--- ray/doc/notes/ReleaseNotes	2005/07/27 05:34:17	1.89
+++ ray/doc/notes/ReleaseNotes	2007/12/13 07:03:37	1.142
@@ -1491,3 +1491,158 @@ behavior so I documented it rather than changing it.
 =====================
 Version 3.8
 
+Fixed Russian roulette ray termination, which was doing the opposite of
+what it should have been doing, resulting in overestimated values below
+minimum ray sampling weight (serious).
+
+Changed mksource to use 98th percentile (rather than 99th) for threshold.
+
+(Above changes included in 3.7.1 patch release.)
+
+Eliminated writing of EXPOSURE= line in ra_tiff -r when source TIFF
+does not have STONITS set.
+
+Fixed bug in ambient calculation that caused black values for multiple
+bounces in some scenes.  Thanks to Mark Stock for creating and sharing
+the sophisticated test scene that demonstrated this problem.
+
+Fixed bug in antimatter using Russian Roulette sampling.
+
+(Above changes included in 3.7.2 patch release 22 Aug 2005.)
+
+Created fieldcomb script to combine alternate fields in a rendering
+sequence.  (Development generously sponsored by Iebele Abel.)
+
+Added missing preload of mesh objects for more efficient memory use
+during parallel rendering -- this was a serious oversight.
+
+Created ran2tiff program to automate exposure setting of animation
+sequences using a moving histogram in combination with pcond, and
+converting frames to TIFF.
+
+Eliminated SPEED macro from makeall and source tree.
+
+Added -pj option to vwrays to allow jittering of sample locations.
+
+Fixed bug in rtcontrib where it occasionally missed writing the
+final bin file.  (Pretty serious, actually.)
+
+Added -Dfseeko=fseek to MACH= line for linux, as fseeko(3) seems to be
+broken in current versions of Linux.  (Need to remove this later.)
+
+Rearranged struct's for better efficiency on 64-bit architectures.
+
+Added -i option to rlam so it works with binary input formats.
+
+Added rtcontrib -bn option to specify number of bins and fixed bug
+in recovery of files containing multiple values per record.
+
+Added rtcontrib -fo option to write over existing files, which it now
+refuses to do without this option.
+
+Made rtcontrib flush output when handed a zero-length ray vector,
+to match behavior of rtrace.
+
+Improved color scale for falsecolor output.
+
+Added threshold test for Fresnel approximation to prevent its use
+on smooth surfaces with unrealistically low specularities (less than 1.8%).
+Water has a specular reflectance of 2.1% at room temperture.
+
+Removed NICE macro settings in rpict and rtrace, which was slowing down
+the processes under Mac OS X (10.4) by about 60% due to processor napping.
+
+Made xform, replmarks and rad2mgf handle quoted modifiers and names properly.
+
+Fixed problem in genblinds, where it was generating slats that were too
+wide with the +/-r option.  (Thanks to Helena Buelow-Huebe for discovering
+this bug.)
+
+Fixed bug in ra_ppm for 16-bit/sample i/o, where it was getting the byte
+order wrong (apparently).
+
+Added -V option to rtcontrib to report actual contributions, rather than
+just ray coefficients.
+
+Fixed bug in ranimate that caused it to fail in continuation of renderings.
+
+Renamed librt.a to librtrad.a, to avoid conflict on Solaris. Almost
+everything now gets linked to -lrtrad instead of -lrt.
+
+Fixed problem with large sample sets, so -u+ option now is truly random.
+Thanks to Rick Mistrick and Younju Yoon for identifying this problem,
+which was related to a poorly designed hash function.
+
+Changed -u+ to be the default setting in rtrace (only).
+
+Fixed binary i/o for rlam and vwrays under Windows.
+
+Added tmCvLums() to tone-mapping routines using a 16-bit lookup table
+on IEEE floats, which is also used to improve speed of COLOR conversions,
+especially on machines with slow log() calls.  The first call to
+tmCvLums(), tmCvGrays(), or tmCvColors() initializes the table.
+
+Created optics2rad.csh script to convert from Optics 5 material output
+to something correct for Radiance.  Built on old glazing.cal function,
+which is actually incorrect so removed from the distribution.
+
+Fixed nuisance bug in ximage under some SGI machines that caused it to
+choose an inappropriate visual and display incorrect colors.  (Thanks
+to Oskar Itzinger for debugging the problem.)
+
+Fixed long-standing inaccuracy of pcompos -t and +t thresholding options.
+
+Fixed bug in handling of views with tabs rather than spaces.
+
+Fixed bug related to alias of alias.
+
+****** COMPATIBILITY CHANGE ********
+Changed default sky turbidity in gensky from the original default of
+2.75 to the CIE 110-1994 standard value of 2.45.  Thanks to Marilyne
+Anderson of MIT for pointing out this discrepancy.
+
+Added -o option to ra_xyze to produce original pixel values.
+
+Fixed bug in tracing of virtual sources, where intervening glazing was
+not being accounted for properly.
+
+Added -h option to pcompos and pcomb to eliminate verbose headers.
+
+=====================
+Version 3.9
+
+Added -spec option to falsecolor to restore original spectral color scale.
+Also added "-s auto" option to automatically set the maximum value.
+
+Added byte-swapping input and output to rcalc.
+
+Additional error-checking for source surfaces.
+
+Re-fixed bug in octree code that caused it to allocate 8 times as
+much memory as it actually used(!) -- this was pointed out way back
+in 1999 by Erik Reinhard and Erik Jansen, but the fix they gave me
+got lost in the slightly borken release process at the time.  This
+should make a big difference in memory usage, especially for the
+new mesh primitive.  (I measured 30% less memory usage.)
+
+Corrected ray weights for material mixtures.
+
+Reduced the default -n setting for obj2mesh to take better advantage
+of the smaller octree structure size.
+
+Fixed bug in shadow cache where it was adding surfaces with antimatter
+holes cut in them to the obstructor list.  Thanks to John Mardaljevic
+for discovering and reporting the problem.
+
+Adding ability in mkillum to incorporate BSDF data in calculation.
+Internal interface (raypcalls) now used in place of rtrace process(es).
+
+Fixed pixel alignment in pcomb for size reduction and enlargement.
+
+Allow rtcontrib user to set -P and -PP options for persistent rendering.
+The only caveat is that -n setting > 1 requires -PP rather than -P.
+
+Added -c option to rtcontrib to facilitate cumulative calculations.
+
+Added ability to incorporate LBNL Window 6 bidirectional scattering
+distribution function (BSDF) data in mkillum.