Release version: LibRaw 0.21.2

Source code for Unix (Linux, Mac OS X, FreeBSD) and Windows (32 and 64 bits). Mac OS X 11+ binaries (and sources), 64-bit ARM/Intel, compiled with Xcode 15 . Windows binaries (and sources), compiled by MSVC 2022, 64 bit

LibRaw 0.21.2 changes (relative to 0.21.1)

  • New compile-defined limit LIBRAW_MAX_PROFILE_SIZE_MB: limits allocation/read size for embedded color profile (default: 256Mb)
  • Embedded color profile allocation/read size: limited by input file size.
  • Multiple fixes (mostly inspired by oss-fuzz) to improve library stability and/or input checks.
  • raw-identify: use fallback if PATH_MAX not available
  • Disabled color conversion for Canon 16-bit thumbnails
  • docs/changelog: explained the case when no thumbnail is found in specific file
  • swapXX renamed to libraw_swapXX to avoid name conflict
  • better striped thumbnails handling

LibRaw 0.21.1 changes (relative to 0.21.0)

  • Fixed typo in Panasonic metadata parser that prevents correct data processing for some Panasonic cameras.

LibRaw 0.21.0 changes (relative to 0.20)

Camera format support

  • Phase One/Leaf IIQ-S v2 support
  • Canon CR3 filmrolls/RawBurst
  • Canon CRM (movie) files
  • Tiled bit-packed (and 16-bit unpacked) DNGs
  • (non-standard) Deflate-compressed integer DNG files are allowed

Camera support

  • Canon EOS R3, R7 and R10
  • Fujifilm X-H2S, X-T30 II
  • OM System OM-1
  • Leica M11
  • Sony A7-IV (ILCE-7M4)
  • DJI Mavic 3
  • Nikon Z9: standard compression formats only

Multiple (resultion) thumbnails support

  • New imgdata.thumbs_list data item with data fields:
      int thumbcount: thumbnail count
      libraw_thumbnail_item_t thumblist[LIBRAW_THUMBNAIL_MAXCOUNT]: list of thumbnails
      libraw_thumbnail_item_t fields:
      enum LibRaw_internal_thumbnail_formats tformat: internal thumbnail format (this is not PPM/JPEG, but internal type related to thumbnail reading/unpacking method)
      ushort twidth, theight: thumbnail image size.
      ushort tflip: image rotation (see notes below)
      unsigned tlength: on-disk data size (not uncompressed for compressed formats)
      unsigned tmisc: bit depth and channel count: (color INT64 toffset: Thumbnail data offset in file
    • Only TIFF-based and CR3 files are parsed for thumbnail list, other formats will have thumbcount = 1 (or 0 if no thumbnail found in file).
    • Thumbnail image size may be unknown (not recorded in metadata), in this case twidth and theight are zero. Usually small(er) thumbnails will always have twidth/theight filled, while largest one may have these fields set to zero.
    • Thumbnail rotation (tflip) is filled only for TIFF-based RAWs (if Orientation tag is not present in IFD, default zero value is used) For non-tiff images, tflip is initialized to 0xffff (not known...)
    • There is no code to select thumbnail based on some criteria (e.g. minimal resolution), it is left to user to implement such a criteria, if needed (see unpack_thumb_ex() call below)
    • If you want to get largest possible thumbnail: just use old unpack_thumb() call to get it.
  • new API call: LibRaw::unpack_thumb_ex(int i): Unpacks i-th thumbnail if it exists into imgdata.thumbnail structure i should be non-zero and less then imgdata.thumbs_list.thumbcount.
  • samples/simple_dcraw.cpp: new -E command line switch to extract all thumbnails from input file(s)

