When using EinsteinInitialData/TwoPunctures with CCTK_REAL set to single-precision memory-access errors and incorrect results were encountered. The incorrect results could potentially affect double-precision code too.
The memory access errors occur due to double / CCTK_REAL confusion.
The incorrect results are due to rounding errors in PunctTaylorExpandAtArbitPosition that push an acos argument outside of [-1,1]. Though this was encountered with single-precision code, perhaps it could occur with DP code too.
The attached patch fixes the type mismatch and clamps the acos argument to [-1,1].