Add the atlas folder to LD_LIBRARY_PATH

[ogrisel]

To make the build pack work and able to unpickle sklearn models built offline I add to:

$ heroku config:set BUILDPACK_URL=https://github.com/dbrgn/heroku-buildpack-python-sklearn/
$ heroku config:add LD_LIBRARY_PATH=/app/.heroku/vendor/lib/atlas-base/atlas:/app/.heroku/vendor/lib/atlas-base:/app/.heroku/vendor/lib/

otherwise I get a:

ImportError: liblapack.so.3gf: cannot open shared object file: No such file or directory

when the application loads scipy.

Presumably setting LD_LIBRARY_PATH directly in the build back will remove that manual configuration step.

[dbrgn]

Hm, will that work without having the user-env-compile lab turned on?

[ogrisel]

It just tried to use my master branch where this patch-1 branch has been merged on a newly created heroku app and it seems to work. I don't know where the user-env-compile is set.

[dbrgn]

Alright, I'll test it next week and merge it afterwards :)

[ghost]

Has this been fixed/merged? I was able to push successfully after the issue #2 fix but now I'm getting this same import error ogrisel describes above when I try to migrate my heroku database.

[dbrgn]

@ogrisel Sorry, I was in my vacations (camping in the mountains -> no internet) when this PR was posted and totally forgot about it afterwards...

Adding the atlas folder to LIBRARY_PATH seems to make sense, but I don't quite understand what this pull request does about it. First you're adding the path, but then you're removing it again in the second commit. If you take a look at the PR diff, you can see that you're not really changing anything (besides some refactoring steps concerning references to environment variables).

In the second request, you're writing (LD_)LIBRARY_PATH should be updated by sourcing the numpy/scipy step, but you're not sourcing anything. Did you forget to push some commits?

@buildingspeak Did you try setting the Heroku config variables as described in this pull request description?

[ghost]

Yes, adding the config variable LD_LIBRARY_PATH as ogrisel posted worked for me on removing the liblapack error. I didn't use his actual fork; I just set the config variable in Heroku, pushed again, and it worked.

However, could also be the sequential installing (see my post on issue #2) that really did it; I don't know for sure and I haven't tried to clear that config variable and push again. I'm just happy to have it working! If it would help, I can clear the variable and see if I can still push.

[ogrisel]

but you're not sourcing anything. Did you forget to push some commits?

No, the numpy install script is already sourced in the current master of the buildpack.

[ogrisel]

Here it is:

https://github.com/dbrgn/heroku-buildpack-python-sklearn/blob/master/bin/compile#L158

[dbrgn]

Ah, now I get it. It is sourced in between the two changes.

Would you mind squashing your commits into a single one? With two commits, the easiest way to do that would be:

git reset --soft HEAD^
git commit --amend

Then (after checking the result with git log) do a force push to your Github branch.

[ogrisel]

Done!

[dbrgn]

Great, thank you!

@buildingspeak could you test whether it works now without having the Heroku config variables defined?

[ghost]

I unset the LD_LIBRARY_PATH config var and was able to push and then to migrate the db on heroku, so it seems to be working fine without the config var defined.

[dbrgn]

Very good, thanks for testing it, @buildingspeak.