Conversation
|
So the errors happening on this one are due to the fact that it's legal (though unusual) to query the modulestore for a CourseKey where nothing is set but the version_guid (since that points to a unique structure document): I don't know why on earth we would be pulling this value from the |
386dd61 to
0c50196
Compare
|
@bradenmacdonald, @kdmccormick, @feanil: Please take a quick look at this fix for the release blocker. I was planning to push this to #37825 because that's the PR that should eventually merge (it has a lot of useful discussion). But I don't seem to have permission to push to that branch, so I'm putting the code up here for now. |
|
Oh dear. I wish we didn't have such flexible From a quick look, this seems sane and reasonable, but I am unclear how the partition services changes fix the bug nor how that was being used. |
|
@bradenmacdonald: The What this would have meant in terms of real-world impact is that if you made a publish and then tried to do an access check on another piece of content that had its runtime created before your publish, that access check would blow up. Which isn't a common case, but could happen. |
I'm experimenting on top of @marslanabdulrauf's #37825 to see if I can get the test failures ironed out without having to change the outward behavior of modulestore.
(And hopefully without breaking other stuff.)