"Array Subscript Out Of Range" Exception Can Be Thrown Under A Floating Point Rounding Edge Case.

[SimonCSmith]

I'm getting an "array subscript out of range" exception (Windows, Debug Build, OCIO 2.5.0) inside of reach_m_from_table due to a hue of 360.0 being passed in. This causes a i_lo=361, and i_hi=362. Given the ACES2::Table1D rt is only 362 in size, you get the overrun access scenario.

The reason you can get a hue of 360.0 is due to floating point rounding issues. These step from the calculations in the fwd call.

The RGB values coming into Renderer_ACES_OutputTransform20::fwd

0.742242277
0.0931933373
0.321542144

ACES2::RGB_to_Aab converts to

0.659631252
103.849297
-2.28881836e-05

Then ACES2::Aab_to_JMh converts it on to

62.3068
103.850
360.000

The atan2() in the Aab_to_JMh returns -2.20398064e-07, which then results in -1.26278792e-05 going into the _wrap_to_hue_limit call, and adding that to 360.000000 leaves it at a value of 360.000000.

Note that the random data obtained from the buffer over-read is technically immaterial (as it's multiplied by 0 in the lerp call) but it does mean you get an exception firing off due to Windows _STL_VERIFY macro which halts execution in the debugger.

For context, the processor I had at the time of the exception had a source of OCIO::ROLE_SCENE_LINEAR, and display/view of "sRGB - Display" and "ACES 2.0 - SDR 100 nits (Rec.709)" from the ACES2.0cg configuration (cg-config-v4.0.0_aces-v2.0_ocio-v2.5)

• Simon.