(Experimental) RawSpeed "version 3" (develop branch) support

Build with -DUSE_RAWSPEED3, see details in RawSpeed3/
LibRaw::capabilities will set bit LIBRAW_CAPS_RAWSPEED3 if compiled w/ RawSpeed-v3 support
If file was processed (or tried to process) via RawSpeed-v3, these bits are rised in imgdata.process_warnings:
  • LIBRAW_WARN_RAWSPEED3_PROCESSED - processed via RawSpeed v3
  • LIBRAW_WARN_RAWSPEED3_PROBLEM - not processed (due to exception in RawSpeed library)
  • LIBRAW_WARN_RAWSPEED3_NOTLISTED - file not listed in cameras.xml data
RawSpeed-v3 support is controlled via libraw_decoder_info.t->flags (LIBRAW_DECODER_TRYRAWSPEED3 bit). The bit is set for:
  • lossless jpeg decoder
  • Canon sRAW decoder
  • lossless compressed and packed-compressed DNG decoders
  • Pentax decoder
  • Nikon (compressed) decoder
  • Phase One Compressed and Hasselblad/Compressed decoders
  • bitpacked decoder
  • Panasonic (old 12-bit compression) decoder
  • Olympus decoder
  • Sony ARW (v1 and v2) decoders
  • Samsung (version 1) decoder

(Experimental) Fine control of RawSpeed version selection/use

To enable, build LibRaw with -DUSE_RAWSPEED_BITS
LibRaw::capabilities will set bit LIBRAW_CAPS_RAWSPEED_BITS if compiled with this flag.
If enabled: imgdata.rawparams.use_rawspeed becomes bit-field (instead of 0/1 off/on) with bits:
  • LIBRAW_RAWSPEEDV1_USE - use RawSpeed Version 1 (if compiled with)
  • LIBRAW_RAWSPEEDV1_FAILONUNKNOWN - do not process unknown files (not listed in RawSpeed v1 camera definitions file)
  • LIBRAW_RAWSPEEDV1_IGNOREERRORS - ignore mirror decoding errors (out of range data,etc)
  • LIBRAW_RAWSPEEDV3_USE - Use RawSpeed Version 3 (if compiled with)
  • LIBRAW_RAWSPEEDV3_FAILONUNKNOWN - do not process unknown files
  • LIBRAW_RAWSPEEDV3_IGNOREERRORS - ignore minor decoding errors

Misc changes/improvements

  • New compile time define LIBRAW_OWN_SWAB to use on platforms without swab() in C/C++ runtime
    If set: image orientation is set based on TIFF/IFD0:Orientation tag, makernotes orientation data is ignored
  • Nikon makernotes: read NEFCompression tag for HE/HE* files
  • Nikon orientation tag: more fixed offsets for known cameras
  • Adobe DNG SDK 1.6 support (meaning, just an additional patch for GPR SDK)
  • Removed LibRaw::memerr(), memory allocation failures are already handled in LibRaw_memmgr::*alloc w/ exception thrown
  • Removed memory error callback, out-of-memory error should be handled via returned error code check, all related parameters (e.g. LibRaw constructor option to not set such callback) is also removed
  • imgdata.params.adjust_maximum_thr is settable via C-API libraw_set_adjust_maximum_thr(..)
  • New rawoptions/processing flags for DNG processing if compiled with Adobe DNG SDK:
    LIBRAW_RAWOPTIONS_DNG_STAGE2_IFPRESENT,LIBRAW_RAWOPTIONS_DNG_STAGE3_IFPRESENT If these flag(s) are set: Stage2/Stage3 processing will be performed only if OpcodeList2/OpcodeList3 tags are present in the input DNG file Old (hard) flags (LIBRAW_RAWOPTIONS_DNG_STAGE2 and LIBRAW_RAWOPTIONS_DNG_STAGE3) are not removed and will force Stage2/3 processing if set (regardless of input file tags)
  • New imgdata.rawparams.options bit: LIBRAW_RAWOPTIONS_DNG_ADD_MASKS If set: DNG Transparency Masks will be extracted (if selected via shot_select)

Bugs fixed

  • Fixed possible out-of-buffer read in Nikon orientation tag parser
  • Windows datastream: fixed logic errors (not showing up in real life)
  • Out-of-range read-only array access in postprocessing if output_color is set to 0 (raw color)
  • Minolta Z2 was not recognized correctly on 32-bit systems
  • Fixed possible buffer overflow in Kodak C330 decoder
  • dcraw_process(): check for buffer allocation results to avoid NULL deref
  • Multiple bugfixes inspired by oss-fuzz project

Previous version: LibRaw 0.20

Source code for Unix (Linux, Mac OS X, FreeBSD) and Windows (32 and 64 bits). Mac OS X 10.15 binaries (and sources), 64-bit, compiled with Xcode 11.6 . Windows binaries (and sources), compiled by MSVC 2019, 64 bit

LibRaw 0.20.2 changes (relative to 0.20.1)

Reverted 0.20.1 change:
    const buffer for open_buffer() and open_bayer() calls
Because of 0.20.0 ABI break.

