Bug 2756 – Bad code generation for pure nothrow math functions

Bad code generation occurs when the module is called 'std.math' in a module statement, and the signature is exactly 'pure nothrow double XXX(double)' or 'pure nothrow float XXX(float)', and the parameter is a compile-time constant. I suspect this is related to the special treatment of std.math.sqrt(), allowing it to be evaluated at compile time. ---------------- MINIMAL TEST CASE: ---------------- module std.math; pure nothrow double food(double x){ return 2.0;} void main() { assert( food(1.0) == 2.0 ); } ---------------- FURTHER INFO: ---------------- module std.math; import std.stdio; pure nothrow float foof(float x){return 2.0;} pure nothrow double food(double x){ return 2.0;} pure nothrow real foor(real x) { return 2.0; } void main() { writefln(foor(1.0)); // writes 2. OK. writefln(food(1.0)); // writes -0 !! writefln(foof(1.0)); // writes 0 !! }

Patch: builtin.c line 62 -------- if (ident == Id::_sqrt) // add this line builtin = BUILTINsqrt; --------

Eh, that's an embarassing one!

Fixed DMD 2.027