For the submission and comments, I mainly decided to use a `has_one` association. Based on the database schema, multiple request for comments could be allowed (i.e., for each file or submission), but this won't happen practically (since we always create new submissions and files). Hence, the `has_one` association is representing our relationship better.
* Ensure only one `pp_work_alone` event is stored.
* Disable Turbolinks for Programming Groups Work Alone, so that the implement page is requested normally. Otherwise, Turbolinks would load the page first, just to notice that it needs to reload the page afterwards to include Highlight.js for the tips.
We might receive a `RecordInvalid` error, if one part (such as the external user) is already assigned (not unique). Hence, we just retry.
Fixes CODEOCEAN-TZ
Further, we remove the "check invitation" button and extract some methods to our new ProgrammingGroups object in JavaScript.
Co-authored-by: Sebastian Serth <Sebastian.Serth@hpi.de>
Our investigation in CODEOCEAN-TV showed that we might attempt to close the WebSocket connection twice, if learners press the stop button. With this commit, we fix that issue.
We need to store or update the LTI parameters. However, this operation is not atomic and multiple requests can interfere with our database operation. Therefore, we need to retry in case the record is not unique.
Fixes CODEOCEAN-TC
* no lti session data anymore included in the function
* decided not to delete the pg_id here as it is handled in create_through_lti and implement
* the function is only called once without an exercise id so the values are now directly deleted there
* We want to identify a user that has triggered a testrun. Previously (in regular operation), only submission author who were regular users were able to start a testrun. Now, we want to prepare a future where submission authors are programming groups. Still, a testrun is triggered by an individual user and not a group.
* Further, this commit fixes some missing foreign key constrains.
* We must not close the socket manually (with `client_socket.close`), as this would close it twice. Compare: 2c83794b85/lib/tubesock.rb (L92)
* When the socket is closed twice, nginx registers a `Connection reset by peer` error: `recv() failed (104: Connection reset by peer) while proxying upgraded connection`
* Tubesock automatically closes the socket when the `hijack` block ends and otherwise ignores `Errno::ECONNRESET`: 2c83794b85/lib/tubesock.rb (L161-L162)
