libvips/ChangeLog

16/1/10 started 7.21.2
- "invalidate" is careful to keep images alive, so invalidate callbacks can do
  im_close()
- flood_blob could loop if start point == ink
- added im_meta_remove()
- added remove profile option to vipsthumbnail
- added vips_bandfmt_iscomplex() and friends, im_iscomplex() and friends 
  deprecated
- im_bandjoin()/im_gbandjoin() work with images of varying formats
- added im_copy_native(), deprecated im_copy_from() and friends
- im_check*() name rationalisation
- finally removed old flood stuff
- im_insert*() bandalike and formatalike
- im_*join() bandalike and formatalike
- im_ri2c() bandalike 
- im_vips2png() saves 16-bit PNGs, if necessary
- vipsthumbnail has selectable interpolators, optional sharpen
- moved a lot of stuff (eg. im_iscomplex()) from deprecated to 
  almostdeprecated to avoid breakage
- im_csv2vips(): allow lines that end with EOF rather than \n
- im_vips2tiff has a bigtiff option

15/1/10 started 7.21.1
- added "written" callbacks, used to implement write to non-vips formats

26/11/09 started 7.21.0
- branch for new dev cycle
- argh, missing init from colour.c (thanks Peter)
- argh, im_measure() was not looping over bands correctly (thanks Peter)
- removed mmap_limit, we now always use windows ... reduces VM use hugely,
  because mmap windows are freed when their regions are freed, while images
  are only unmapped when they are closed
- have a min bytes for mmap windows as well, so we don't make too many tiny
  windows
- im_disp2Lab() was broken
- deprecated.h is now defined in terms of current functionality, rather than
  repeating stuff
- im_flood() and friends rewritten, typically 4x faster
- removed --with-cimg option, added --disable-cxx
- added im_system_image() (thanks Roland)
- added postclose callbacks
- added vipsthumbnail 
- oops, generate C++/Python wrappers for deprecated operations by default
- read TIFF images strip-wise, not scanline-wise
- better TIFF YCbCr reading (thanks Ole)
- isanalyze generates fewer silly error messages

26/11/09 started 7.20.3
- updated en_GB.po translation 
- file_length is gint64 to avoid win32 breakage

23/11/09 started 7.20.2
- GETTEXT_PACKAGE now includes lib version number (thanks Jay)

11/11/09 started 7.20.1
- oop, im_clip2fmt() was missing PTOP flag, should get a small speedup
- im_conv() / im_convf() didn't like all-zero masks
- small updates to im_convf() from im_conv()
- im_read_imask() produced an incorrect error message if passed a doublemask
- rename im_convf(), im_convsepf() as _f()
- vips.c drops _f suffix when overloading
- regenerate C++ binding, don't make deprecated package

9/11/09 started 7.20.0
- removed vips-7.x.spec.in, shouldn't really have this in SVN
- bumped version to 7.20
- fixes to get "make dist" working again

3/4/09 started 7.19.0
- version bump
- tiny conv speedup
- catch lcms error messages
- fix includes for gtk+-3.0
- report virtual memory too in im__print_all()
- cosmetic changes to nohalo
- im_magick2vips() needs to invert alpha
- now (more or less) passes -Wextra
- added "fail" option to im_jpeg2vips: fail with an error on any warning
  (thank you Ole)
- started gtk-doc changes
- renamed im_meta_get_type() and im_header_get_type() as 
  im_meta_get_typeof() and im_header_get_typeof() to prevent confusion with
  GObject type definers (was breaking gtkdoc object scan)
- revised more names, limited documented API
- im_buildlut() could segv for non-zero based tables (thanks Jack)
- VIPS_BUF_STATIC() does not take length arg
- check for SetImageOption() so we work with GraphicsMagick too
- "header" sets a non-zero exit code if anything failed
- add and use im_check_uncoded() and friends
- matlab load handles column-major and plane-separated images (thanks Mikhail)
- JPEG save allows "none" for profile, meaning don't attach a profile
- saner, simpler, faster typecasting for im_add(), im_subtract(),
  im_multiply(), im_divide(), im_remainder()
- im_remainder() has a float result for float types
- im_measure() allows sel == NULL, meaning all patches
- added "deprecated" package
- faster, simpler, better im_max(), im_min, im_avg(), im_deviate()
- im_max() returns true modulus, not square of modulus, for complex images
- im_avg() works for complex, returning average modulus
- im_system() fix (thanks Roland)
- im_system() rewrite
- im_maxpos()/im_minpos() are partial and work for complex
- im_max()/im_min() are now convenience functions
- im_maxpos_avg() handles complex and multi-band images
- added im_point(), rewrite im_point_bilinear() in terms of this
- close callbacks now happen *after* images have closed resources (such as
  open files) ... this lets them delete temps and stuff. Expect breakage :(
- added vips_interpolate_get_window_offset()
- boolean revised: smaller, more general, faster
- im_remainderconst_vec() renamed to im_remainder_vec() for consistency
- added im_shift*_vec()
- renamed im_eor_vec() as im_eorimage_vec() for consistency, also and, or
- renamed im_eorconst() as im_eorimage_const() for consistency, also and, or
- relational revised: smaller, more general, faster
- im_blend()/im_ifthenelse() allows many-band conditional, 1-band then/else
- im_blend()/im_ifthenelse() allows band and format to differ between then 
  and else parts
- better im_check() functions
- added im_flood_other() as start of simple segmentation operator
- added im_label_regions() 
- im_printlines(), im_debugim() deprecated (use im_vips2csv() instead)
- meta, header, callback, error, region, check, generate, memory gtkdocs
- removed printlines tool, vips2csv is much better
- removed other useless tools as well: debugim, binfile
- fix up addr calcs on 64-bit machines with >2gb images and inplace ops
  (thanks Christoph)
- im_generate() checks that im_demand_hint() has been called for this image
- im_jpeg2vips.c, set scale_num on shrink (thanks Guido)
- heh argh reading history always stopped after the first line (thanks Haida)
- added im_histindexed
- new im_iterate() calls start and stop functions from workers so resources 
  they make are owned by the worker thread ... this makes it possible to have
  start functions which create mutiple regions and therefore allows
  im_iterate() to scan more than one image at once
- threadgroup no longer has any default action, you must attach a work
  function
- added im_copy_file()
- added im_insertset()
- im_insertplace() allows small to be outside big
- added im__colour_difference(), colour ops now work on any image format
- added im_col_display_get_table(), so display tables are now shared by name
- added im__colour_unary()
- drop "set" postfix from names, so "insert" can now take a vector of positions
- deprecate all the "_raw" variants, not really necessary now
- removed "contrib", not very useful anymore
- added im_header_as_string()
- im_malloc()/im_free() now call g_try_malloc()/g_free() ... removes confusion 
  over whether to use im_free() or g_free() for things like im_header_string()
- added im_history_get(), im_getexp(), im_printdesc() as wrapped functions ...
  so you no longer need the "header" program
- image vectors from Python work, woo

25/3/09 started 7.18.0
- revised version numbers
- updated vipsmanual
- revised manpages
- removed name and "changed" from vipsobject since we don't use them yet
- explicitly link with stdc++ for nohalo etc. stuff
- wrap im_gauss_imask_sep in C++/Python

6/3/09 started 7.17.3
- revised nohalo
- remove fading stuff from im_render() -- cleaner and simpler
- configure spots support for "restrict"
- reset dcm:display-range on magick read to help DICOM
- saner im_buildlut() behaviour
- added im_gauss_imask_sep()
- allow open and view of truncated images (thanks Joe & Rachel)
- revising rounding on im_affine*() coordinate transforms to make them more
  stable
- added Radiance write
- added im_float2rad()
- added IM_CODING_RAD, support where it makes sense (extract, flip, rotate,
  etc.)
- IM_PROGRESS env var
- docs for rad2float and IM_CODING_RAD

3/3/09 started 7.17.2
- im_magick2vips.c: allow funky bit depths, like 14 (thanks Mikkel)
- isradiance was returning TRUE too often
- radiance loader now loads packed RGBE/XYZE instead of unpacking to float
- added im_rad2float()

