New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
FAIL: distribution_tests #10
Comments
Hello, |
So? |
I'm also unable to replicate on an up-to-date Fedora box with gsl 1.15. @cicku, any hints as to how we can replicate the failure? OS, number of threads, using the latest version of the library? If you set .verbose='y' in tests/distribution_tests.c:208, the test should log the parameters tried; what was the last set at failure? If you're running |
I tested it on Fedora 22 32 bits. I will re check this soon. |
It appears that the tests failed on Debian i386 boxes (for the least at the time of writing): |
I recall having trouble with the GSL's beta-related functions when Many of the tests are based on using random draws to arrive at a statistic, and comparing the statistic to its expected value to within some tolerance. There are two types of error:
When a test fails, it's worth investigating which type of error it is, and if it's the second type, it may be reasonable to just adjust the tolerance of the test. Raising the number of draws is also an option, depending on how long we are willing to wait for a test to finish. |
Finally I could grab the test-suite.log file on a i386 box: I got the same error. |
Yes, I still get the same error with 0.999a. |
The error was fix in the last debian packages, and the fix was brought to GitHub a couple of days ago. |
Hi @jgmbenoit and all, the bug was fixed, but the 1.0 version introduced the issue again (I checked only i386) I checked, the patches were indeed applied, so I looked at the other commits, and the bisect found the commit responsible for the regression: reverting this commit on top of 1.0 makes the testsuite succeed on i386
I'm not sure about what needs to be fixed, but right now the package is stuck in unstable, and won't migrate without a fix. cheers! Gianfranco |
BTW this seems to be a regression on 32bit architecture, because at a first look 64 bit builds are fine. |
Thanks for your reports. I will come back to this bug latter, once I can concentrate: it is very low level. I have my idea, but I have to test it on different architectures. |
actually the FTBFS is RC for Debian policy, and the package build failure is preventing the old libgsl from being removed from unstable/testing (cruft report). so, if you can, please upload the revert of that commit (I can do it if you don't have upload privileges), and then take your time to properly fix the issue |
I will not assume that the revert is safe. I was not aware of the GSL issue. Is there a list of packages that block the GSL package ? I will have a look and try to fix the issue this week-end. |
we are working on the decruft of the old library, there is no hurry, but please note that the next ubuntu LTS should have 1.0, and it is currently in proposed because of this bug. |
Fortunately enough, this timeline is greater than my timeline. |
@LocutusOfBorg, would you please post or send me |
I hope this helps! cheers! |
Thanks, that's very useful. I'll have a closer look in a few hours and try to get this resolved today. |
Commit 4bb0399 addresses the bug. Please test on your side and let me know how it goes. What happened: line 801 of The commit @LocutusOfBorg pointed out added a page to the parameter set and thus caused this bug to be active. |
I am ready to buy it :-) When do you plan to make a new release ? |
@b-k @jgmbenoit thanks to you both!
|
I have just updated the debian package material and ask for sponsorship to the usual sponsor. Thanks for your help. |
thanks! |
OK, have updated the pkg branch with the above changes (and two others regarding Kullback-Leibler divergence and mixture distributions). |
Thanks! |
Hi,
The testsuite is nearly perfect except one below:
Bernoulli distribution: ...
PASS. Bernoulli distribution: ......
PASS. Beta distribution: ...
PASS. Beta distribution: gsl: simplex.c:265: ERROR: non-finite function value encountered
Default GSL error handler invoked.
gsl version: 1.16
The text was updated successfully, but these errors were encountered: