, hinting, or excluding individual components and dependent packages. The new build scripts now require a CMake minimum version of cmake 3.12. Please read the notes below for details.
• Support for HEIC/HEIF images.
• A new notation for getting and setting attributes in ParamList and ImageSpec. For details, see #2204 and #2297.
• Improved performance particularly for DPX reading, reading individual frames from movie files, and TextureSystem improvements under high threading.
• Although we still use Travis and Appveyor (for now), much of the CI heavy lifting is now being done by GitHub Actions CI.

We will continue to issue new (compatible) 2.1.x releases approximately monthly, with bug fixes or minor/safe/nonbreaking enhancements.

Detailed release notes are below. Enjoy.

Release 2.1 (8 Dec 2019) -- compared to 2.0

New minimum dependencies:

• CMake minimum is now 3.12. #2348 (2.1.5)

Major new features and performance improvements:

• Support for HEIC/HEIF images. HEIC is the still-image sibling of HEVC
  (a.k.a. H.265), and compresses to about half the size of JPEG but with
  higher visual quality. #2160 #2188 (2.1.0)
• oiiotool new commands: -evaloff -evalon --metamerge --originoffset
• ImageCache/TextureSystem improved perf of the tile and file caches under
  heavy thread contention. In the context of a renderer, we have seen
  improvements of around 7% in overall render time, averaged across a suite
  of typical production scenes. #2314, #2316 (2.1.3) #2381 #2407 (2.1.8)
• Fix huge DPX reading performance regression. Technically this is a bug
  fix that restores performance we once had, but it's a huge speedup.
  #2333 (2.1.4)
• Reading individual frames from very-multi-image files (movie files) has
  been greatly sped up (10x or more). #2345 (2.1.4)

Public API changes:

• ImageSpec new methods getattribute() and getattributetype(). #2204
  (2.1.1)

• ImageSpec and ParamValueList now support operator ["name"] as a way
  to set and retrieve attributes. For example,

  myimagespec["compression"] = "zip";
  myimagespec["PixelAspectRatio"] = 1.0f;
  int dither = myimagespec["oiio:dither"].get<int>();
  std::string cs = myimagespec["colorspace"];
  

  See the documentation about "Attribute Delegates" for more information,
  or the new header attrdelegate.h. #2204 (2.1.1) #2297 (2.1.3)

• ImageSpec::find_attribute now will retrive "datawindow" and "displaywindow"
  (type int[4] for images int[6] for volumes) giving the OpenEXR-like bounds
  even though there is no such named metadata for OIIO (the results will
  assembled from x, y, width, height, etc.). #2110 (2.1.0/2.0.4)

• "Compression" names (where applicable) can now have the quality appended
  to the name (e.g., "jpeg:85") insead of requiring quality to be passed
  as a separate piece of metadata. #2111 (2.1.0/2.0.5)

• Python: define __version__ for the module. #2096 (2.1.0/2.0.4)

• Python error reporting for ImageOutput and ImageBuf.set_pixels
  involving transferring pixel arrays have changed from throwing exceptions
  to reporting errors through the usual OIIO error return codes and queries.
  #2127 (2.1.0/2.0.5)

• New shell environment variable OPENIMAGEIO_OPTIONS can now be used to
  set global OIIO::attribute() settings upon startup (comma separated
  name=value syntax). #2128 (2.1.0/2.0.5)

• ImageInput open-with-config new attribute "missingcolor" can supply
  a value for missing tiles or scanlines in a file in lieu of treating it
  as an error (for example, how OpenEXR allows missing tiles, or when reading
  an incompletely-written image file). A new global OIIO::attribute()
  setting (same name) also accomplishes the same thing for all files read.
  Note that this is only advisory, and not all file times are able to do
  this (OpenEXR is the main one of interest, so that works). #2129 (2.1.0/2.0.5)

• ImageCache::invalidate() and TextureSystem::invalidate() now take an
  optional force parameter (default: true) that if false, will only
  invalidate a file if it has been updated on disk since it was first opened.
  #2133, #2166 (2.1.0/2.0.5)

