-
Notifications
You must be signed in to change notification settings - Fork 6
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
Enable tests in build #113
Conversation
769836e
to
6b33fa8
Compare
@nhartland, could you produce one theory with low precision and a couple of experiments? Then we can download it during the testing phase and use it to check fktables and whatnot. |
How big would be acceptable? We already have theory 162 which is super-low precision, but it includes all sets. It's coming in at 125 Mb/ |
As opposed to the usual ~5 GB. |
Could we have a subset of that with a couple of experiments? In any case 125Mb is less than some conda packages, so it's probably fine. |
We can do, but I'm lazy and adding theories is a pain. |
Let's try with 162 and see how long takes for travis to kick as out. |
f8dcfac
to
dc91b55
Compare
Great, getting two different build failures on linux and mac. I really have no patience for understanding how linking in mac works but it is really annoying me how absurd it is. https://travis-ci.com/NNPDF/nnpdf/jobs/113751056 |
This now works on linux (albeit it is somewhat absurd to run tests in release mode). @nhartland I would appreciate some help with the Mac failure, which looks frankly nonsensical. |
This error? |
Yes. Also, apparently I haven't fixed travis so it refuses to build anything.... |
What is the deal with travls? I keep getting emails saying we're running out of trials. Is this something we'll need to pay for at some point? |
Apparently I have it for free, but the repo must belong to me. I though it would be enough to put my credentials. |
I have set up travis on my fork. I have also sent an email asking if we could have it for NNPDF. For now a solution would be:
which causes to push both to the NNPDF and to my repository, while pulling is always from NNPDF. This is not ideal since e.g. we don't get the flag in the PR. |
The build of the test keeps failing: https://travis-ci.com/Zaharid/nnpdf/builds/67740102?utm_source=email&utm_medium=notification I suppose some permutation of the options here would fix it: https://cmake.org/Wiki/CMake_RPATH_handling but I have ran out of patience. |
Travis is working now, so it would be good to have it do something useful! |
Did they reply to your email? |
Haha, cheers!
… —
You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub <#113 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AE_NDFEwyVPvU5Qi2X6yPeyYemBIdu3-ks5teqcPgaJpZM4SWjWk>.
|
So I'm not even sure how to reproduce this error let alone try and resolve it. I can't see the travis output for the nnpdf version in your repo:
because I don't have access to it |
On the travis logs, there is:
In particular, |
Haha yes indeed. I'll try just overwriting LDFLAGS without that. |
So now there is a different error. I don't know if overwriting the LDFLAGS without that flag has fixed the problem or if we've just covered it up with another problem, but let's be optimistic! The error now is an rpath one which I suspect would be fixed by installing the tests before running them (if that makes sense?). You've set |
5db2a30
to
457557d
Compare
Seems like this worked!
|
Are these flags pretty much hard coded into conda then, not something we can trivially remove from some config? Overwriting the LDFLAGS is probably asking for trouble further down the line. Maybe something as simple as running LDFLAGS through sed and cutting out that one flag would be marginally more robust. |
I was going to do the sed thing. |
So my understanding of the issue with Mac and conda is the following:
@nhartland, can you get a working dev environment with that information? |
This makes self testing easier.
Set up the requirement environment to: - Be able to test validphys with pytest. - Be able to compile and run the catch test, with all the debug flags enabled. This has the added benefit of testing that the debug build actually works.
Apparently these are not defined and we want to have the undefined check.
This requires updating all the regression tables with the new theory. We also put all the tests in a file so that we have clear control over the order. In particular we want python test first so that we download theories and pdfs.
One of the many useless missfeatures of setuptools is that it doesn't copy all of the files by default.
I am only half sure this is a good idea...
If it compiles the library maybe it will compile the tests?
I'd like to merge this because it is annoying to write tests that depend on it otherwise. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sure, sorry, obviously fine by me. I'll see if I can finally resolve #98.
This implements some but not all of the points in #95. In order to be able to merge it, we require a low precision theory that we can use it tests. Currently, it predictably fails with:
It is also missing enabling coverage reports, but that's out of scope here.