11/10/08 started 7.17.0
- merge vips-7.16 brach back into trunk
- bumped version number to 7.17.0
- re-added type.[hc]
- added vipsinterpolate and im_affinei
- added yafrsmooth interpolation
- added yafrtest
- added yafrnohalo
- ubuntu 8.10 changes
- interpolators get an output pointer, not region
- tuning for bicubic
- revised transform / clip code, we now do corner not centre
- yafr-smooth reworked along the lines of bicubic
- cleanups after yafr hacking
- added affinei_all
- don't set im_error() on failed callback
- moved im_format_t to VipsFormat, now sits over VipsObject
- IM_FORMAT_FLAG_PARTIAL -> VIPS_FORMAT_PARTIAL
- updated docs
- interpolators use type introspection
- added vips --list classes, does formats too
- include sys/param.h to get PATH_MAX in more places
- added vips_format_get_flags()
- oop, forgot to check for cancel during tiled tiff write
- don't use mmap for tiff read: no performance advantage, chews up VM
- VIPS_INTERPOLATE_SHIFT bumped to 12, we need the extra precision for u16 gel
  data
- added string->double transform for cmdline args
- merged class-params branch back into trunk
- IM_FREE() can do "const char*" variables
- im_buf_t renamed as VipsBuf
- added vips_object_to_string()
- added "nickname" and "description" properties to VipsObject
- shift/and/or/eor ops were broken for non-int types
- added nohalo interpolator
- updated format docs
- IM_INPUT_INTERPOLATE() now used by affinei and affinei_all
- added vips_object_new
- resamplers/interpolators now in a resample package
- removed yafrnohalo.c
- added matio as a dependency
- added Matlab save file read
- added Radiance file read
- better file-not-found messages

11/9/08 started 7.16.3
- oop typo in manpage for im_project()
- doc fixes
- returning non-zero from eval callbacks was not always stopping computation

11/9/08 started 7.16.2
- added --without-v4l option
- added -no-undefined to libsrcCC build, so we get a libvipsCC.dll
- removed the swig dependency: we include the generated bindings in the
  distribution tarball

6/9/08 started 7.16.1
- trigger eval callbacks on tiled tiff write
- added vips as an im_format_t
- added im_format_write()/_read() convenience functions
- more cleanups for the format API, argh
- removed win32/ directory now mingw finally makes DLLs correctly
- removed the windowed spcor for now, it has some edge effects

24/8/08 branch for 7.16
- renames and version numbers
- load plgs from libdir as well as libdir/vips-x.x for compat
- complex -> complex conversion was broken
- refstring <-> gstring transforms
- better behaviour with Magick non-presence
- added --enable-links switch to configure ... we no longer make the bin/im_*
  links by default
- started a 'format' section in the docs
- configure fails if no gettext found
- revised po/
- released as 7.16.0!

25/5/08 fork for loadable image format branch
- image load/save in non-vips format code moved to own dir
- simple format searching added
- some cleanups for vips load
- im_open() simplified
- add im_format_flags
- only consider formats with a save method in im_format_for_name()
- oops, format sort order was reversed
- im_filename_suffix() includes "." in suffix
- merge back into trunk for 7.15.1
- remove im_ispng(), im_png2vips_header() etc. & friends
- add "vips --list formats"
- rename VBuf as im_buf_t for consistency
- add type.[hc], start of new type system
- removed man pages for IM_MIN, MAX and RINT to avoid case confusion on 
  OS X / win

7/3/08 started 7.15.0
- MAGIC constants should be tagged as unsigned
- write MAGIC correctly on sparc/powerpc machines (thanks Joe)
- oop, we were still making fade threads even when not fading
- tiny cond jump fixes for valgrind in colour.c
- remove -lstdc++ from libs, except on windows 
- push Magick cflags earlier in the include order to make it easier to pick
  GraphicsMagick over ImageMagick (thanks Mikhail)
- fix the en_GB translation
- use meta to preserve resunit between tiff load and save
- small doc improvements
- read and write CMYKA tiff (thanks Doron)
- performance improvements for morphology ops, esp. when zooming out
- oop, im_render() was broken for mask == NULL
- better support for multiple Python installs (thanks Jay)
- better IM_SETSTR() stops some warnings
- im_histcum() works for signed histograms
- better rounding for im_conv(), im_convsep()
- tiny speedup for im_conv()
- better /0 test for remainderconst
- revise i18n configure, get rid of intltool 
- command-line IMAGVEC input could segv with non-vips image output
- added .tobuffer()/.frombuffer(), .tostring()/.fromstring() to Python binding
- add PIL_mode_from_vips () and vips_from_PIL_mode () utility functions
- update docs for new Python stuff
- FIND_ macros no longer search for stuff, you have to specify prefixes if the
  packages are not on the default path (or in $prefix). This avoids some
  accidents on some platforms
- configure prints a summary of optional packages found at the end
- im_lhisteq() checks for window too small
- added invalidate callbacks
- now tests for MagickWand before ImageMagick (thanks Adam) 
- added "-rotate" option to vips2dj
- added man page for im_resize_linear
- better jpeg-in-tiff YCbCr read (thanks Ole)
- oops, invalidatefns were not being freed on im__close()
- VMask() can init from std::vector, so Python can init from []
- added IM_LIBDIR, im_guess_libdir()
- load plugins from libdir/vips-x.x on startup
- added meta get/set int/double/string/area/blob/GValue to C++ API
- include time_t in vips.h, thanks nicola
- lock global image list (thanks lee)

25/1/08 started 7.14.0
- bump all version numbers for new stable
- better CMYK JPEG read (thanks Ole)
- add __str__ to VError in Python (thanks Ole)
- revert the dynamic wrapping for Python :-( next version!
- added VImage::convert2disc (thanks Ole)
- you can now set the jpeg quality factor for tiff pyramids (thanks Joe)
- you can now shrink jpegs during read, see "man im_jpeg2vips"
- added CMYK JPEG write
- optionally use GraphicsMagick (thanks Bob Friesenhahn)
- look for MAGICKCORE_HDRI_SUPPORT (thanks Marcel)
- set icc profiles in tiff pyramids explicitly (thanks Joe)
- add --without-cimg configure option
- add im_maxpos_subpel 
- make im_abs compile without liboil 
- add im_align_bands 
- fix type overflow in complex division 
- fix im_cross_phase and im_addgnoise (Tom)
- updated docs, C++ and python
- header no longer stops on error
- C++ dummy2.cpp helps OS X linking

12/12/07 started 7.13.3
- added "include <cstring>" to VImage.cc to help gcc 4.3
- started moving the python binding to dynamic wrapping
- added im_wrap(), im_wraptwo(), im_phasecor_fft(), im_cross_phase() (Tom)
- memleak plugged in im_save_string_setf()
- bugfix in and docs for im_gradcor() (Tom)
- use Glib macros to make im_msb work on big-endian platforms (Tom)
- use Glib macros to get rid of needless compile warnings (Tom)
- fix type overflow in im_c2ps/im_abs (Tom)

31/10/07 started 7.13.2
- build cimg on windows fixes
- various include cleanups, updated man pages
- break im_wbuffer() out to a separate API
- use im_wbuffer() to make im_vips2jpeg() compress in the background
- also im_vips2png(), im_vips2tiff(), im_vips2ppm()
- revised evaluation progress system
- new evalstart/evalend/preclose callbacks fix over/underflow reporting
- but the meaning of evalend has changed in a non-backwards-compatible way :(
  use preclose instead ito get the old behaviour
- added "--vips-progress" flag to turn on a simple eval progress tracker
- make im_spcor[12] static, im_spcor is wrapper (Tom)

28/9/07 started 7.13.1
- vips2dj can print RGB images
- oop, include <stdexcept> missing
- add protos for generate/iterate function args, fix warnings
- add cimg package, we now have C++ source inside VIPS, sigh
- added OUTPUT_DOUBLEVEC and OUTPUT_INTVEC, use for im_maxpos_vec() and friends

29/8/07 started 7.13.0
- we now have a trunk and the version is 7.13.x, woo!
- move manpages into a separate man/ dir ... speeds up builds a lot on windows
- don't install malkovich
- don't fail on unknown args for python (thanks Simon)

1/8/07 started 7.12.5
- im_embed() is more general ... x and y can be negative
- predicate.c is smaller and cleaner
- libsrcCC link improvement from Pablo
- support 32/64-bit ImageMagick as well (thanks Marcel)
- better im_magick2vips() for Q8 ImageMagick
- split repository to trunk/branches ready for a stable 7.12.x branch

27/7/07 started 7.12.4
- proto.h had vars called small, breaking on win32
- more python fixing, we now have working matricies too

17/7/07 started 7.12.3
- fix to VImage.i for gcc 4.2 (thanks Damir)
- eek, off by 1 for more than 1 band hists
- needed a slightly larger worst-case buffer in im__b64_encode()
- tiny cleanup for make_hI() prevents cond jump on ui in valgrind
- --disable-threads was broken again
- remove .svn dirs from dist
- now passes distcheck again

17/7/07 started 7.12.2
- added im_bandmean()
- added support for TIFFTAG_PREDICTOR (Andrey Kiselev)
- fix TIFFOpen() mode snafu (Andrey Kiselev)

11/5/07 started 7.12.1
- memory.c abort()s with DEBUG
- oops, im_bits_of_fmt() manpage was not being installed
- im_histcum() can do all image types
- updated NEWS
- added im_csv2vips_header() 
- command-line csv read was broken (thanks Tom)
- removed length limit on argument vectors (Tom)
- added IM_PREFIX, configure-time install prefix
- oop, turned off memory.c DEBUG
- fix some bogus gcc 4.1 warnings with im_open_local_array()
- better vips usage message
- oops, IM_ANY missing from im_demand_hint() manpage (thanks Shahid)
- just warn if plugins fail to load in im_init_world()
- expose Vargv and make refblock public rather than private so that 
  subclasses of VImage can add vips-style member operations (thanks Pablo)
- oops, im_initdesc() needed to have bbits set correctly (thanks Shahid)
- make VError derive from std::exception more officially
- woo, got exceptions working in SWIG
- soname version bumped to 12.x.x
- oops, added vector ops to Python
- check for overflow of int in return of strtol in dispatch system (Tom)
- add im_[di]mask_[xy]size to dispatch system (Tom)
- add im_gradcor(), im_grad_[xy] (Tom)

26/3/07 started 7.11.21
- ooo, added %include "std_except.i" & friends to VError.i, VImage.i
- im_init_world() is more lenient about recursive invocation
- im_gbandjoin() falls back to im_copy() for 1 input image
- race condition fixed in im_render.c (thanks Simon)
- bump for 7.12!!!

26/1/07 started 7.11.20
- another fix to im_region_image() (thanks Mikkel)
- tiny speed up to im_rect_includesrect()
- avoid recursive invocation in im_init_world() (thanks Christian)
- fix to extract_prefix (thanks Christian)
- buffer cache is now per thread
- combine buffer unref and ref in a single operation to reduce malloc/free
  cycles
- new internal API for passing regions between threads means we can remove
  buffer locks
- more buffer/region sanity checks, plus a memory barrier
- lock around error buffer changes
- im_vips2mask() was wrong for nx1 m-band images
- liboil back to "test"
- add buffer_cache_list to avoid GHashTable _insert()s
- oop, --vips-concurrency was broken
- renamed (in dispatch layer) im_and/or/eor_const -> im_and/or/eorimageconst 
  for consistency 
- C++ API wraps IMAGEVEC, DOUBLEVEC, INTVEC arguments
- oop, IMAGE % vector<double> was broken

21/12/06 started 7.11.19
- added im_linreg() (Tom)
- various C++ API polishes, plus a bugfix (Dennis Lubert)
- vips.spec split to devel and python too (Dennis Lubert)
- be more explicit about sizeof(magic)
- init magic to native order by default (thanks Dennis)
- Hist becomes im_history_get() 
- new history mechanism is faster, uses much less memory, and removes 
  duplicate lines
- added im_get_option_group()
- added official im_concurrency_set()/_get()()
- don't read bbits from vips files ... set ourselves from bandfmt
- oops add RGB16 and GREY16 to C++ header
- --list packages option to vips.c
- updated docs to 7.12
- oops, im_region_image() snafu was causing a lot of recomputation
- make im_mpercent() suck a little less
- EXIF save was a bit bOrked ... cause of mac crashes?
- im_histgr(), im_heq(), im_hist() all number bands from zero now
- fix stride in liboil calls
- set RGB16 on 16-bit RGB ICC export

29/11/06 started 7.11.18
- added im_buffer_t so regions can share calculated pixels: 2-3x speedup on 
  the benchmark
- im_region_local() -> im_region_buffer()
- im_sharpen() order change to help sharing
- im_invalidate() clears buffer caches
- add sentinel attributes
- add some missing im_demand_hint()s
- paint ops invalidate the output image
- fix nothread eval
- raise threads limit to 1024 (thanks Christian)
- manual redone
- vipsCC python init() hooks 
- add liboil dependency
- use liboil for im_abs(), im_add(), im_divide(), im_floor(), im_multiply(),
  im_subtract(), im_lintra(), im_avg(),im_deviate()
- quiet libtoolize test (thanks Tom)
- im_benchmarkn now regrows image each time
- strip meta from sample2.v ... saves a lot of mem (esp. Hist)
- added im_isscalar() (Tom)
- added IM_REGION_ADDR_TOPLEFT() (Tom)
- reduce size of im_rightshift_size.c to help compile (Tom)
- make im_stop_many(NULL) safe (Tom)

24/11/06 started 7.11.17
1;5Q
- better benchmark script makes graphing easier
- double-buffer image file writes
- reuse write threads
- clean up threadgroup / iterate / generate
- added im_benchmarkn to make it easier to make a CPU-bound op on large
  machines
- im_cache() failed for cpus > 1

1/11/06 started 7.11.16
- moved im__convert_saveable() into im_copy() (thanks Christian)
- missing gobject dependency (thanks Christian)
- --enable-threads was broken (thanks Christian)
- eval without theads was broken (thanks Christian)
- LIBADD libvips.la to libvipsCC.la (thanks Simon)
- benchmark.sh is now plain sh, not bash
- set ORIENTATION_TOPLEFT in im_vips2tiff (thanks Josef)
- oops, im_vips2csv() output separator was broken
- added im_benchmark2
- move XYZ2Lab LUT build outside the eval thread

30/10/06 started 7.11.15
- print leaked windows
- oops, race condition in im_window_unref()
- integrated im_region_window() into im_region_image(), tiny speed up

6/10/06 started 7.11.14
- ifthenelse and affine dhints revised
- buildlut no longer outputs x cods
- configure asks for glib >= 2.6 (we need GOption)
- configure uses AC_TOOL_CHECK to find tool names to help cross-compiling.
- better configure test for libexif
- add </libexif/ prefix if required
- test for glibtoolize during bootstrap to help OS X
- add --version to vips main prog
- don't set Type in im_scale, want to be able to scale hists, for example
- im_vips2mask() allows multiband hists as well
- im_vips2csv() lets you specify an output separator
- vips.c cleanups in arg handling
- removed python binding to C layer, too hard to get it working right
- leak test python test program
- added throw() decls to C++ to help SWIG
- no longer derive VError from std::exception, can't get it to work with SWIG
- added mmap window manager to share windows between regions ... big reduction
  in VM use for large workspaces and for SMP
- add im_spcor2 (with sliding window) (Tom)

15/9/06 started 7.11.12
- python vips wrapper hooks into init to start vips
- oop, python detect was broken again
- im_file_length() error return was wrong (thanks Christian)
- im_csv2vips() isnored last column if there was no trailing
  whitespace/quoting
- test for NULL filename/mode, common if you forget to check argc
  (thanks Bruno)
- im_break_token() returns NULL on trailing whitespace ... helps read_dmask
  work with trailing whitespace on matrix headers
- added im_buildlut(3)
- removed extra im_free() in im_copy() fallback in im_msb()
- added GValue input/output args
- various C++ binding improvements (thanks Dennis): namespace support, 
  VError inherits from std::exception, small fixups
- remove im_msb() warning on fallback to copy for uchar ... not very serious,
  happens a lot in nip2
- added <vips/vips> C++ include ... include in a namespace
- added im_benchmark / SMP benchmark script
- add im_maxpos_avg() and im_point_bilinear() (Tom)
- make im_region_free(NULL) safe (Tom)
- link in manpages for im_contrast_surface (Tom)

8/9/06 started 7.11.11
- add im_norm_dmask()
- removed old code for gradient and lindetect
- internal decls split from proto.h to help SWIG
- test for python and SWIG during configure
- added python dir for the binding
- python binding done!
- oops, --without-python was broken (thanks Tom)
- added python/test
- add im_lu_decomp() im_lu_solve(), rewrite NR type functions in terms of these (Tom)

23/6/06 started 7.11.10
- still more im_affine() rounding/clipping tweaks
- ignore "--" arguments to vips.c
- im_init_world() also sets g_*_prgname() and loads plugins
- add manpage for im_init_world() (oops)
- error_exit() prints prgname
- various cygwin fixes
- fix cache thread assert failure (thanks Joe)
- "header" now uses GOption, slightly different args, will loop over args
- fixed assert() overenthusiasm in im_prepare()
- im_csv2vips() now has separate whitespace / separator tables
- add im_rightshift_size() (Tom)
- add im_maxpos_vec(), im_minpos_vec() (Tom)
- add im_norm_dmask() (Tom)
- make im_free_[di]mask(NULL) safe (Tom)

23/6/06 started 7.11.9
- back on sourceforge CVS again
- require openexr 1.2.2 or greater 
- range check xy on im_insert*() for sanity
- VMask::invertlut decl removed (thanks Jean)
- added \"all\" option to vips.c

17/5/06 started 7.11.8
- debrokened openexr read
- added im_tile_cache()
- added tiled read to im_exr2vips()
- im_tiff2vips() now uses im_tile_cache() rather than its own cache ...
  faster in some cases, less RAM use in some cases, saves 200+ lines
- removed 'broken' read option from im_tiff2vips()
- read/write doubles with g_ascii_strtod() and friends where appropriate
- add a "thread" member to region to help sanity check region ownership
- saner threadgroup fixes a race problem on gcc 4.0.3 / amd64
- added im_vips2csv()
- im_open() now does CSV read/write too
- oops, broke vips main prog for function name in argv1 case

