make check on windows: assertion failed: (equal? 1.0 (numerator 0.0))

[Mario Domenech Goulart]

======================================== library tests ...

Error: assertion failed: (equal? 1.0 (numerator 0.0))

	Call history:

	<syntax>	  (##core#undefined)
	<syntax>	  (##sys#error "assertion failed" (##core#quote (equal? +inf.0 (denominator 0.0))))
	<syntax>	  (##core#quote (equal? +inf.0 (denominator 0.0)))
	<eval>	  (equal? +inf.0 (denominator 0.0))
	<eval>	  (denominator 0.0)
	<syntax>	  (assert (equal? 1.0 (numerator 0.0)))
	<syntax>	  (##core#if (##core#check (equal? 1.0 (numerator 0.0))) (##core#undefined) (##sys#error "assertion fa...
	<syntax>	  (##core#check (equal? 1.0 (numerator 0.0)))
	<syntax>	  (equal? 1.0 (numerator 0.0))
	<syntax>	  (numerator 0.0)
	<syntax>	  (##core#undefined)
	<syntax>	  (##sys#error "assertion failed" (##core#quote (equal? 1.0 (numerator 0.0))))
	<syntax>	  (##core#quote (equal? 1.0 (numerator 0.0)))
	<eval>	  (equal? 1.0 (numerator 0.0))
	<eval>	  (numerator 0.0)
	<eval>	  (##sys#error "assertion failed" (##core#quote (equal? 1.0 (numerator 0.0))))	<--
make[1]: *** [check] Error 70
make[1]: Leaving directory `c:/src/chicken-core'
make: *** [check] Error 2

(c) 2008-2013, The Chicken Team
(c) 2000-2007, Felix L. Winkelmann
Version 4.8.3 (rev b663e07)
windows-mingw32-x86 [ manyargs dload ptables ]
compiled 2013-11-09 on foobar (MINGW32_NT-5.1)

[sjamaan]

This seems to be caused by a bug in MingW's fpclassify():

​http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56404
​http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56397

So far I haven't found anything in the mingw bug database relating to this

[sjamaan]

Found the MingW issue: ​http://sourceforge.net/p/mingw/bugs/1927/

[sjamaan]

If we fix this, we should probably restore the commented-out test as well.

[sjamaan]

Bug also seems to exist in mingw-w64 (which is apparently the maintained version of mingw)

[sjamaan]

Mingw-w64 bugreport: ​http://sourceforge.net/p/mingw-w64/bugs/367/

[sjamaan]

"fixed" by disabling the tests in changeset c368462. Properly trying to fix all broken systems is a rabbit-hole that goes too deep; for example iOS even sets a hardware flag on the FPU to treat denormalised floating-point numbers as zero. There may be more OS/CPU combinations that do similarly fubar things. Our industry's madness knows no limits.