https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118499
--- Comment #16 from kargls at comcast dot net --- On 1/17/25 03:47, tkoenig at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118499 > > --- Comment #15 from Thomas Koenig <tkoenig at gcc dot gnu.org> --- > (In reply to kargls from comment #14) >> (In reply to anlauf from comment #13) >>> (In reply to kargls from comment #12) >>>> (In reply to Thomas Koenig from comment #9) >>>>> Question is, what should we permit... >>>>> >>>>> For 'normal' operations, only unsigned op unsigned is permitted, >>>>> so unsigned**unsigned is obviously ok. >>>>> >>>>> What about (integer|real|complex)**unsigned? >>>>> >>>>> What about unsigned**integer? >>>>> >>>>> Since exponentiation is special (and also does not involve >>>>> type conversion) my feeling is to allow it all. >>>>> >>>>> Comments? >>>> >>>> Agree with the others. Supporting the above is fine. >>>> What about allowing unsigned**(real|complex)? The Fortran standard >>>> allows integer**(real|complex). >>> >>> I would vote against this. Mathematically, the result cannot be an >>> unsigned. >> >> As I noted, the standard allows integer**(real|complex). >> >> % cat a.f90 >> program foo >> integer u >> real x >> complex c >> x = 1.25 >> u = 2 >> c = (1.25,3.14) >> print *, u**x, u**c >> end program >> % gfcx -o z -std=f2018 a.f90 && ./z >> 2.37841415 (-1.35409331,1.95532227) >> >> The result is not integer. It literally is the expected >> mathematical expression exp((real|complex)*log(unsigned)) > > Do I understand correctly that the integer argument is promoted > to real first? This would be the sort of implicit conversion that > is generally excluded from UNSIGNED. > F2023 does not explicitly state a conversion occurs. There is Table 10.2 on page 161. This shows the types of the operands and the results of **. I'll need to look at the J3 text again, but if 'unsigned+(real|integer|complex)' is only allowed with explicit conversion of the 'unsigned' to the type of operand 2, then exponentiation would also require explicit conversion.