22/4/06 started 7.11.7
- split vips_png.c to im_vips2png.c and im_png2vips.c
- added OpenEXR dependency
- added im_exr2vips(), im_exr2vips_header()
- added im_isexr(), im_open() knows about OpenEXR
- added im_contrast_surface(), im_contrast_surface_raw() (Tom)
- added im_msb(), im_msb_band() (Tom)
- im_scale() sets Type on output
- added RGB16, GREY16 types
- im_*2vips() set these types if appropriate
- configure fixes for mac
- vips main prog uses GOption
- im_icc_* locks around calls to cmsDoTransform() to avoid corruption on SMP
  machines
- add im_prepare_many() (Tom)

10/3/06 started 7.11.6
- typo in manpage and header for im_rect_dup() (Tom)
- don't abort image load if XML read fails
- added im_video_test() ... test video source
- oops, lcms .pc finder was not working
- clipping problem in im_affine() fixed (thanks Clare)
- test for attr support in libmagick
- im_text() returns an error for empty string 
- im_falsecolour() scale reversed
- im_remosaic() could crash on bad mosaics
- configure changes to fix --without-magick, lcms and fftw (but sadly we now
  require .pc files for these libs)
- im_vips2jpeg() automatically converts to 1 or 3 band sRGB uchar for write
- also im_vips2png()
- added im_project()

20/2/06 started 7.11.5
- added im_csv2vips()
- commas in filename options can be escaped with a '\'
- raise tile buffer limit (thanks Ruven)
- im_spcor() and im_fastcor() have prettier borders
- im_fastcor() returns sum of squares of differences, not sum of abs of 
  differences

18/11/05 started 7.11.4
- small win32 fixes, thanks Juan
- added im_flood_blob_copy() ... a temporary hack
- much faster im_histplot()
- read RGBA palette-ized PNG images more robustly (thanks Tom)
- turn on -ms-bitfields automatically for mingw

26/9/05 started 7.11.3
- better error recovery for im_binfile() file too large
- all raw files now use mmap windows, so (eg.) ppm and analyze reads can go
  >2GB
- remove DISABLE_COMPAT ... you now have to define IM_ENABLE_DEPRECATED to get
  broken #defines
- fix to build without exif, thanks Chas
- use native win32 API for seek()/truncate() to work with large files
- use attribute to check printf-style args with gcc
- fix gcc4 warnings
- removed ebuild, since it's in gentoo now
- im_magick2vips() sets meta from attributes (good for dicom)
- im_magick2vips() writes many-frame images as tall thin VIPS images
- im_histcum() was broken for vertical histograms
- im_histnorm() is neater
- simpler and faster inner loop for im_conv() and im_convf() avoids gcc4 bug 
- appendc() was reading past the end of the buffer on MSB machines

13/6/05 started 7.11.2
- im_copy_set() was messed up in 7.11.1
- put into CVS, phew
- fixed a rounding bug in im_affine() ... should no longer get black edges on
  image resize
- if TIFF open fails in im_open(), try going through libmagick
- merge requires all bands == 0 for transparency (used to just check 1st
  band)
- 16 bit LAB TIFF read/write was wrong
- new GType for refstring makes it visible from im_header_map()
- jpeg loader attaches exif data (and human-readable meta fields)
- jpeg saver writes any exif data 
- meta not wiped by im_*incheck() in a "w" image
- meta keeps traverse order
- now require glib >= 2.4
- require libxml-2.0 for meta save and new history mechanism
- no more .desc files, history saved in XML after pixel data
- i/s/d meta fields saved there too
- added base64 encode/decode
- added blob header write
- added a save string type: types which define transforms to and from the
  save format get serialized
- GValue meta API now exposed, since we can serialise anything
- jpeg loader loads ICC profiles
- jpeg saver saves ICC profiles from the VIPS header
- src/header.c knows about meta system
- added im_analyze2vips(), im_grid(), im_raw2vips()
- extract/grid/replicate/zoom were not setting TRANSFORM flag
- better falsecolour LUT
- less stupid + more portable read/write of VIPS header
- better im_updatehist()
- jpeg load sets vips xres/yres from exif, if possible
- jpeg save sets exif xres/yres from vips, if possible
- icc_export and icc_transform  attach profiles and intents to output images
- added im_icc_import_embedded() to import with an embedded profile
- split vips_jpeg.c into two, it was getting too big
- added im_cp_descv(), im_cp_desc_array(), funcs use them
- removed im_append_Hist() from API
- fixed meta copy bug
- better history copy, removed nonsense about 1st line of Hist being special
- tiff read/write now reads/writes ICC profile from meta
- edvips rewritten to remove stupidness, and can now set xml
- header can now print xml extension block
- IM_ prefix for colour temp names

1/6/05 started 7.11
- added im_copy_morph()
- im_region_region() allows Bands and BandFmt to differ, provided
  sizeof( pel ) is the same ... makes im_copy_morph() work
- added im_meta*() functions (MW)
- im_header_*() rewritten for meta 
- added im_header_exists(), im_header_map()
- use pkg-config to find libpng and ImageMagick
- added im_lineset()
- added im_extract_areabands() (JF)
- added im_copy_from() (JF)

15/4/05 started 7.10.12
- im_ifthenelse just evals left/right for region all zero/all one
- also im_blend
- swap g_setenv() back to plain setenv() so we work with glib 2.2

9/4/05 JC started 7.10.11
- docs no longer have broken links
- fixed memleak in im_text()

8/4/05
- one bit tiff read was sometimes reading a byte too far in each scanline

14/1/05 started 7.10.9
- im_filename_split() will now usually work for filenames containing ':'
  characters
- added im_render_fade() for fancier nip2 image repaint
- added "ccittfax4" as a TIFF compression option
- fix all 64-bit compiler warnings
- "," allowed as column separator in mask read
- better at spotting singular matricies
- small im_render() tidies
- glib dependancy reduced to just 2.0, but untested ... helps people building
  on older systems who aren't interested in nip2
- removing leading spaces from IMAGEVEC arguments
- load non-interlaced PNGs more efficiently
- 1 point mosaic functions work on more image types
- better memory allocation debugging info
- local memory on regions can shrink as well as grow
- shut down threadgroups on render if no dirty tiles
- limit number of simultaneous renders
- higher mmap window threshold 
- allow max == -1 for unlimited render cache
- 'priority' marks non-suspendable renders
- im_embed() mode == 4 paints white pels
- im_tiff2vips() was broken with --disable-threads
- oops, im_errormsg() compat macros were GCC only
- larger default tile size and strip height
- tiff write sets PHOTOMETRIC_CIELAB for vips TYPE_LAB images ... so we can 
  write float LAB as well as float RGB (thanks Ruven)
- also 16 bit LAB TIFF write
- im_render() rewritten

20/11/04 started 7.10.8
- im_sharpen() is ~15% faster
- more quoting for MAGICK finder
- im_XYZ2Lab() uses a LUT rather than cbrt(), 5x faster
- --disable-threads removes gthread dependency completely (thanks Simon)
- intercept TIFF warnings as well as errors ... stops occasional libMagick
  exceptions
- add im_init_world() to im_init() as well to help backwards compat (thanks 
  Simon)
- im_icc_present() function description was broken, thanks Jay
- oops, libtool library versioning was wrong, thanks Jay
- can now make TIFF pyramids of any non-complex image type (was uchar and LAB
  only), thanks Ruven
- 1st order mosaic code now works for LABQ too
- build system changes to make "make distcheck" work
- RPM .spec files fixed up and updated by configure (thanks Simon)
- tiny cleanups for vdump
- use g_setenv()/g_getenv()
- tiny improvements to IM_FREE*()
- tiny VImage debug print fixes (thanks Jay)
- swap off_t for gint64 to fix LARGEFILE support on win32
- computation feedback now uses gint64 for number of pels, so we give feedback
  correctly on images with >2**31 pels
- other small fixes for >2**31 pels in an image

10/11/04 started 7.10.7
- im_histnD() was not checking BandFmt (thanks Kirk)
- improvements to threading system speed up non-vips output in some cases
- use cbrt(x) where we can ... 10x faster than pow(x,1.0/3) on win32
- typeo in im_text() when built without PANGOFT2 (thanks Stefan)

1/11/04 styarted 7.10.6
- tiny doc fixes
- scripts now only depend on 'vips' program
- im_open( "r" ) is now lazier ... for non-VIPS formats, we delay read until 
  the first ->generate()
- so im_open_header() now deprecated since im_open("r") is identical
- now looks for fftw3 as well as fftw2 ... slightly faster ffts

19/10/04 started 7.10.5
- fix to light_correct (thanks Jay)
- edvips knows about xoffset/yoffset
- better vips enum<->char conversions

4/10/04 started 7.10.4
- man page fixes (thanks Jay)
- removed last csh scripts (thanks Jay)
- scripts default VIPSHOME to $prefix (thanks Jay)
- doc build system tidies
- im_rank() edge padding was wrong
- im_vips2tiff() can now embed ICC profiles

22/9/04 started 7.10.3
- mildly better im_vips2tiff()
- *, -, +, /, % between two images now work for mixed number of bands
- im_free() was missing a man page
- revised documentation

