JanWielemaker
commented
6 years ago Thanks for reporting. The dtoa() function is only used for quoted write that tries to write a float with as few as possible digits. This one is handled by formatFloat() in os/pl-fmt.c, which eventually calls
snprintf(). I recall there is a problem with that on Windows. Your examples work fine on Linux, but also for the current version fail on Windows, both 32-bit and 64-bit. The character following the string differs. I'll have a look into this snprintf story ...
hello !
i am using SWI-Prolog (threaded, 32 bits, version 7.4.2) on windows 7. using format/3 to atom/codes for a float in non-exponential notation produces a weird result if the number is > 1.e19 :
?- format(atom(X), '~f', [1e20]). X = '100000000000000000000.00000\004\'.
the last character of X is the ASCII control code 4 (EOT). the same happens for codes(X) instead of atom(X). it works fine for numbers <= 1.e19. for numbers >= 1e20 the atom length seems to be 28 in all cases.
?- format(atom(X), '~f', [1e19]), atom_length(X, L). X = '10000000000000000000.000000', L = 27.
?- format(atom(X), '~f', [1e20]), atom_length(X, L). X = '100000000000000000000.00000\004\', L = 28.
?- format(atom(X), '~f', [1e30]), atom_length(X, L). X = '100000000000000000000000000\004\', L = 28.
?- format(atom(X), '~f', [1e100]), atom_length(X, L). X = '100000000000000000000000000\004\', L = 28.
?- format(atom(X), '~f', [1e200]), atom_length(X, L). X = '999999999999999970000000000\004\', L = 28.
i understand that floats are printed by a C function called dtoa. may be there is a buffer size issue with that function ?
thanks, -attoparsec-