14.26.3 Bugs Fixed

• format/3 could crash with certain parameters to the ‘~*’ construct (and the equivalent inline form). For some format control options, this was fixed by restricting the magnitude of the ‘~*’ parameter.
• Profiling could crash.
• Locating source info for exceptions could crash.
• Minor efficiency bug in backtracking in dynamic code.
• Efficiency bug in append(L1,L2,L3) if called with L1 not a proper list.
• Reading a term could crash under low memory conditions.
• Multiple problems with the compatibility library system3.
• In some cases, large integers could overflow when converting to floating point values. This could lead to incorrect behavior:
  • - For arithmetic expressions, e.g. sin/1 and (/)/2, which implicitly take floating point values, a float overflow error will now be thrown if the argument evaluates to a large integer that will not fit in a floating point value.

    In some cases this now throws an error, where earlier versions computed a sensible result, e.g. X is 1/(1<<1024) would compute X = 0.0 but now throws an error (because the value of (1<<1024) is a large integer that does not fit in a floating point number).

  • - SP_get_float() now fails if the argument term is a large integer that does not fit in a C double. It used to return an infinity value.
  • - Foreign resource code using the type +float, or code that use the library(structs) floating point types, will now give an error if the argument is a large integer that does not fit in the destination type.
• In some cases, floating point negative zero, not a number (NaN) and positive or negative infinity could enter the system. These values are not supported, and caused various incorrect behavior:
  • - Reading the text ‘-0.0’ now creates an ordinary (unsigned) floating point zero.
  • - SP_put_float(-0.0) now creates an ordinary (unsigned) floating point zero.
  • - SP_put_float(-inf), SP_put_float(+inf) and SP_put_float(NaN) fail².
  • - Foreign resource code using the type -float, or code that use the library(structs) floating point types, will now throw an error if the value is not finite. Negative zero will treated as an ordinary (unsigned) floating point zero.

    This means that a call to a foreign function can throw an exception when the function returns to Prolog, if it attempts to return a non-finite result.

• Foreign resource code using the type +integer, or code that uses the library(structs) integer types, will now give an error if the argument is a large integer that does not fit in the destination type.
• spld --extended-rt failed unless --license-file was specified. Now it tries to find the license in the “usual” places.
• ZINC: With the timeout/1 option of various predicates and the spfz option -time Time together with one-solutions search, no solution was reported at time-out, even if some solution had been found during optimization search.
• CLPFD:
  • - Incorrect special case for bool_channel/4.
  • - Propagation bug when mixing dif/2 and reified constraints.
  • - Idempotency and correctness bugs in multiple global constraints.