maxima-5.39.0/ecl-16.1.3 new testsuite failure
Hello *, There is a new testsuite failure in maxima-5.39.0 compiled with ecl-16.1.3: Running tests in rtest16: ********************** Problem 561 (line 2153) *************** Input: 2 - %pi ev(closeto(li (1.0 %i) - (------ + %i 0.915965594177219), 1.3878e-16), numer) 2 48 Result: error-catch This differed from the expected result: true 612/613 tests passed With ecl-16.1.2 there were no unexpected failures in the maxima testsuite. Andrey
Hey, I have already tested with Maxima, see: https://gitlab.com/embeddable-common-lisp/ecl/issues/307 . ECL doesn't accept denormals by default now, I suspect that this is the main cause of this error (but I didn't know how to investigate maxima problems further – i.e what error was cought). To ignore all fpe signals issue: (si:trap-fpe t nil) to turn off some of them, do: (si:trap-fpe :floating-point-overflow nil) (si:trap-fpe :division-by-zero nil) (si:trap-fpe :floating-point-invalid-operation nil) (si:trap-fpe :floating-point-inexact nil) (si:trap-fpe :floating-point-underflow nil) second argument is whenever this particular exception should be signalled, or simply ignored – NIL means "ignore". Best regards, Daniel Andrey G. Grozin writes:
Hello *,
There is a new testsuite failure in maxima-5.39.0 compiled with ecl-16.1.3:
Running tests in rtest16: ********************** Problem 561 (line 2153) *************** Input: 2 - %pi ev(closeto(li (1.0 %i) - (------ + %i 0.915965594177219), 1.3878e-16), numer) 2 48
Result: error-catch
This differed from the expected result: true
612/613 tests passed
With ecl-16.1.2 there were no unexpected failures in the maxima testsuite.
Andrey
-- Daniel Kochmański ;; aka jackdaniel | Przemyśl, Poland TurtleWare - Daniel Kochmański | www.turtleware.eu "Be the change that you wish to see in the world." - Mahatma Gandhi
On Mon, Dec 19, 2016 at 02:17:52PM +0100, Daniel Kochmański wrote:
I have already tested with Maxima, see: https://gitlab.com/embeddable-common-lisp/ecl/issues/307 . ECL doesn't accept denormals by default now, I suspect that this is the main cause of this error (but I didn't know how to investigate maxima problems further – i.e what error was cought).
Trying to evaluate this in Maxima built with ECL 16.1.3 leads to this error: (%i1) li[2](1.0*%i); Condition of type: DIVISION-BY-ZERO I built ECL 16.1.2 and 16.1.3 from sources on my 64-bit Linux box. It looks like the problem comes down to this: On ECL 16.1.2:
(log #c(0.0 -1.0)) #C(0.0 -1.5707964) (log #c(-0.0 -1.0)) #C(0.0 -1.5707964)
while on 16.1.3:
(log #c(0.0 -1.0)) #C(0.0 -1.5707964) (log #c(-0.0 -1.0)) Condition of type: DIVISION-BY-ZERO
Cheers, Kris Katterjohn
Hey, Thank you for narrowing this down! I've added an issue to gitlab: https://gitlab.com/embeddable-common-lisp/ecl/issues/329 Best regards, Daniel -- Daniel Kochmański ;; aka jackdaniel | Przemyśl, Poland TurtleWare - Daniel Kochmański | www.turtleware.eu "Be the change that you wish to see in the world." - Mahatma Gandhi
This wasn't a regression, just reinforcing fpe shown already existing problem in atan function. The problem is fixed in the develop branch. -- Daniel Kochmański ;; aka jackdaniel | Przemyśl, Poland TurtleWare - Daniel Kochmański | www.turtleware.eu "Be the change that you wish to see in the world." - Mahatma Gandhi
participants (3)
-
Andrey G. Grozin -
Daniel Kochmański -
Kris Katterjohn