When converting an image with highlight recovery (without blending) and automatic white balance with libraw 0.13.5, we see major artifacts being introduced in areas.

Testing with the same version of dcraw shows no problem; it sounds like highlight recovery incorrectly occurs at the wrong position in the pipeline.

Testcase:

$ wget http://quora.org/2011/highlight-testcase.nef
$ dcraw_emu -w -H 2 highlight-testcase.nef
$ eog highlight-test.nef.ppm

$ dcraw -w -H 2 highlight-test.nef
$ eog highlight-test.ppm