LibRaw 0.20.1 changes (relative to 0.20.0)


  • exif callback is called on EXIF GPS and EXIF Interop IFDs
  • open_bayer call documented
  • Canon (ColorDatsSubver==32): parse Specular White instead of hardcoded value

Fixes for normal files processing:

  • Olympus XZ-1: do not provide linear_max (it is wrong in metadata)
  • Nikon Z cameras: added space in camera list
  • raw-identify: fixed wb-preset print
  • Pentax Optio 33WR: maker index was incorrect
  • dcraw_emu: corrected help line for -6 option.
  • raw-identify: corrected range check for color matrices print
  • use_camera_matrix option: fixed a bug introduced when making compiler more happy.

Fixes for damaged/special crafted files processing:

  • Fix for truncated CR3 files parsing
  • DNG metadata merger: all color loops are limited to MIN(4,colors)
  • Check for marings: should be less than raw image size
  • Check for xmpdata present in Samsung Lens ID assignment
  • Check for column range in leaf_hdr decoder
  • Additional checks in Hasselblad model parser
  • Fuji rotate: better limits check
  • DNG files: limit tiff_samples

Not fixes, but makes ASAN/compilers/etc happy:

  • corrected GPS EXIF output
  • const buffer for open_buffer() and open_bayer() calls

Camera Format support

  • Canon CR3
  • GoPro (via GPR SDK)
  • Panasonic 14-bit
  • Fujifilm compressed/16bit
  • Rapsberry Pi RAW+JPEG format (if USE_6BY9RPI defined)
  • Foveon X3F support changed: it is supported only if USE_X3FTOOLS defined at build (see below for 'Imported code policy disclaimer')

Camera support (1133 total)

  • Canon: PowerShot G5 X Mark II, G7 X Mark III, SX70 HS, EOS R, EOS RP, EOS 90D, EOS 250D, EOS M6 Mark II, EOS M50, EOS M200, EOS 1DX Mark III (lossless files only)
  • DJI Mavic Air, Air2, Osmo Action
  • FujiFilm GFX 100, X-A7, X-Pro3, X100V, X-T4 (uncompressed and lossless compressed only), X-T200
  • GoPro Fusion, HERO5, HERO6, HERO7, HERO8
  • Hasselblad L1D-20c, X1D II 50C
  • Leica D-LUX7, Q-P, Q2, V-LUX5, C-Lux / CAM-DC25, SL2, M10 Monochrom
  • Nikon D780, Z50, P950
  • Olympus TG-6, E-M5 Mark III, E-PL10, E-M1 Mark III,
  • Panasonic DC-FZ1000 II, DC-G90, DC-S1, DC-S1R, DC-S1H, DC-TZ95
  • PhaseOne IQ4 150MP
  • Ricoh GR III
  • Sony A7R IV, A9 II, ILCE-6100, ILCE-6600, RX0 II, RX100 VII
  • Zenit M
  • also multiple smartphones (the tested ones are listed in LibRaw::cameraList)

Source code re-arranged

  • dcraw.c is not used in the generation and build processes
  • dcraw_common.cpp and libraw_cxx.cpp are split into multiple code chunks placed in separate subfolders (decoders/ for raw data decoders, metadata/ for metadata parsers, etc)
  • dcraw_common.cpp and libraw_cxx.cpp remain to preserve existing build environments (these files are now just a bunch of #include directives).
  • It is possible to build LibRaw (It may be useful to reduce library memory/code footprint.)
    • without postprocessing functions (dcraw_process() and called function)
    • without postprocessing and LibRaw::raw2image() call (and called function).
    • See Makefile.devel.nopp and Makefile.devel.noppr2i for the list of source files needed to build reduced/stripped library.

Normalized make/model

There is a huge number of identical cameras sold under different names, depending on the market (e.g. multiple Panasonic or Canon models) and even some identical cameras sold under different brands
(Panasonic -> Leica, Sony -> Hasselblad).

To reduce clutter, a normalization mechanism has been implemented in LibRaw:

In imgdata.idata:

  • char normalized_make[64]; - primary vendor name (e.g. Panasonic for Leica re-branded cameras)
  • char normalized_model[64]; - primary camera model name
  • unsigned maker_index; - primary vendor name in indexed form (enum LibRaw_cameramaker_index, LIBRAW_CAMERAMAKER_* constant).