1/9/04 started 7.10.2
- C++ .pc files were still set for 7.9, grr
- im_insertplace() didn't check compatibility of images (thanks Matt)

27/7/04 started 7.10.1
- set default stack size explicitly to help platforms with a very low default
- 16 bit RGB tiff read was broken (bug introduced in 7.9.5, thanks Haida)
- !pangoft2 was broken, thanks Kenneth
- win32 build fixes

12/7/04 renamed as 7.10.0
- added NOCACHE function flag ... stops nip memoising video & paint ops
- added im_extract_bands() ... takes out many bands from an image
- im_vips2tiff() scanline write speed up for area pipelines

10/6/04 started 7.9.6
- tiny polishing of im_ppm2vips()
- im_blend() can now work on labq
- boolean ops all work on float/complex images (by casting to int)
- im_maplut() was broken for 1 band image + many band lut + >8 bit output
- im_lintra_vec() now handles 1 band image and many band vector to make many
  band image
- oops, im_lintra_vec() was missing a man page
- im_measure() can work on labq
- im_lhisteq() uses new embed mode, _raw() version is one pixel smaller, sets
  Xoffset/Yoffset for new origin scheme
- generalised im_tone_build() to any image type to make im_tone_build_range()

20/5/04 started 7.9.5
- tiff output res can be a single number too 
- added im_text() to make a bitmap from a string with pango
- im_tiff2vips() does 16 bit RGBA
- im_binfile() was broken since 7.9.2 due to im_mapfile() change
- im_ppm2vips() now works for 16 bit images
- added im_copy_swap() ... copies, reversing byte order
- im_resize_linear() was broken for some images, thanks Javi

8/3/04 started 7.9.4
- oops, config.h include missing in a few places
- im_vips2tiff() can now write 1 bit images
- im__find_lr/tboverlap() now exported to nip
- better edge tile handling for tiff read/write (thanks Ruven)
- added extend-pixels-from-edge mode to im_embed()
- im_conv*(), im_rank(), im_stdif(), im_dilate(), im_erode() all use it to
  expand their input, so their output now has guess borders, not black borders
- im_fastcor() now does an im_embed( 1 ) on the output ... the zero borders
  were very annoying before, since you would usually be searching for the
  minimum point
- no change to im_spcor(), since you will usually be searching for the maximum
- better im_render() cache behaviour under heavy loads
- im_affine() revised 
	* clip, resample and transform is now pixel-perfect for all 
	  inputs (I hope) 
	* uses the new embed to make sure there are no black borders 
	  from edge interpolation
	* about 20 - 30% faster
- policy change: Xoffset and Yoffset are now set by all operations to record
  the position of the input origin in the output
- im_replicate() is much faster for some cases
- added tile and mirror flags to im_embed()
- added im_cache() convenience function
- better ETA for image calculation
- im_tiff2vips() now has a "broken" option so it can read tiled tiffs made
  with earlier versions of vips
- on convert float to int format, now does floor() not rint() ... more
  'mathematical'
- added im_rint()
- im_sharpen() now uses a gaussian mask
- im_convsep() more resistant to int overflow problems
- added im_make_xy(), avoids rounding problems with the old float-based thing
- im_profile() now makes vertical images for a vertical profile
- added im_vips2tiff() option to set the resolution in inches not cm (thanks
  Andrey)
- im_binfile() is now exported 

6/2/04 started 7.9.3
- added an im_init_world() to im_open(), to help old progs
- renamed VSemaphore as im_semaphore_t
- started using libtool library versioning
- now uses g_module_*() in place of dlopen()
- now uses pkg-config instead of vips-config (thanks Simon)
- fixes to vips.h for _ADDR() with DEBUG on (thanks Konrad)

10/12/03 started 7.9.2
- patches for freebsd, thanks Lev Serebryakov
- vips2dj knows about my colour laser printer
- added i18n support, glib/gmodule/gthread dependency
- im_error*() API revised to be more i18n friendly
- List type removed, now uses g_slist
- VBuf added, some more utility funcs pushed down from nip
- im_thread stuff removed, now uses g_thread
- im_lock stuff removed, now uses g_mutex
- im_semaphore_t renamed to VSemaphore, not sure this is a good idea
- build with --disable-threads to turn off threaded render 
- #include <vips/vips.h> now pulls in most of the public API, you shouldn't
  need other vips includes very often
- im_close() is better at cleaning up if there's an error
- inverse FFTs could fail for wider-than-high images with fftw
- better im_icc_transform error messages
- bug fix in im_render with large caches
- im_binfile() now has an offset parameter
- im_mapfile()/im_unmapfile() now work on IMAGE and record the length of the
  file they mapped ... this lets VIPS successfully unmap a file if it changes
  size while it's open

20/10/03 started 7.9.1
- threadgroups now have their own kill flag
- im_plotmask() now does anti-aliasing
- im_iterate() fix for operations on mmap window images (thanks Clare)
- im_writeline() stops on kill
- fix for im_fwfft() segv for wider-than-high real images (thanks Andrey)
- fix for im_fwfft() to work for non-square real images (thanks Andrey)
- can now read and write 32-bit IEEE float TIFF (Andrey Kiselev)
- clean-ups for colour.c (Andrey Kiselev)
- no longer lets you make an image with width|height|bands == 0 (thanks Joe)
- im_vips2tiff(), im_vips2*jpeg*(), im_vips2png(), im_vips2ppm() could
  sometimes fail for mmap window input images (thanks David)
- added IM_RECT_HCENTRE(), IM_RECT_VCENTRE() macros

20/8/03 JC
- started 7.9.0
- added im_rank_image() ... im_maxvalue() a special case of this
- im_subtract() goes up to int earlier for better value preserving, thanks
  Haida
- im_rank() much faster for large windows, correct result on all platforms 
  (dratted memcpy() was causing problems before)
- fixed problem with libMagick config if installed somewhere strange
- fixed problem with include order in library compile 
- added --without-magick configure option
- added im_render(), threaded background image paint
- added im_replicate(), replicate an image horizontally and vertically

31/5/03 JC
- started 7.8.11
- fixed a problem with relational operators and some combinations of input
  types (bug introduced in 7.8.9), thanks Haida
- vips-7.8 script overrides VIPSHOME environment variable
- better im_guess_prefix
- stupid light_correct script no longer uses /pics/tmp
- added batch_crop script

22/5/03 
- started 7.8.10
- the JPEG writer can embed ICC profiles in output images ... although I've
  yet to see it make any difference :-( test this carefully at some point
- fixed a possible coredumper in jpeg write
- jpeg read now spots truncated files
- im_invertlut() now makes an image, not a mask ... sorry :-( 
- im_histnD() makes an n-dimensional histogram from an n-band image
- im_col_pythagoras() patch
- IM_NUMBER() now returns int not size_t
- new win32 build system from Juan and friends, based on tmake
- sample project files for MSVC added, thanks Juan
- win32/ subdir now has the win32 build systems
- spec/ subdir now has the spec files for building RPMs
- dist now includes formatted documentation
- license change: VIPS is now LGPL, nip stays GPL ... this means proprietary
  programs can link against the vips library 
- had a report of a working VIPS build on a 64 bit system (!)
- im_log_dmask() now includes all of the negative lobe, thanks matt
- vips-7.8 start script now auto-relocates
- im_spcor_raw(), im_fastcor_raw() now exported

29/4/03
- started 7.8.9
- changes to build to help MSVC 
- oops, makedef.pl missed out function names with an initial cap, and
  error_exit()
- im_min() and im_max() gave random wrong results for >1 thread on >1 CPU
  machines (bug introduced in 7.7.20), thanks Joe
- vips.c no longer generates C++ wrappers for functions with no image argument
  (thanks Haida)
- im_invertlut() now wrapped by hand in VMask.cc
- C++ docs updated
- added im_open_header(), returns an IMAGE with just width/height/etc and no
  data
- ... so now "header" will print useful stuff even on truncated files
- tiff writer knows about alpha (thanks Jenny)

7/2/03 
- started 7.8.8
- build failed with lcms turned off
- im_spcor() could segv for 16bit images (thanks Joe)
- im_tiff2vips() read resolution expressed as pixels/cm incorrectly
- im_vips2tiff() tries not to write mad resolutions
- header and im_open file type tests reordered for slight speedup
- im_copy_set() had a broken dispatch function for xres/yres
- im_fwfft() exploits libfftw real -> complex transform if possible for a 2x
  speed-up (thanks Matt)
- im_invfftr() added for complex -> real inverse transform for 2x speed-up
  (thanks Matt)
- im_freqflt() now uses im_invfftr() for real result and speedup
- im_flipver() could segv on some inputs, thanks Clare
- relational operators now work on complex
- relational rewritten ... now fractionally slower, but 1/3 the size
- vips2dj -1:1 produced incorrect height
- better overlap-too-small detection in mosaicing code
- im_system() can have NULL output
- global balance ignores overlaps with only transparent pixels