• New filter name "nuke-lanczos6" matches the "lanczos6" filter from Nuke.
  In reality, it's identical to our "lanczos3", but the name alias is
  supposed to make it more clear which one to use to match Nuke, which uses
  a different nomenclature (our "3" is radius, their "6" is full width).
  #2136 (2.1.0/2.0.5)

• New helper functions in typedesc.h: tostring() converts nearly any
  TypeDesc-described data to a readable string, convert_type() does data
  type conversions as instructed by TypeDesc's. #2204 (2.1.1)

• ImageBuf:

  • Construction from an ImageSpec now takes an optional zero parameter
    that directly controls whether the new ImageBuf should have its buffer
    zeroed out or left uninitialized. #2237 (2.1.2)
  • set_write_format() method has a new flavor that takes a
    cspan<TypeDesc> that can supply per-channel data types. #2239 (2.1.1)

• ColorConfig:

  • Added getColorSpaceFamilyByName(), getColorSpaceNames(),
    getLookNames(), getDisplayNames(), getDefaultDisplayName(),
    getViewNames(), getDefaultViewName(). #2248 (2.1.2)
  • Added Python bindings for ColorConfig. #2248 (2.1.2)

• Formal version numbers are now four parts: MAJOR.MINOR.PATCH.TWEAK.
  #2313,#2319 (2.1.3)

• ImageInput now sets "oiio:subimages" attribute to an int representing the
  number of subimages in a multi-image file -- if known from reading just
  the header. A positive value can be relied upon (including 1), but a
  value of 0 or no such metadata does not necessarily mean there are not
  multiple subimages, it just means it could not be known from inexpensively
  reading only the header. #2344 (2.1.4)

• The imagesize_t and stride_t values now have revised definitions.
  It should be fully API/ABI compatible (at least for 64 bit systems), but
  is a simpler, more modern, more platform-independent definition.
  #2351 (2.1.5)

• DeepData has been altered to make pixel indices and total counts int64_t
  rather than int, in order to be safe for very large images that have > 2
  Gpixels. #2363 (2.1.5)

• On OSX, we now expect non-embedded plugins to follow the convention of
  naming runtime-loaded modules foo.imageio.so (just like on Linux),
  whereas we previously used the convention of foo.imageio.dylib. Turns
  out that dylib is supposed to be only for shared libraries, not runtime
  loadable modules. #2376 (2.1.6)

Fixes and feature enhancements:

• oiiotool:
  • New -evaloff and -evalon lets you disable and enable the expression
    substitution for regions of arguments (for example, if you have an
    input image filename that contains {} brace characters that you want
    interpreted literally, not evaluated as an expression). #2100 (2.1.0/2.0.4)
  • --dumpdata has more intelligible output for uint8 images. #2124
    (2.1.0/2.0.4)
  • Fixed but that could prevent -iconvert oiio:UnassociatedApha 1 from
    correctly propagating to the input reader. #2172 (2.1.0/2.0.6)
  • -o:all=1 (which outputs all subimages to separate files) fixed a
    crash that would occur if any of the subimages were 0x0 (it could
    happen; now it just skips outputting those subimages). #2171 (2.1.0)
  • Improved support of files with multiple subimages: Several commands
    honored -a but did not respect individual allsubimages= modifiers
    (--ch, --sattrib, --attrib, --caption, --clear-keywords,
    --iscolorspace, --orientation, --clamp, -fixnan); Several commands
    always worked on all subimages, but now properly respect -a and
    allsubimages= (--origin, --fullpixels, --croptofull, --trim);
    Several commands were totally unaware of subimages, but now are so and
    respect -a and allsubimages= (--crop, --fullsize, --zover, --fill,
    --resize, --resample). #2202 #2219, #2242 (2.1.1, 2.1.2)
  • --ociodisplay: empty display or view names imply using the default
    display or view. #2273 (2.0.10/2.1.3)
  • --metamerge option causes binary image operations to try to "merge"
    the metadata of their inputs, rather than simply copy the metadata
    from the first input and ignore the others. #2311 (2.1.3)
  • --colormap now supports a new "turbo" color map option. #2320 (2.1.4)
  • Expression evaluation has been extended to support operators // for
    integer division (whereas / is floating point division), and %
    for integer modulus. #2362 (2.1.5)
  • New --originoffset resets the data window origin relative to its
    previous value (versus the existing --origin that sets it absolutely).
    #2369 (2.1.5)
  • --paste has two new optional modifiers: :all=1 pastes the entire
    stack of images together (versus the default of just pasting the top
    two images on the stack), and :mergeroi=1 causes the result to have
    the merged data window of all inputs, instead of the foreground image
    clipping against the boundary of the background image data. #2369 (2.1.5)
  • --paste now works with deep images. #2369 (2.1.5)
  • --paste semantics have c...

