Guile 3.0.7 test failures on i686-linux, glibc 2.33 [core-updates]

  • Done
  • quality assurance status badge
Details
2 participants
  • Ludovic Courtès
  • Maxime Devos
Owner
unassigned
Submitted by
Ludovic Courtès
Severity
important
L
L
Ludovic Courtès wrote on 3 Jul 2021 23:41
(address . bug-guix@gnu.org)
87tulbxegp.fsf@inria.fr
On current ‘core-updates’
(ca. 39f1486efd70712416ca784f9014132644b04155), Guile 3.0.7,
specifically (@@ (gnu packages commencement) guile-final) fails tests on
i686-linux:

Toggle snippet (13 lines)
Running numbers.test
FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 10/7)
FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 -10/7)
FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (130.0 10/7)
FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (-130.0 -10/7)
FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (130.0 -10/7)
FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (-130.0 10/7)
FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 10/7)
FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 -10/7)
FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 10/7)
FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 -10/7)

Note that this doesn’t happen on ‘master’ (glibc 3.31).

Ludo’.
L
L
Ludovic Courtès wrote on 3 Jul 2021 23:45
control message for bug #49368
(address . control@debbugs.gnu.org)
87sg0vxe9f.fsf@gnu.org
severity 49368 important
quit
M
M
Maxime Devos wrote on 4 Jul 2021 14:06
Re: bug#49368: Guile 3.0.7 test failures on i686-linux, glibc 2.33 [core-updates]
c3d4cad63eaef080975b83afcd0c4c60161263dd.camel@telenet.be
Ludovic Courtès schreef op za 03-07-2021 om 23:41 [+0200]:
Toggle quote (21 lines)
> On current ‘core-updates’
> (ca. 39f1486efd70712416ca784f9014132644b04155), Guile 3.0.7,
> specifically (@@ (gnu packages commencement) guile-final) fails tests on
> i686-linux:
>
> --8<---------------cut here---------------start------------->8---
> Running numbers.test
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (-130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (-130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 -10/7)
> --8<---------------cut here---------------end--------------->8---
>
> Note that this doesn’t happen on ‘master’ (glibc 3.31).

In a REPL:

(euclidean/ 130.0 10/7)
(euclidean/ 130.0 -10/7)
(floor/ 130.0 10/7)
(floor/ -130.0 -10/7)
(ceiling/ 130.0 -10/7)
(ceiling/ -130.0 10/7)
(truncate/ 130.0 10/7)
(truncate/ 130.0 -10/7)
(truncate/ -130.0 10/7)
(truncate/ -130.0 -10/7)

Output on core-updates (using --keep-failed and ./meta/uninstalled-env guile):

scheme@(guile-user)> (euclidean/ 130.0 10/7)
$1 = 90.0
$2 = 1.4285714285714257
scheme@(guile-user)> (euclidean/ 130.0 -10/7)
$3 = -90.0
$4 = 1.4285714285714257
scheme@(guile-user)> (floor/ 130.0 10/7)
$5 = 90.0
$6 = 1.4285714285714257
scheme@(guile-user)> (floor/ -130.0 -10/7)
$7 = 90.0
$8 = -1.4285714285714257
scheme@(guile-user)> (ceiling/ 130.0 -10/7)
$9 = -90.0
$10 = 1.4285714285714257
scheme@(guile-user)> (ceiling/ -130.0 10/7)
$11 = -90.0
$12 = -1.4285714285714257
scheme@(guile-user)> (truncate/ 130.0 10/7)
$13 = 90.0
$14 = 1.4285714285714257
scheme@(guile-user)> (truncate/ 130.0 -10/7)
$15 = -90.0
$16 = 1.4285714285714257
scheme@(guile-user)> (truncate/ -130.0 10/7)
$17 = -90.0
$18 = -1.4285714285714257
scheme@(guile-user)> (truncate/ -130.0 -10/7)
$19 = 90.0
$20 = -1.4285714285714257

It appears that truncate/, ceiling/ and floor/ give the same output ...
that doesn't seem correct.

Greetings,
Maxime.
-----BEGIN PGP SIGNATURE-----

iI0EABYKADUWIQTB8z7iDFKP233XAR9J4+4iGRcl7gUCYOGkPBccbWF4aW1lZGV2
b3NAdGVsZW5ldC5iZQAKCRBJ4+4iGRcl7pb/AP4iErBmWRYgxDQmlFpOgaItThKB
v61U/iNfeIA0mGNNqwD9Fm27V20yPTIfUNH1HFKo54Tesd/5RADeIMOpWyAR2wA=
=GU8x
-----END PGP SIGNATURE-----


M
M
Maxime Devos wrote on 4 Jul 2021 14:31
707865c726394c527a9141e742a575c2e4e19a67.camel@telenet.be
Ludovic Courtès schreef op za 03-07-2021 om 23:41 [+0200]:
Toggle quote (10 lines)
> On current ‘core-updates’
> (ca. 39f1486efd70712416ca784f9014132644b04155), Guile 3.0.7,
> specifically (@@ (gnu packages commencement) guile-final) fails tests on
> i686-linux:
>
> --8<---------------cut here---------------start------------->8---
> Running numbers.test
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 10/7)
> [...]

