~ chicken-core (master) 22860196304c17a687c3287a76a91365e6cf3833

commit 22860196304c17a687c3287a76a91365e6cf3833
Author:     Peter McGoron <code@mcgoron.com>
AuthorDate: Thu Jul 30 11:04:33 2026 -0400
Commit:     Felix Winkelmann <felix.winkelmann@bevuta.com>
CommitDate: Wed Sep 9 13:45:43 2026 +0200

    Modify special cases for asin
    
    The constants come from Gambit, but Gambit has mixed-exactness complex
    numbers and has its numeric system designed with that in mind. CHICKEN
    does not have those, although it has real numbers, which are a subset
    of mixed-exactness complex numbers. This upgrades unsigned 0 to signed
    zero because that is the most sensible thing to do (in the absense
    of mixed exactness). This causes a change to the constants.

diff --git a/library.scm b/library.scm
index 4a5fbdfc..9647b27f 100644
--- a/library.scm
+++ b/library.scm
@@ -2778,10 +2778,15 @@ EOF
 	      ;; These should fall out of the algorithm below,
 	      ;; but inexactness-promotion rules end up generating
 	      ;; a NaN somewhere.
-	      ((eqv? n -inf.0)
-	       -1.5707963267948966+inf.0i)
-	      ((eqv? n +inf.0)
-	       1.5707963267948966-inf.0i)
+	      ;;
+	      ;; These are the special cases -inf.0+0.0i and
+	      ;; +inf.0+0.0i from Gambit. Since Gambit has mixed exactness numbers,
+	      ;; this doesn't copy -inf.0+0i and +inf.0+0i. Basically, unsigned
+	      ;; zero is approached counterclockwise, and signed zero from the
+	      ;; side with that sign. So unsigned zero approches from the bottom
+	      ;; and matches -0.0, which is probably not what we want.
+	      ((eqv? n +inf.0)  1.5707963267948966+inf.0i)
+	      ((eqv? n -inf.0) -1.5707963267948966+inf.0i)
 	      (else
 	       (let* ((x (real-part n))
 	              (s:1-n (sqrt (- 1 n)))