igraph_hrg test fails on i386 arch in Debain Sid environment

[jgmbenoit]

Dear Maintainers,

it appears that the latest igraph source (version 0.10.9) fails the igraph_hrg test in the Debian Sid environment on the i386 architecture. The corresponding auto-build log is available at the Debian igraph auto-build webpage

[szhorvat]

This is likely harmless, and has to do with the known numerical instability of these functions ... I'll have a look in a couple of days. If anyone knows an easy way to test on that platform, let me know. We do run CI tests with x87 math, and those tests do not fail.

[szhorvat]

Copying the relevant part of the test output before the build log disappears:

521/554 Test #508: test::igraph_hrg ...................................***Failed    0.20 sec
-- Test case output differs from the expected output
-- See diff below:
-- -------------------------------------------------------
--- /<<PKGBUILDDIR>>/tests/unit/igraph_hrg.out	2024-02-04 21:40:29.000000000 +0000
+++ /<<PKGBUILDDIR>>/_BUILD_STATIC/tests/test_igraph_hrg.out	2024-02-05 11:14:47.195150846 +0000
@@ -1,24 +1,24 @@
-Vertex # 25, left: # 29, right: # 43, prob: 0.0067
-Vertex # 26, left: # 13, right: # 46, prob:      1
-Vertex # 27, left: # 14, right: # 19, prob:      1
-Vertex # 28, left: # 15, right: # 16, prob:      0
-Vertex # 29, left: # 28, right: # 41, prob:   0.23
-Vertex # 30, left: # 33, right: # 37, prob:   0.13
-Vertex # 31, left: #  8, right: # 34, prob:      1
-Vertex # 32, left: # 23, right: # 24, prob:      0
-Vertex # 33, left: # 22, right: # 27, prob:      0
-Vertex # 34, left: #  7, right: # 44, prob:      1
-Vertex # 35, left: # 17, right: # 39, prob:    0.1
-Vertex # 36, left: # 12, right: # 20, prob:      0
-Vertex # 37, left: # 26, right: # 36, prob:      0
-Vertex # 38, left: # 42, right: # 45, prob:      1
-Vertex # 39, left: # 21, right: # 48, prob:      0
-Vertex # 40, left: #  2, right: #  5, prob:      1
-Vertex # 41, left: # 32, right: # 35, prob:      0
-Vertex # 42, left: #  4, right: # 31, prob:      1
-Vertex # 43, left: #  0, right: # 47, prob:      1
-Vertex # 44, left: #  6, right: #  9, prob:      1
-Vertex # 45, left: #  1, right: # 40, prob:      1
-Vertex # 46, left: # 11, right: # 18, prob:      0
-Vertex # 47, left: #  3, right: # 38, prob:      1
-Vertex # 48, left: # 10, right: # 30, prob:   0.25
+Vertex # 25, left: # 33, right: # 36, prob:  0.029
+Vertex # 26, left: # 19, right: # 20, prob:      0
+Vertex # 27, left: # 14, right: # 26, prob:      1
+Vertex # 28, left: # 10, right: # 37, prob:   0.15
+Vertex # 29, left: # 21, right: # 30, prob:      0
+Vertex # 30, left: # 18, right: # 35, prob:      0
+Vertex # 31, left: #  1, right: # 42, prob:      1
+Vertex # 32, left: # 16, right: # 41, prob:      1
+Vertex # 33, left: # 32, right: # 39, prob:      0
+Vertex # 34, left: #  0, right: # 45, prob:      1
+Vertex # 35, left: # 11, right: # 17, prob:      0
+Vertex # 36, left: # 27, right: # 28, prob:      0
+Vertex # 37, left: # 43, right: # 48, prob:      0
+Vertex # 38, left: #  6, right: #  9, prob:      1
+Vertex # 39, left: # 13, right: # 29, prob:   0.75
+Vertex # 40, left: #  5, right: # 44, prob:      1
+Vertex # 41, left: # 23, right: # 24, prob:      0
+Vertex # 42, left: # 40, right: # 47, prob:      1
+Vertex # 43, left: #  3, right: # 31, prob:      1
+Vertex # 44, left: #  8, right: # 34, prob:      1
+Vertex # 45, left: #  2, right: #  4, prob:      1
+Vertex # 46, left: # 12, right: # 22, prob:      0
+Vertex # 47, left: #  7, right: # 38, prob:      1
+Vertex # 48, left: # 15, right: # 46, prob:      1
-- -------------------------------------------------------
CMake Error at /<<PKGBUILDDIR>>/etc/cmake/run_legacy_test.cmake:67 (message):
  Exiting test.

[jgmbenoit]

If anyone knows an easy way to test on that platform, let me know.

On Debian based distribution running on a amd64 architecture, we can create i386 chroot environments.
The following two links might be good starting point:
https://wiki.debian.org/Schroot/
https://wiki.ubuntu.com/SimpleSbuild/

[szhorvat]

Currently I don't have access to any x86 machines at all, and I'm not sure when I'd have the sort of access that allows a chroot. I simply don't have the resource to debug this.

I looked at the commit history, and it seems that I re-enabled this test in this commit:

ffb499f#diff-790ac84eea526803c98ecb2b2e62d424cb42899f24e7ecbbe53bb4a0295b36e4

It is possible that the difference in behaviour was already present. I wonder why it was disabled in the first place.

For the moment I suggest removing tests/unit/igraph_hrg.out to disable output checking for this test.

[szhorvat]

This part of the test was already commented out in the very first version of this test, 13 years go, see fdf756f

[ntamas]

Removing the expected output of the test for the time being, see the commit above. We can revert it later if/when we have time to debug this.