3/1/03
- started 7.8.7
- worked in patch from Hans Breuer (thanks!)
	- png read/write with im_png2vips(), im_png2vips_header(), 
	  im_vips2png(), im_ispng()
	- im_errorstring() and im_col_displays() are now functions not externs
	  (helps DLLs)
	- many include fixes to help native win32 build
- added libMagick support, 78 file formats now loadable with
  im_magick2vips(), im_magick2vips_header() and im_ismagick(), w00t
- now installs vips.m4 to $prefix/share/aclocal
- added im_icc_export_depth() ... export to device space with a specified bit
  depth (8 or 16)
- vips.def now rebuilt with custom rule in libsrc/Makefile.am
- removed externs im_Type, im_BandFmt, im_Coding, im_Compression to simplify
  DLL build
- im_mmap() -> im__mmap(), since it's supposed to be an internal function
- new vips-7.8.x/proj directory holds unsupported sample makefiles and 
  config.h for building with the MSC toolchain 
- new scripts batch_image_convert and batch_rubber_sheet (thanks Joe)
- added the RPM .spec files to the main distribution
- InitializeMagic() now passed "" rather than NULL to avoid assert() problems 
  on some libMagic versions

2/12/02
- started 7.8.6
- now reads 8-bit RGBA tiff
- C++ build guide fixes (thanks fsicre)
- im_Type2char array text slightly messed up
- global_balance is safer for complex mixed mosaics
- removed im_lintra() fallback to im_copy() for scale == 1, offset == 0 ...
  too confusing
- im_tiff2vips() now reads 16-bit LAB
- added im_Lab2LabS() and im_LabS2Lab()

5/11/02
- started 7.8.5
- fix for mmap window of local region ... caused im_iterate() to break
  sometimes for large images, in turn occasionally breaking 
  im_max()/im_min()/etc. (thanks Joe)
- tiny speed up for im_rot90()/270()
- on install on win32, add .exe suffix for links
- vips.c knows to remove .exe suffix for linked commands
- added im_errormsg_system() ... decode win32 error codes too
- pagesize calcs for roving mmap windows were messed up on win32 (thanks Kirk)
- some TODO cleanups
- global balance broke horribly if you had filenames with spaces in (thanks
  Clare)

31/10/02
- started 7.8.4
- im_unmapfile() includes mixed up on mac os x
- libtool patched for mac os x
- vips.c sets numeric locale to "C"

27/10/02
- started 7.8.3
- configure fixes help mac os x
- im_guess_prefix() adds ".exe" suffix on w32 if not there
- changed im_measure() error messages to number bands from 1
- added func descriptor for im_read_dmask() to help nip, updated C++ API, docs

21/10/02 JC
- started 7.8.2
- tries rand() if random() is not available
- tries mktemp() if mkstemp() is not available
- turns off realpath() if not available
- added IM_DIR_SEP/IM_DIR_SEP_STR directory separator character/string
- added IM_PATH_SEP/IM_PATH_SEP_STR path separator character/string
- added im_path_is_absolute()
- vips.c knows to link to vips.exe on win32
- spot mingw* and set BINARY_OPEN
- open images in binary too (since we now read() the header)

10/10/02 JC
- im_lintra() and im_lintra_vec() were broken for complex images :-( thanks
  matt
- renamed im_and() as im_andimage(), im_eor() as im_eorimage() and im_or() as
  im_orimage() ... avoids breakage in the C++ layer
- added im_dE00_fromLab()
- limited release as vips-7.8.0

2/10/02
- renamed as vips-7.8, woohoo
- revised documentation

19/9/02 JC
- started sorting out VIPS #defines ... there are now a sensible set of new
  names (eg. NOCODING becomes IM_CODING_NONE, LAB becomes IM_TYPE_LAB)
- define IM_NO_VIPS7_COMPAT to turn off the old names
- added im_mmap()/im_munmap() layer for windows portability
- removed the contents of history.h .. obsolete
- added IM_IMAGE_ADDR() macro

10/9/02 JC
- handle errors from TIFF lib correctly
- configure fixes for cygwin
- CMYK TIFF write fixed
- configure fixes for mingw

5/9/02 JC
- im_cp_desc() now copies Xoffset/Yoffset

21/8/02 JC
- started 7.7.24
- reads CMYK TIFF
- reads dpi from TIFFs
- better float Xres/Yres

14/8/02 JC
- new header fields Xoffset and Yoffset ... used by functions to hint 
  the position of the origin in output images
- support added to c++ api and to header
- im__lrmerge(), im__tbmerge(), im__affine(), im_insert(),
  set Xoffset/Yoffset
- now uses <lcms.h>, not <lcms/lcms.h> for better suse w0rkage
- better configure for fftw (uses libdfftw name if libfftw not found)

8/8/02 JC
- large file support with mmap() windows ... had to change 
  im_prepare_inplace() to im_prepare_to()

  benchmark:

  - system

	hardware: 2 x 2.5GHz P4, 1GB RAM, 15k SCSI, ReiserFS
	os: suse 8 (kernel 2.4.18)
	compiler: gcc 2.95.3, -O2, threads turned on
	images: fred.v, fred2.v; both 4k by 4k LABPACK (64MB)
	images: jim.v, jim2.v; both 15k by 15k LABPACK (900MB)
	time: smallest real of 5 runs, system idle
	vips: 7.7.23, debug on in im_openin.c, window limit set with an
	  environment variable

  - benchmarks

	cpu-bound: im_sharpen fred.v fred3.v 11 1.5 20 50 1 2
	io-bound: im_insert fred.v fred2.v fred3.v 4000 0
	worst-case: im_rot90 fred.v fred3.v 

  - results

	desktop:

			no mmap windows		mmap windows
	
	cpu-bound	real    0m3.712s	real    0m3.970s
			user    0m6.010s	user    0m6.390s
			sys     0m0.900s 	sys     0m1.110s

	io-bound	real    0m1.813s	real    0m1.865s
			user    0m0.900s	user    0m0.990s
			sys     0m1.720s 	sys     0m1.520s

	worst-case	real    0m1.344s	real    0m3.039s
			user    0m1.270s	user    0m2.230s
			sys     0m0.850s	sys     0m3.050s

  not quite sure why sharpen is a little slower (4%?) ... IO speed is about
  the same though ... worst-case is having to constantly move windows about
  (500,000 page faults, vs 10,000 for no windows)

  again, with an image larger than RAM 

  			no mmap windows		mmap windows

	io-bound 	real    2m52.759s	real    2m11.172s
			user    0m14.940s	user    0m14.890s
			sys     0m29.940s	sys     0m26.560s

	worst-case	real    3m35.391s	real    3m50.760s
			user    0m19.850s	user    0m26.600s
			sys     0m12.650s 	sys     0m43.130s

  mmap windows actually slightly faster in this case ... plus they stress the
  OS less

31/7/02 JC
- added -lm for better lcms detect
- README notes for fftw on suse8
- im_profile() sets HISTOGRAM for output image
- im_copy()/im_copy_set() function descriptor no longer sets PTOP ... helps 
  avoid LUT problems
- im_subsample()/im_zoom() fall back to im_copy() for shrink/grow == 1
- im_lintra() falls back to im_copy() for scale == 1, offset == 0
- no longer use Type == LUT ... all just Type == HISTOGRAM now
- im_blend() was messed up for > 1 band images :(

16/7/02 JC
- started 7.7.23
- im_XYZ2sRGB() wasn't setting Type = sRGB
- im_icc_import() was broken for rgb
- im_header_string() had wrong return type in function database

13/7/02 JC
- added im_flood_blob() 
- added im_open_local_array() ... C API convenience function
- oop, im_flood() was missing a man page
- Type == FOURIER added to help visualisation
- released as 7.7.22

30/6/02 JC
- JPEG, TIFF and PPM import all now set sRGB Type for RGB import
- im_header_int(), im_header_double() and im_header_string() added to aid
  UIs
- now uses gettimeofday(), not time()
- for consistency with other trig functions, im_c2amph() now returns degrees 
  not radians (ouch)
- added im_c2rect() ... turn (amp, phase) to rectangular
- added im_sign() ... unit vector in direction of value
- better im_scaleps() ... old code was terrible
- rewritten im_rotquad() ... now partial
- im_icc_export()/_import() now do ABSOLUTE correctly
- added im_icc_ac2rc() ... converts absolute to relative colorimetry 

25/6/02 JC
- added im_copy_set(3) ... like im_copy(), but set informational header fields

20/6/02 JC
- added im_ceil(), im_floor()
- im_Lab2LabQ was not clipping a/b range correctly
- im_icc_export(), own ABSOLUTE mode
- released as 7.7.21

28/5/02 JC
- im_remainderconst_vec broken for float/double
- added Yxy colourspace

