msw@ and pkasting@ also mentioned the possibility of interpolation search.  It's a good idea if we can find the right guessing method.  Inverting the function to calculate directly would still be ideal, though, if possible.

We should also simplify the API from three functions to one.  Not only is this simpler to use, it guarantees consistent behavior.  All callers probably want to use GetResultingPaintColor(), but FindBlendValueWithMinimumContrast() doesn't(); and the other two functions don't let you retrieve exact alpha.

I think the right API is roughly:

struct BlendResult {
  SkColor color;
  SkAlpha alpha;
};

// Alpha-blends |default_foreground| toward |high_contrast_foreground| until the
// result has a contrast ratio against |background| of at least
// |contrast_ratio|.  Returns both the blended color and the alpha used to
// achieve that blend, since different callers want one or the other and it's
// more efficient to compute them both at once.  If |default_foreground| already
// has sufficient contrast, returns {default_foreground, 0}.
BlendResult GetBlendWithMinimumContrast(
    SkColor background,
    SkColor default_foreground,
    SkColor high_contrast_foreground =
        BlendTowardOppositeLuma(base, SK_AlphaOPAQUE),
    float contrast_ratio = kMinimumReadableContrastRatio);

And the implementation would use GetResultingPaintColor() first, ensure that we only run one contrast check in the case where there's sufficient contrast already, and compute exact alpha.

***UI Mass Triage***

Adding appropriate label for expert review.