Bug 356 – std,math.tan doesn't preserve NaN payloads [fix included]

Currently in tan(x), it effectively does if (isnan(x)) return real.nan; when it should be if (isnan(x)) return x; The good news is that all math functions preserve NaN payloads, with this one exception. And it's a one-line fix. In the asm code for tan, at the end of the code, add the line "jnp Lret". ------- trigerr: jnp Lret ; // if x is NaN, return x. <<< ADD THIS LINE fstp ST(0) ; // dump x, which will be infinity } return real.nan; Lret: ; } --------- Also, tan() currently fails one if its unit tests ( tan(1e100)). Since the isnan(tan(1e100)) is not correct mathematically, I think that line should just be removed. -------- // overflow [ real.infinity, real.nan], [ real.nan, real.nan], [ 1e+100, real.nan], <<< REMOVE THIS LINE ]; --------

[email protected] wrote: > http://d.puremagic.com/issues/show_bug.cgi?id=356 > > Summary: std,math.tan doesn't preserve NaN payloads [fix > included] > Product: D > Version: 0.167 > Platform: PC > OS/Version: Windows > Status: NEW > Severity: minor > Priority: P1 > Component: Phobos > AssignedTo: [email protected] > ReportedBy: [email protected] > > > Currently in tan(x), it effectively does > if (isnan(x)) return real.nan; > when it should be > if (isnan(x)) return x; > > The good news is that all math functions preserve NaN payloads, with this one > exception. And it's a one-line fix. I spoke too soon. The version(linux) form of pow() should have two lines changed: if (isnan(y)) return y; // <<< instead of "return real.nan" <<<<<<<< if (y == 0) return 1; // even if x is $(NAN) if (isnan(x) && y != 0) return x; // instead of "return real.nan;" <<<<<<

Fixed DMD 0.168