Increase test tolerance#797
Conversation
Needed for i386. Most likely needed because of the 80-bit floating point precision on that platform. See https://buildd.debian.org/status/fetch.php?pkg=tasmanian&arch=i386&ver=8.1-1&stamp=1739206789&raw=0
| if (std::abs(sum - 96.0 * 512.0 / 27.0) > 10.0 * Maths::num_tol){ // without 10.0 the test fails on dpcpp with error 1.E-12 | ||
| #else | ||
| if (std::abs(sum - 96.0 * 512.0 / 27.0) > Maths::num_tol){ | ||
| if (std::abs(sum - 96.0 * 512.0 / 27.0) > 2.0 * Maths::num_tol){ |
There was a problem hiding this comment.
IEEE 754 can be deceptively accurate even if it doesn't make much sense on the math side.
I think it is safe to just remove the #ifdef and go with 10.0 * Maths::num_tol in all cases. This tests a rule on an unbounded domain (0, inf), higher rounding error is expected.
There was a problem hiding this comment.
I can do this myself, just close the PR if you want me to do it.
There was a problem hiding this comment.
Thanks, I let you fix this.
There was a problem hiding this comment.
@svillemot Thanks for putting Tasmanian into the Debian repo.
Let me know if I can do anything else to help. I normally don't backport bugfixes but I can backport both this and he Python install dir to the 8.2 release and make it 8.2.1, let me know if this would be helpful.
There was a problem hiding this comment.
Thanks for your offer to help. Merging Debian-specific patches into the GitHub repo is what matters most, because it means I’ll be able to drop those patches for the next release.
In the meantime I don’t really need a new bugfix release, because the patches are already applied to the Debian package, but feel free to make one if you think it can benefit other users outside Debian.
2 participants
Needed for i386. Most likely needed because of the 80-bit floating point precision on that platform.
See https://buildd.debian.org/status/fetch.php?pkg=tasmanian&arch=i386&ver=8.1-1&stamp=1739206789&raw=0
This patch has been applied to the Debian package.