16/05/02 JC
- auug, libtool was all messed up ... redone all the autotools stuff 
- uses libtool convenience libraries to build vips in sections
- uses config subdir for temp files and .m4 things
- patched stupid suse config.guess
- vips2dj patched for better raw cmyk
- released as 7.7.20

12/5/02 JC
- im_vips2jpeg*() and im_vips2ppm() now both partial
- started updating the C++ guide
- had to change the location of the C++ headers :-( all C++ progs should now
  have:

	#include <vips/vipscpp.h>

  this is so things can work on systems which do not have case sensitive
  file systems
 
- changes for Mac OS X 
  * im_system() TRUE/FALSE removed
  * searches /*/[lib|include] to get fink libs for tiff and jpeg

30/4/02 JC
- several functions were missing IM_FN_PIO in their descriptor ... this was
  harmless for nip/ip/C, but broke the ref counting in the C++ layer 
- im_system() now defaults "/tmp" for temp files
- STRING input and output args were broken for C++ :-(
- threads exit more quickly on error
- im_min()/im_max() now partial (at last)
- im_remainderconst()/im_remainderconst_vec() added
- --with-dmalloc configure switch
- vips2dj does CMYK and mono too
- im_vips2tiff() allows any number of bands (but not the right way to
  write CMYK, see TODO)

26/4/02 JC
- old ICC profile reader removed
- little cms wrapped ... configure spots it, im_icc_transform() uses it to map
  between two images
- also im_icc_import() and im_icc_export() so you can see PCS images
- im_icc_present() to test for existence of lib
- README fixes

4/4/02 JC
- TODO changes
- oops, DEBUG left on in im_invertlut()

2/4/02 JC
- im_fwfft.c/im_invfft.c now use libfftw if available ... about 5x speed up
  and double precision
- added FIND_FFTW autoconf macro
- include/vips/proto.h changes

26/3/02 JC
- started 7.7.19

25/3/02 JC
- im_log_dmask() was broken (thanks matt)
- casts between VDMask and VIMask were broken (thanks matt)
- various error msgs improvements and tiny man page fixes

13/3/02 JC
- tb/lr merge first/last cache moved to per-call state for better sharing
- im_remosaic() bails out faster on error and makes better error messages

13/3/02 ruven
- im_vips2tiff() pyramids stop at tilesize, not 64x64

25/02/02 JC
- im_remosaic() is smarter, and works better with im_global_balance()
- im_affine() 2x faster

14/2/02 JC
- started 7.7.18
- vips.m4 and libsrc/Makefile.am fixes for IRIX

11/02/02 JC
- vips/thread.h and vips/threadgroup.h were missing extern "C" for C++ 
- VImage::write() now tracks dependencies, so you can write() to a partial
  safely ... although it's not a very useful thing to do (thanks Mike)
- new VImage::print() function for debugging
- added im_print()

22/01/02 JC
- started 7.7.17

15/01/02 JC
- im_rect_unionrect() and im_rect_intersectrect() safer for repeated args
- im_video_v4l() no longer perrror()s on ioctl fail for less spam

03/01/02 JC
- started 7.7.16
- im_version_string() really does return the date as well now

12/12/01 JC
- im_guess_prefix() extra smartness for relative path names
- VImage() no longer uses tmpnam() (thanks Paul)

11/12/01 JC
- renamed im_fexists() as im_existsf()

7/12/01 JC
- ppm man pages added (doh)

28/11/01 JC
- warnings on g++ 2.96 fixed

22/11/01 JC
- started 7.7.15
- im_video_v4l() failed to compile on non-linux platforms

7/11/01 JC
- im_remosaic() added
- im_*merge() are more intelligent about transparency in bizarre overlaps
- grr! putenv() semantics change on more recent clibs ... should be safer
  now

19/10/01 JC
- VDisplay( "display name" ) segved on unknown display :-( thanks mike

26/9/01 JC
- contrib tools get data files from share/vips/xxx area now
- im_vipshome() renamed as im_guess_prefix(), reworked for new package layout
- doc/ build sorted out
- ... but of course, docs still need updating for 7.8

20/9/01 JC
- fix to im_vipshome()
- ip2 renamed as nip
- split to library only ... separate ip and nip packages
- new VIPS_VERSION_* macros set from configure.in in vips/version.h.in
- vips.m4 VIPS finder
- reworked README, doc/README and TODO
- now installs to /usr/local/ by default
- fmask4th.c was including varargs.h ... d'oh
- include area reorganised: everything inside <vips/..> now ... hopefully the
  only user-visible change is that all plain C progs need to change:
  	
	#include <vips.h>

  to:

	#include <vips/vips.h>

  the C++ API should be unaltered

21/8/01 ruven
- im_setupout() was missing some #includes

20/8/01 JC
- started 7.7.14

15/8/01 JC
- added libxml dependency for ip2

27/7/01 JC
- im_conv(), im_convf(), im_convsep(), im_convsepf() now reject masks with 
  scale == 0

26/7/01 JC
- started 7.7.12

25/7/01 JC
- started 7.7.11
- oop, im_histeq() and im_tonemap() also missed
- better error messages from im_run_command()

23/7/01 JC
- started 7.7.10
- im_sharpen() failed due to change in im_band_extract() offset

20/7/01 JC
- started 7.7.9

4/7/01 JC
- im_open(,"w") open() delayed until im_setupout(), very slightly safer
- updated im_open() man page
- im_tiff2vips() now embeds index in filename ... and it's page number (from
  0), not subsample factor
- finally bit the bullet ... im_extract()/im_extract_band() now number from 
  zero (sorry!)
- and im_lrmosaic()/im_tbmosaic() bandno param too

29/6/01 JC
- im_region_free() now frees immediately

27/6/01 JC
- im_vips2tiff() man page updated for deflate, 2 years late

22/6/01 JC
- oops, limit wrong on im_rank()

21/6/01 JC
- better post_install for --prefix outside VIPS's tree
- -ltiff needs -lm in acinclude.m4 ... fixes configure on redhat 7.x

13/6/01 JC
- started 7.7.8

6/6/01 JC
- im_invertlut() added 

31/5/01 JC
- im_colour_temperature, im_XYZ2Lab_temp, im_Lab2XYZ_temp added
- ... colour temp stuff needs sorting out properly

25/5/01 JC
- added vips-config script, cf. gtk-config
- --without-threads option added
- did a bit of work on the C++ API docs

24/5/01 JC
- added im_tiff2vips_header(), im_jpeg2vips_header() and im_ppm2vips_header()
- header uses these to print fields quickly
- switched to config.h
- configure.in rewritten ... much nicer, fewer options, more automatic

17/5/01 JC
- im_matinv() didn't free stuff correctly on singular matrix

16/5/01 JC
- vips2dj now knows about 5000ps printers
- allow RW mode for non-native VIPS image files, for 8 bit images

2/5/01 JC
- started 7.7.7

1/5/01 JC
- im_addgnoise() did not work for >1 band images

23/4/01 JC
- configure options to remove support for JPEG and TIFF ... helpful for a no-
  dependencies build

20/4/01 JC
- im_(v)snprintf() added
- all sprintf()s removed

15/4/01 JC
- im_affine() had a rounding problem

11/4/01 JC
- tiny mosaicing bug fixed in im__lrcalcon
- started 7.7.6

21/3/01 JC
- new iblend code in im_tbmerge() was typo-d 
- mosaic1 was broken by affine too

20/3/01 JC
- im_image() failed for FMTUCHAR

12/3/01 JC
- started 7.7.5
- im_sharpen() uses seperable convolution for big speed up
- new "Print" menu

11/3/01 JC
- REALVEC renamed as DOUBLEVEC
- added IMAGEVEC
- added IM_INPUT_IMAGEVEC
- gbandjoin now has function description
- new function im_maxvalue()
- im_compass()/im_lindetect() reimplemented with im_conv()/im_maxvalue(), 
  about 15% faster, works for any type, partial
- im_gradient() reimplemented with im_conv()/im_abs()/im_add(), about 30%
  slower, works for any type, partial

10/3/01 JC
- new function, im_clip2fmt() converts between any image formats ...
  slightly faster than the old im_clip()
- legacy im_clip2us() etc. functions now just call this

9/3/01 JC
- im_conv() rewritten, simpler, about 10% faster
- im_convsep() rewritten, now does any non-complex type, partial, 20% faster
- new functions: im_convf(), im_convsepf() for DOUBLEMASK
- raw versions of each
- legacy convolvers (eg. im_convbi()) removed

8/3/01 JC
- new function im_blend()
- new function im_lab_morph()
- speed up to im_ifthenelse()
- speed up to im_*merge() (uses integer arithmetic for integer blends)

4/3/01 JC
- tiny speed ups to im_histgr() 
- speed ups to im_maplut() 

3/3/01 JC
- new functions: im_histnorm(), im_histcum()
- im_histeq() more general
- im_vipshome() --- better behaviour for relative paths

2/3/01 JC
- new video package
- im_video_v4l1() (video for linux) added
- configure.in switches to turn v4l1 on and off

1/3/01 JC
- new im_histspec() implementation ... more general, bugs removed

14/2/01 JC
- better vips2dj usage message

13/2/01 JC
- im_image_sanity() added, called in various places in iofuncs

9/2/01 JC
- added 'check' and 'name' class member stuff to ip from ip_gtk2
- new "Plot" menu
- new "Overlay" menu
- more stuff in _stdenv/_list ... curried forms of head/tail etc.

