Split the CUDADriver recipe into a download and a munki part.

[mkuron]

Split the CUDADriver recipe into a download and a Munki part. That way, I can use it with the Importer I wrote for MunkiServer.

[jessepeterson]

Sorry I didn't see this before now (wasn't watching the repo). It appears you've got non-download things in the download recipe. I think only the three processors (CUDADriverURLProvider, URLDownloader, and EndOfCheckPhase) ought to be in the download.

[mkuron]

I wasn't sure about whether determining the version should be part of the download or part of the import. The way it is in my pull request made more sense to me as determining the version is generic to all importers.
I've looked at some of the sample recipes and you're right, this is commonly done in the import/package build recipe and not in the download recipe.

I've noticed that https://partners.download.nvidia.com/activation/cuda_update_macos.xml contains the exact version number in the kServerVersion key, so how about simply exporting that as the version variable straight from the Python script and dropping all the pkg extraction stuff to find the version number?

[jessepeterson]

The way it is in my pull request made more sense to me as determining the version is generic to all importers.

This makes sense — I hadn't considered that. I was focused on the way it's usually done.

so how about simply exporting that as the version variable straight from the Python script and dropping all the pkg extraction stuff to find the version number?

This introduces an element of trust that Nvidia know what they're doing. I would think to the spirit of autopkg it would want to get what the package itself thinks it's version is (since that's whats going to be installed). However I'd be willing to give it a try. Perhaps @timsutton might have a perspective (I think he uses this package, too). All that being said it'd be nice to skip all the pkg extraction crap.

[mkuron]

I'm guessing Nvidia's auto-updater checks that exact same key against the installed version. So as long as that XML file stays there and keeps its current format (both of which we already depend on), it should be safe to depend on the kServerVersion being sensible.

[jessepeterson]

Okay, you've convinced me. :) Can you update the pull to reflect all that?

[mkuron]

Here's the commit for using kServerVersion.

On a different note, the urllib2.urlopen() call appears to have stopped working recently because Nvidia's server no longer accepts mixed SSLv2 and SSLv3 negotiation. curl is also affected by this if you don't force SSLv3:

$ curl 'https://partners.download.nvidia.com/activation/cuda_update_macos.xml'
curl: (35) Unknown SSL protocol error in connection to partners.download.nvidia.com:443 
$ curl -2 'https://partners.download.nvidia.com/activation/cuda_update_macos.xml'
curl: (35) Unknown SSL protocol error in connection to partners.download.nvidia.com:443 
$ curl -3 'https://partners.download.nvidia.com/activation/cuda_update_macos.xml'
<?xml version="1.0" encoding="UTF-8"?>
[...]

As urllib before Python 3 does not support customized openers, I figured using curl through subprocess would be easiest. Alternatives like using NSURL* via the PyObjC bindings or using the socket and ssl modules directly would have required dozens more code lines.

[jessepeterson]

Yikes, sorry for not seeing updates on this! What's the current status? The CUDADriver package appears to be downloading fine now — perhaps they relaxed the requirement?

[mkuron]

This probably depends on the system Python version. Most web servers now disable SSLv3 for security reasons, and probably only newer Python versions (OS X 10.10 and higher maybe?) default to TLS. So I'd suggest keeping the curl for better compatibility and in case Nvidia ever decides to make another change to their HTTPS configuration.

[timsutton]

For me the recipe currently works without curl. A couple comments to follow in @mkuron's most recent commit..

[mkuron]

Here's the commit that merges the latest changes from master. I think the pull request is ready to merge now.