These fields are always filled upon LibRaw::open_file()/open_buffer() calls.

const char* LibRaw::cameramakeridx2maker(int index): converts maker_index to normalized_make.

We recommend that you use these normalized names in a variety of data tables (color profiles, etc.) to reduce the number of duplicate entries.

New vendor index values will be added strictly to the end of the LibRaw_cameramaker_index table, ensuring that the numbers assigned to vendors that are already known to LibRaw will not change.

DNG frame selection (and other changes)

DNG frames selection code re-worked:

  • by default all frames w/ the NewSubfileType tag equal to 0 (high-res image) are added to the list of available images (selection performed via imgdata.params.shot_select field, as usual)
  • the special case for Fuju SuperCCD (SamplesPerPixel == 2) works as before: shot_select=1 will extract second sub-image.
  • Additional flags to imgdata.params.raw_processing_options:
    • LIBRAW_PROCESSING_DNG_ADD_ENHANCED - will add Enhanced DNG frame (NewSubfileType == 16) to the list of available frames
    • LIBRAW_PROCESSING_DNG_ADD_PREVIEWS - will add previews (NewSubfileType == 1) to the list.
  • By default, DNG frames are not reordered and are available in same order as in DNG (LibRaw traverses IFD/Sub-IFD trees in deep-first order). To prioritize the largest image, set LIBRAW_PROCESSING_DNG_PREFER_LARGEST_IMAGE bit in imgdata.params.raw_processing_options.

DNG Stage2/Stage3 processing via DNG SDK (request via flags in raw_processing_options)

Imported code policy disclaimer

We've changed the policy regarding 3rd party code imported into LibRaw:

We (like other authors of open-source RAW parsers) gladly import support code for various RAW formats from other projects (if the license allows it).
This is done to expand camera support.
Unfortunately, not all imported code can tolerate truncated or otherwise damaged raw files, as well as arbitrary conditions or arbitrary data; not all authors handle rejecting unexpected input well.
LibRaw is now widely used in various projects, including ImageMagick, which, in turn, is often used on web sites to process any input images, including arbitrary data from unknown users. This opens up wide possibilities for exploiting the various vulnerabilities present in the code borrowed from other projects into LibRaw. In order to avoid such security risks, - the borrowed code will no longer compile by default.
We are not able to support it in general case, and the authors refuse to add code to reject unexpected input.
Thus, if you use some kind of camera for which the support is disabled by default, you need to recompile LibRaw for your specific case.

Formats currently affected:

  • X3F (Foveon) file format.
    Code is imported from Kalpanika X3F tools:
    To turn the support on, define USE_X3FTOOLS
  • Rapsberry Pi RAW+JPEG format.
    Code is imported from
    To turn the support on, define USE_6BY9RPI

Format support is indicated via LibRaw::capabilities() call with flags:
LIBRAW_CAPS_X3FTOOLS - Foveon support

GoPro .gpr format support

GoPro format supported via open-source GPR SDK See README.GoPro.txt for details.

Windows support/Windows unicode (wchar_t*) filenames support

  • (old) LibRaw's WIN32 external define split into 3 defines to fine tune compiler/api compatibility:
  • LIBRAW_WIN32_DLLDEFS - use to compile DLLs (__dllimport/__dllexport attributes)
  • LIBRAW_WIN32_UNICODEPATHS - indicates that runtime has calls/datatypes for wchar_t filenames
  • LIBRAW_WIN32_CALLS - use Win32 calls where appropriative (binary mode for files, LibRaw_windows_datastream, _snprintf instead of snprintf, etc).

If the (old) WIN32 macro is defined at compile time, all three new defines are defined in libraw.h
If not, these defines are defined based on compiler version/libc++ defines

LibRaw::open_file(wchar_t*) is always compiled in under Windows, but if LIBRAW_WIN32_UNICODEPATHS (see above) is not defined, this call will return LIBRAW_NOT_IMPLEMENTED.