7/2/01 JC
- started 7.7.4
- vips2dj and vdump now use im_vipshome()

5/2/01 JC
- new im_vipshome() function
- min()/max() macros renamed as MIN()/MAX()
- new im_load_plugins() function
- vips.exe and ip now load $VIPSHOME/lib plugins at startup

2/2/01 JC
- mosaicing functions now have an extra max blend width parameter

30/1/01 JC
- fixed tbmerge no overlap detect

13/12/00 JC
- started 7.7.3

30/12/00 JC
- vips.h fixes for cygwin/wingdi conflict

27/11/00 JC
- added im_vips2ppm(), im_open() imports and exports it
- fixed nasty implicit output conversion problem for PIO dispatch() calls

21/11/00 JC
- added im_ppm2vips()

16/11/00 JC
- configure.in fixes ... jpeg found correctly now
- searches for libz as well

16/11/00 JC
- started 7.7.2

5/11/00 JC
- speed up to lab2labq

19/10/00 JC
- started 7.7.1

13/1/00 JC
- oops, im_open() was missing an 'else' in jpeg/tiff load

5/8/00 JC
- im_vips2tiff() now has mode string embedded in output filename
- im_vips2jpeg() now has qfac in output filename
- im_open() understands this

11/7/00 JC
- new im_image() function ... wraps a VIPS image around a memory buffer
- C++ layer changes:
	* now use #include <vips/Vips.h>
	* error renamed as VError
	* new VMask() constructors
	* new VImage() constructor
	* new VImage::data() access member
	* more operator equivalences: <, >, <=, >=, ==, !=, &, |, ^, %

17/6/00 JC
- more consts added to vips protos

10/5/00 JC
- minor configure.in changes to help solaris
- removed _TIFFmalloc() and _TIFFfree() calls

9/3/00 JC
- fixed rounding problem in generate grid

8/3/00 JC
- fixup to im_system() temp dir

7/3/00 JC
- added im_system()
- small tidies

1/3/00 JC
- better plugin test in configure.in

26/2/00 JC
- fixes to Makefile.am in ip/src* and configure.in, to help ip find the right
  gtk includes on systems with more than one gtk-xxx installed

21/2/00 JC
- now builds ip and ip_gtk2

16/2/00 JC
- configure.in fixes for xil and some TIFF/JPEG strangeness

15/2/00 JC
- im_global_balancef() was broken! d'oh
- note in README about enabling video cards

10/2/00 JC
- configure now searches for xil ... FIND_XIL macro
- im_zoom() spots integer overflow
- better shape set on region drag end in ip
- better zoom-too-far handling in ip

13/1/00 JC
- fixes to configure etc. to help cygwin
- jpeg is now searched for too ... FIND_JPEG macro

10/1/00 JC
- global_balance() now uses new affine() atuff

27/12/99 JC
- mosaic1 stuff now uses new affine() funcs

21/12/99 JC
- added Joe's docs
- im_LabS2LabQ() rounding on a/b slightly broken for a/b == 0

20/12/99 JC
- new function: im_affine()
- similarity*() now in terms of affine()
- tests for error return in reduce.c from maplut
- small clean-ups

15/12/99
- im_version() added, new iofuncs package for it
- ip did not call zero-input-arg vips functions 
- vips.c did not like zero-input-arg functions

6/12/99 JC
- Sobel filter was a bit broken

3/12/99 JC
- menu reorganisation
- generate gauss mask dialog

2/12/99 JC
- reworked text file IO, better error messages

1/12/99 JC
- reworked .iprc filename stuff, cleaner

30/11/99 JC
- better menu set switcher in calc preferences
- ip now thinks it's 7.7
- did a spellcheck on the ip guide

29/11/99 JC
- 7.7 started!
- srgb D65->D50 converter
- Negate added to arith
- find similar pixel value dialog
- find similar colour dialog
- paste into background dialog

19/11/99 JC
- adjust labq was mising
- shrink image defaulted to /2 not /1
- sharpdropshadow broken
- Rotate.* broken
- vips-7.6.3 release

18/11/99 JC
- ip didn't report space free on >2GB filesystems correctly
- ip did not link statically against libXpm on solaris7

17/11/99 JC
- im_histplot() failed for all 0's histogram
- new profile_image dialog
- man pages for im_sRGB2XYZ/im_XYZ2sRGB were broken
- new colourize image dialog
- new shrink image image dialog
- new expand image image dialog
- better doc Makefiles

16/11/99 JC
- more menu reorganising
- new image_to_mask/mask_to_image buttons
- new match_two_images dialog
- new measure_colour_chart dialog

15/11/99 JC
- new custom LABQ sharpen in Image menu
- configure support for SGI video (thanks Ruven)
- menu fiddling in ip
- new "resize canvas" menu item
- im_insert_noexpand() added, no docs tho'

11/11/99 JC
- new ip menu item: Image=>Adjust white/black/saturation of LabQ 
- fixed bug in decompose complex number
- browse-icons now ignores errors
- better error msg for "12 12" etc cases

10/11/99 JC
- open hi-res gives proper error msg if no file found
- made file-select boxes a bit more compact and clearer
- updated configure.in for vips-examples-7.6

9/11/99 JC
- new column start point moved
- vdump now defaults to subsample 1, portrait
- navigation boxes were broken
- no longer shrink-to-fit if loading as high res
- better positioning of zoom windows
- removed some old cruft

8/11/99 JC
- ooops, im_remainder() got lost somehow
- refguide converted to latex and updated 
- cppguide updated

5/11/99 JC
- libguide converted to latex and updated

4/11/99 JC
- ip guide now latex2htmls cleanly ... still needs updating tho'
- cpp/app guide latex2html redone

2/11/99 JC
- configure no longer adds -32 to IRIX builds for you
- some more stuff in README
- simple DEBUGM malloc tracking, made libsrc/iofuncs/memory.c
- changed im_malloc() to return void *
- removed duplicate im_malloc() proto from util.h
- changed all malloc/free to go through im_malloc()/im_free()

23/10/99 JC
- double-click on error image now pops a load browser and an error dialog
- add-new-column no longer scrolls to right edge of workspace
- better scroll-to-bottom on item add
- load ws twice does not cause 'already open' errors
- memorise directory button in fsb

18/10/99 JC
- new FIND_TIFF/FIND_MOTIF macros for acinclude.m4
- new ./configure switches, see ./configure --help
- enabled static libs

8/10/99
- broke action_proc_bop() into smaller functions, stops bad code gen on 
  gcc2.95.1 (and others, prolly)
- restored old Makefiles in doc/src/ipguide
- vips7.6 script renamed as vips-7.6

5/10/99
- replace image was broken
- some menu reorganisation
- 'reload all menus' button

4/10/99
- computed regions in ip were broken
- updated system.iprc defaults
- new menu item: adjust white/black points

2/10/99
- browse icons was broken
- greyscale 16-bit tiled tiff was broken
- extra mutex locks for TIFF*() in im_tiff2vips()

1/10/99
- some automake probs fixed
- squished two ip bugs

30/9/99
- fixed problem with pthreads, now works on suse6.2 as well
- new API stuff for threaded evaluation with im_threadgroup_t
- new public interface provide platform independent threads/locks/semaphores
- no error box if you zoom out too far now

26/9/99
- lr/tb merge blend was not quite right ... should be smoother now
- histplot broken for float images

24/9/99
- better mono->labq converter
- more portable ispoweroftwo detect for freq filter stuff

23/9/99
- better graphics expose handling

17/9/99
- >/< stuff in ip was a bit mixed up
- ink preview fixed for mono images
- help popup fixed

15/9/99
- linedetect and sobel filters for ip, thnx Kirk

14/9/99
- Find_histogram was broken
- im_profile() man pages was broken
- ooops, ip had old set of macros

10/9/99
- im_and/im_or/im_eor now work for any integer type

Summer hols: (2nd half August '99)
- initial heap block larger to avoid start gcs
- def slicer fixed
- larger max heap
- toolkits with initial '_' hidden by default
- custom recomb
- dialog.def removed
- ... other menu fixes
- generate grid menu
- im_remainder() added
- new cursor change code, hglass rotates during comp!
- dialog.c handles cursor changes better
- rubber band in paintbox displays
- mag widget stays on right!
- undo/redo single pixel paint ops
- heap size control from prefs
- cancel for reductions as well as for image calc
- browse stuff reworked, no more .icon.v files
- save stops you overwriting open .v files
- better animate_countdown() handling
- better resize behaviour for bars added to images
- better code generation, bug fixed in state tracking
- auto recover from crash