OpenImageIO 2.1.8.2-RC3

• Fix important bug that surfaced with the new fmt code update.
• Remove obsolete PDF docs, rely on readthedocs now.
• Turn off default for STOP_ON_WARNING.

I am expecting to make a final release on 8 Dec.

OpenImageIO 2.0.13

This is the newest stable supported release, guaranteed to be API/ABI back-compatible with prior 2.0.x releases. This will likely be the LAST update to 2.0 as the "supported release" family, though there may be subsequent bug fixes as it transitions to the obsolete (but occasionally supported) branch. Release notes are below.

Release 2.0.13 (1 Dec 2019) -- compared to 2.0.12

• Bug fix in deep image compare (IBA::compare() or oiiotool --compare)
  would fail to notice differences in deep images where the corresponding
  pixels had differing numbers of samples. #2381 (2.1.8/2.0.13)
• DPX: Fix bugs related to int32 math that would lead to incorrect behavior
  in very high-resolution files. #2396 (2.1.3/2.0.13)
• When converting images to texture (via maketx or IBA::make_texture),
  correctly handle color space conversions for greyscale images. #2400
  (2.1.8/2.0.13)
• Build: suppress warnings with libraw for certain gcc versions.
• Build: Fix compiler warnings in ustring.h when _LIBCPP_VERSION is not
  defined. #2415 (2.1.8.1/2.0.13)

OpenImageIO 2.1.8.1-RC2

This is the second and hopefully final release candidate for the IMMINENT transition of 2.1 to be the supported release branch. I plan to retag it as the final release on 8 December. The API/ABI is hereby and irrevocably frozen; subsequent 2.1.x releases should all be fully back compatible for both API and ABI. The only changes since RC1 are a few minor build/warning fixes, and switch of the embedded "fmt" string formatting library to a newer version. Full release notes will be published on 8 Dec when the final release is announced (or you can check the CHANGES.md that comes with it).

OpenImageIO 2.1.8.0-RC1

This is the release candidate for 2.1. Please test. We will strive to keep the API locked and also to only add critical bug fixes between now and the scheduled release on 1 December.

OpenImageIO 2.0.12

Release 2.0.12 (1 Nov, 2019) -- compared to 2.0.11

• Fix compiler warnings on some platforms. #2375
• Work around bug in OpenEXR, where dwaa/dwab compression can crash when
  used on 1-channel tiled images with a tile size < 16. This can crop up for
  MIP-maps (high levels where rez < 16), so we detect this case and switch
  automatically to "zip" compression. #2378

OpenImageIO 2.1 Beta (2.1.7-beta)

We have also branched 2.1 and tagged Release-2.1.7-beta. This has many big changes (and potential compatibility breaks) compared to 2.0. I encourage you all to try it and report any problems. Please see the CHANGES.md in that branch for the full list of enhancements and fixes.

My plan is to let the beta span the course of November, with a goal of declaring a true 2.1 release on December 1, at that point 2.1 then becoming the fully supported stable release branch and 2.0 fading to be an obsolete branch that will only get sporadic updates to address the most critical bugs.

During the beta period, there may be some last minute touch-ups that change the API/ABI, but once we hit release candidates at the end of the month, the API and ABI will be frozen as befitting a supported release.