Use (LibRaw::capabilities() & LIBRAW_CAPS_UNICODEPATHS) on runtime to check that this call was really implemented (or check for #ifdef LIBRAW_WIN32_UNICODEPATHS after #include <libraw.h>)

LibRaw*datastream simplified

  • tempbuffer_open, subfile_open are not used, so removed from LibRaw_abstract_datastream and derived classes.
  • jpeg_src() call implemented using ->read() call and own buffering (16k buffer).
  • buffering_off() call added. It should be used in derived classes to switch from buffered reads to unbuffered.

Minor/unsorted changes

  • new flag LIBRAW_WARN_DNGSDK_PROCESSED to indicate decoder used
  • LibRaw::open() call, max_buf_size special meaning:
    • 1 => open using bigfile_datastream
    • 2 => open using file_datastream
    • New processing flag LIBRAW_PROCESSING_PROVIDE_NONSTANDARD_WB If set (default is not), and when applicable, color.cam_mul[] and color.WB_Coeffs/WBCT_Coeffs will contain WB settings for a non-standard workflow. Right now only Sony DSC-F828 is affected: camera-recorded white balance can't be directly applied to raw data because WB is for RGB, while raw data is RGBE.
    • New processing flag: LIBRAW_PROCESSING_CAMERAWB_FALLBACK_TO_DAYLIGHT If set (default is not), LibRaw::dcraw_process() will fallback to daylight WB (excluding some very specific cases like Canon D30). This is how LibRaw 0.19 (and older) works. If not set: LibRaw::dcraw_process() will fallback to calculated auto WB if camera WB is requested, but appropriate white balance was not found in metadata.
    • Add support for zlib during configure
    • Fixed multiple problems found by OSS-Fuzz
    • Lots of changes in imgdata.makernotes (hope someone will document it)
    • DNG SDK could be used (if enabled) to unpack multi-image DNG files.
    • DNG whitelevel calculated via BitsPerSample if not set via tags.
    • DNG: support for LinearDNG w/ BlackLevelRepeat.. pattern
    • Generic Arri camera format replaced w/ list of specific camera models in supported cameras list.
    • new samples/rawtextdump sample: allows one to dump (small selection) of RAW data in text format.
    • samples/raw-identify:
      • +M/-M params (same as in dcraw_emu)
      • -L <file-w-filelist> parameter to get file list from a file
      • -m paramerer to use mmap'ed IO.
      • -t parameter for timing
    • samples/dcraw_emu: fixed +M handling
    • better support for Nikon Coolscan 16-bit NEF files.
    • Visual Studio project files: re-generated to .vcxproj (Visual Studio 2019), different intermediate folders for different sub-projects to allow 1-step rebuild.
    • imgdata.makernotes...cameraspecific: removed the vendor name prefix from variables.
    • Bayer images: ensure that even margins have the same COLOR() for both the active sensor area and the full sensor area.
    • raw processing flag bit LIBRAW_PROCESSING_CHECK_DNG_ILLUMINANT inverted and renamed to LIBRAW_PROCESSING_DONT_CHECK_DNG_ILLUMINANT. If not set, DNG illuminant will be checked.
    • New libraw_decoder_t flags:
      • LIBRAW_DECODER_FLATDATA - in-file data could be used as is (if byte order matches), e.g. via mmap()
      • LIBRAW_DECODER_FLAT_BG2_SWAPPED - special flag for Sony ARQ: indicates R-G-G2-B channel order in 4-color data
    • Camera-recorded image crop data is parsed into imgdata.sizes.raw_inset_crop structure:
      • ctop,cleft,cwidth,cheight - crop size.
      • aspect - LibRawImageAspects enum (3to2, 4to3, etc)
    • New define LIBRAW_NO_WINSOCK2 to not include winsock2.h on compile
    • Google changes cherry-picked (thanks to Jamie Pinheiro)
    • speedup: ppg interpolate: const loop invariant
    • Bugs fixed
      • Fixed several UBs found by OSS Fuzz
      • Fixed several problems found by other fuzzers.
      • Thumbnail size range check (CVE-2020-15503). Thanks to Jennifer Gehrke of Recurity Labs GmbH for problem report.
      • fixed possible overflows in canon and sigma makernotes parsers
      • fixed possible buffer overrun in crx (cr3) decoder
      • fixed memory leak in crx decoder (if compiled with LIBRAW_NO_CR3_MEMPOOL)
      • fixed possible overrun in Sony SRF and SR2 metadata parsers
      • Fixed typo in longitude (member of parsed GPS structure), update required for code that uses it.