Curiously, on core-updates, ceiling/ is inconsistent with ceiling-quotient
and ceiling-remainder:

scheme@(guile-user) [1]> (ceiling/ 130.0 -10/7)
$28 = -90.0
$29 = 1.4285714285714257
scheme@(guile-user) [1]> (ceiling-quotient 130.0 -10/7)
$30 = -91.0
scheme@(guile-user) [1]> (ceiling-remainder 130.0 -10/7)
$31 = 0.0

But on 'master', the results are consistent:

scheme@(guile-user)> (ceiling/ 130.0 -10/7)
$1 = -91.0
$2 = 0.0
scheme@(guile-user)> (ceiling-quotient 130.0 -10/7)
$3 = -91.0
scheme@(guile-user)> (ceiling-remainder 130.0 -10/7)
$4 = 0.0

To be investigated ...

Greetings,
Maxime.
Toggle quote (1 lines)
>
-----BEGIN PGP SIGNATURE-----

iI0EABYKADUWIQTB8z7iDFKP233XAR9J4+4iGRcl7gUCYOGqOhccbWF4aW1lZGV2
b3NAdGVsZW5ldC5iZQAKCRBJ4+4iGRcl7oYtAP4lFjMOZyfCyJp9+W9e/RoS2kNg
35ZmRF8jK8jV3WemtgD/Vpfi1ljFf2TEJm4tKh7G/2OzQ0TMoEhxDXfhGCRpUQU=
=A9Zb
-----END PGP SIGNATURE-----


M
M
Maxime Devos wrote on 6 Jul 2021 15:24
b6b4a1391d032a4f01ab5901e12464493e1c10a2.camel@telenet.be
Ludovic Courtès schreef op za 03-07-2021 om 23:41 [+0200]:
Toggle quote (23 lines)
> On current ‘core-updates’
> (ca. 39f1486efd70712416ca784f9014132644b04155), Guile 3.0.7,
> specifically (@@ (gnu packages commencement) guile-final) fails tests on
> i686-linux:
>
> --8<---------------cut here---------------start------------->8---
> Running numbers.test
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: floor/: mixed types: (-130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: ceiling/: mixed types: (-130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (130.0 -10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 10/7)
> FAIL: numbers.test: Number-theoretic division: truncate/: mixed types: (-130.0 -10/7)
> --8<---------------cut here---------------end--------------->8---
>
> Note that this doesn’t happen on ‘master’ (glibc 3.31).
>
> Ludo’.

I wondered if the implementation of the C function 'ceil' is buggy,
so I tried the following (spoiler: 'ceil' doesn't seem to be the issue,
or I need another test case:

Run "guix build --system=i686-linux -e (@@ (gnu packages commencement) guile-final) --keep-failed".
In the build directory, write to a file "test.c":

#include <stdio.h>
#include <math.h>

int
main(void)
{
double x;
double y;
if (scanf("%lf %lf", &x, &y) < 0) /* < 0 is actually incorrect here */
return 1;
if (printf("%lf/%lf=%lf\n", x,y, x/y) < 0)
return 1;
if (printf("ceil(%lf/%lf)=%lf\n", x,y, ceil(x/y)) < 0)
return 1;
return 0;
}

Run ((exact->inexact 10/7) = 1.4285714285714286 on both master and core-updates):

$ gcc -Wall -Wextra -lm test.c -o test
$ echo "130.0 1.4285714285714286" | ./test
130.000000/1.428571=91.000000
ceil(130.000000/1.428571)=91.000000

On master, the output is:
130.000000/1.428571=91.000000
ceil(130.000000/1.428571)=91.000000
(idem)

Greetings,
Maxime.
-----BEGIN PGP SIGNATURE-----

iI0EABYKADUWIQTB8z7iDFKP233XAR9J4+4iGRcl7gUCYORZnhccbWF4aW1lZGV2
b3NAdGVsZW5ldC5iZQAKCRBJ4+4iGRcl7q/WAQD4dfWMlKM9QjPclKIrEhyjZ5fr
lUbDUyS3ybODRQXJyAD8CbgxbnFTq0qARX+7Cu/XY2oZxpd+on6g5ddpcFjbJAk=
=3504
-----END PGP SIGNATURE-----


M
M
Maxime Devos wrote on 24 Jul 2021 20:46
(address . 49368-done@debbugs.gnu.org)
7fce1551ca9819053aa094923e6d6f39b9966f22.camel@telenet.be
Ludovic Courtès schreef op za 03-07-2021 om 23:41 [+0200]:
Toggle quote (10 lines)
> On current ‘core-updates’
> (ca. 39f1486efd70712416ca784f9014132644b04155), Guile 3.0.7,
> specifically (@@ (gnu packages commencement) guile-final) fails tests on
> i686-linux:
>
> --8<---------------cut here---------------start------------->8---
> Running numbers.test
> FAIL: numbers.test: Number-theoretic division: euclidean/: mixed types: (130.0 10/7)
> [...]

-----BEGIN PGP SIGNATURE-----

iI0EABYKADUWIQTB8z7iDFKP233XAR9J4+4iGRcl7gUCYPxgARccbWF4aW1lZGV2
b3NAdGVsZW5ldC5iZQAKCRBJ4+4iGRcl7nLJAP4vFJtHqZpJR3x3GqT8iAMBIgSs
0xMTdWLdjMq1riDaCAD/YJInSXX/t1KEyZ91c8vuQO4cv7w1n76hCtTHyaFIAwY=
=9Tkp
-----END PGP SIGNATURE-----


Closed
?