maxima-5.37.3 built with ecl-16.0.0 runs its testsuite without unexpected errors. When buils with ecl-16.1.2, it fails in
Running tests in rtest8: ********************** Problem 104 *************** Input: ev(e5, au = 0, omega = 2)
Result: - u quad_qawf(%e , u, 0, 2, sin, epsabs = 1.0e-10, limit = 32, maxp1 = 100, limlst = 10)
This differed from the expected result: [0.4000000000000001, 2.216570948815925e-11, 175, 0]
162/163 tests passed
Andrey
An additional bit of information: this failure happens in 64-bit Gentoo linux. In 32-bit Gentoo linux there are no unexpected errors in the testsuite.
How can one debug this problem?
Andrey
I'm usually taking the time-consuming process of bisecting ECL, unless I already know what the problem is. Some minimal test without building the whole maxima would be great though.
Regards, Daniel
Andrey G. Grozin writes:
An additional bit of information: this failure happens in 64-bit Gentoo linux. In 32-bit Gentoo linux there are no unexpected errors in the testsuite.
How can one debug this problem?
Andrey
Daniel Kochmański writes:
I'm usually taking the time-consuming process of bisecting ECL, unless I already know what the problem is. Some minimal test without building the whole maxima would be great though.
Regards, Daniel
Andrey G. Grozin writes:
An additional bit of information: this failure happens in 64-bit Gentoo linux. In 32-bit Gentoo linux there are no unexpected errors in the testsuite.
I suspect, that the problem may be caused by Mersenne-Twister RNG 64 bit implementation (which is new).
How can one debug this problem?
Andrey
I can't reproduce it on my 64-bit ArchLinux box. What flags did you built ECL with?
Could you check, if the bug is present on: 4548ed7c4cd9a428d95695b6528e55d6d80cb11c and bff9aced440c47893d283d48cc251391828836a9 ?
(this is before and after the new RNG implementation).
Regards, Daniel
Andrey G. Grozin writes:
An additional bit of information: this failure happens in 64-bit Gentoo linux. In 32-bit Gentoo linux there are no unexpected errors in the testsuite.
How can one debug this problem?
Andrey
Very strange: I cannot reproduce this failure. On the same computer, with the same ecl and maxima. No unexpected errors.
Why I got this failure the first time I rub the testsuite after compiling maxima with the new ecl? It seems impossible to find out.
Sorry for the noise, Andrey
No problem. It's weird indeed. Maybe tests have some timeouts and ECL was running the test for too long (ie your machine was under heavy load)?
Either way it's a great news that we have no regression here :-)
Best regards, Daniel
Andrey G. Grozin writes:
Very strange: I cannot reproduce this failure. On the same computer, with the same ecl and maxima. No unexpected errors.
Why I got this failure the first time I rub the testsuite after compiling maxima with the new ecl? It seems impossible to find out.
Sorry for the noise, Andrey
Thanks,
I'll try investigate it today. Is this file a correct instruction on building and testing Maxima:
https://sourceforge.net/p/maxima/code/ci/master/tree/INSTALL.lisp
or should I perform some other steps?
Regards, Daniel
Andrey G. Grozin writes:
maxima-5.37.3 built with ecl-16.0.0 runs its testsuite without unexpected errors. When buils with ecl-16.1.2, it fails in
Running tests in rtest8: ********************** Problem 104 *************** Input: ev(e5, au = 0, omega = 2)
Result: - u quad_qawf(%e , u, 0, 2, sin, epsabs = 1.0e-10, limit = 32, maxp1 = 100, limlst = 10)
This differed from the expected result: [0.4000000000000001, 2.216570948815925e-11, 175, 0]
162/163 tests passed
Andrey
On Mon, 29 Feb 2016, Daniel Kochmański wrote:
I'll try investigate it today. Is this file a correct instruction on building and testing Maxima:
https://sourceforge.net/p/maxima/code/ci/master/tree/INSTALL.lisp
or should I perform some other steps?
I simply do
./configure --enable-ecl make
After you start maxima, say
run_testsuite();
As I said in my second email, I get this failure in 64-bit Gentoo linux (x86-64, or amd64) but not in 32-bit Gentoo linux (plain x86). Strange.
Andrey
"Andrey" == Andrey G Grozin A.G.Grozin@inp.nsk.su writes:
Andrey> maxima-5.37.3 built with ecl-16.0.0 runs its testsuite without unexpected Andrey> errors. When buils with ecl-16.1.2, it fails in
Andrey> Running tests in rtest8: Andrey> ********************** Problem 104 *************** Andrey> Input: Andrey> ev(e5, au = 0, omega = 2)
Andrey> Result: Andrey> - u Andrey> quad_qawf(%e , u, 0, 2, sin, epsabs = 1.0e-10, limit = 32, maxp1 = 100, Andrey> limlst Andrey> = 10)
Andrey> This differed from the expected result: Andrey> [0.4000000000000001, 2.216570948815925e-11, 175, 0]
This usually means quad_qawf caught some kind of error and gave up and returned a noun form.
Might help to do (trace slatec:dqawf) to see what's happening.
We should probably also add some internal debugging so that we can see what kinds of error is actually encountered.
-- Ray
-
Andrey G. Grozin -
Daniel Kochmański -
Raymond Toy