@amueller sorry, I did not check after @larsmans told me that the memoryview type should be float

what was failing?

@agramfort the C code was not matching the cython code. The cython code did not compile.

I'm not familiar with the code so it is really hard for me to review.

rebase?

I messed with master so you may need to reset. I'll buy you a beer when I'm back ;)

I don't understand this hotfix, why revert to char*?

Can you please remove the merge commit? create a new branch off of the current upstream/master and just cherry-pick fdac94b into it and force push to MechCoder:hot_fix.

As far as the hotfix goes, I do not have time to investigate further right now (exams approaching) but while rerunning cython liblinear.pyx I got this error #5274 (comment) , which somehow looks like numpy returns a char pointer for a data buffer in NumPy, which cython is not able to implicitly cast into a double pointer. (I'm just speculating not sure though)

@amueller Thanks for the offer ! But I am not much of a beer fan and would prefer either rum or whisky :)

The content of b3ab377 is not the cython cast hotfix for the liblinear sample weights.

just a second, fixing it up

fdac94b has the right fix in case you lost it.

Ok for using the char* type, it seems to be a Cython / numpy thingy. At least it's consistent with the rest of the file.

I cancelled the travis build because 25d50a6 still does not have the correct fix. You should use git log -p to check the content of the commit before pushing if you are not sure.

Sorry about the trouble, it should look all right now.

I tried the cherry-pick approach and it did not seem to work. Hence I reverted each of the faulty commits that seemed to show up and squashed.

ping @larsmans @agramfort ?

+1 if travis is green.

travis is green, merging.

I ll rebase my PR ASAP
Le 23 oct. 2015 20:10, "Olivier Grisel" notifications@github.com a écrit :

travis is green, merging.

—
Reply to this email directly or view it on GitHub
#5565 (comment)
.

Sorry for the mess everyone should have checked twice :((

No worries @MechCoder, happens to the best of us ;) @arthurmensch's PR will make sure we won't have this issue in the future.