For the moment, master will continue to track the 2.1 betas, but at the moment that we have a PR that we agree should not be incorporated into 2.1 (because it's risky or breaks compatibility), master will diverge and truly be the pre-2.2 work area.

OpenImageIO 2.0.11

Release 2.0.11 (1 Oct, 2019) -- compared to 2.0.10

• Fixes to build against LibRaw master. #2306
• Fix DPX reading performance regression. #2333
• Guard against buggy pybind11 versions. #2347
• Fixes for safe Cuda compilation of invert<> in fmath.h. #2197

OpenImageIO 2.0.10

Release 2.0.10 (1 Aug, 2019) -- compared to 2.0.9

• ColorConfig improvements: (a) new getColorSpaceFamilyByName(); (b) new
  methods to return the list of all color spaces, looks, displays, or views
  for a display; (c) all of ColorConfig now exposed to Python. #2248
• IBA::ociodisplay() and oiiotool --ociodisplay: empty display or view
  names imply using the default display or view. #2273
• Bug fix in ustring::compare(string_view), in cases where the string_view
  was longer than the ustring, but had the same character sequennce up to
  the length of the ustring. #2283
• oiiotool --stats: Fixed bug where -iconfig hints were not being
  applied to the file as it was opened to compute the stats. #2288
• Bug fix: IBA::computePixelStats() was not properly controlling the
  number of threads with the nthreads parameter. #2289
• Bug fix when reading TIFF bugs: In cases where the reader needed to close
  and re-open the file silently (it could happen for certain scanline
  traversal patterns), the re-open was not properly honorig any previous
  "rawcolor" hints from the original open. #2285
• Nuke txWriter updates that expose additional make_texture controls. #2290
• Build system: Improvements for finding certain new Boost versions. #2293
• Build system: Improvements finding OpenEXR installation.
• Fix bugs when reading TIFF "cmyk" files. #2292.
• DPX: support for reading and writing 1-channel (luma, etc.) DPX images.
  #2294
• IBA::fixNonFinite(): fixed impicit float/double casts to half. #2301
• Build fixes for MinGW. #2304

Release 2.0.9

Release 2.0.9 (4 Jul, 2019) -- compared to 2.0.8

• RAW: Clarification about color spaces: The open-with-config hint
  "raw:ColorSpace" is more careful about color primaries versus transfer
  curve. Asking for "sRGB" (which is the default) gives you true sRGB --
  both color primaries and transfer. Asking for "linear" gives you linear
  transfer with sRGB/Rec709 primaries. The default is true sRGB, because it
  will behave just like JPEG. #2260 (2.1.2)
• Improved oiiotool support of files with multiple subimages: Several
  commands honored -a but did not respect individual allsubimages=
  modifiers (--ch, --sattrib, --attrib, --caption, --clear-keywords,
  --iscolorspace, --orientation, --clamp, -fixnan); Several commands always
  worked on all subimages, but now properly respect -a and allsubimages=
  (--origin, --fullpixels, --croptofull, --trim); Several commands were
  totally unaware of subimages, but now are so and respect -a and
  allsubimages= (--crop, --fullsize, --zover, --fill, --resize,
  --resample). #2202 #2219, #2242
• Fix broken ability to specify compression of multipart exr files. #2252
• Fix Strutil::stof() return type error and other windows warnings. #2254
• IBA::colortmatrixtransform() and oiiotool --ccmatrix allow you to
  perform a matrix-based color space transformation. #2168
• Guard simd.h against shenanigans when Xlib.h having been included and
  #defineing True and False. #2272
• RAW: Clarification about color spaces: The open-with-config hint
  "raw:ColorSpace" is more careful about color primaries versus transfer
  curve. Asking for "sRGB" (which is the default) gives you true sRGB --
  both color primaries and transfer. Asking for "linear" gives you linear
  transfer with sRGB/Rec709 primaries. The default is true sRGB, because it
  will behave just like JPEG. #2260
• Fix inability for python to set timecode attributes (specifically, it was
  trouble setting ImageSpec attributes that were unnsigned int arrays).
  #2279