Pulp

Based on discussions in IRC, we're going to leave the current behavior as-is and close this WONTFIX. Adding the discussion for posterity:

<dalley> dkliban, as resident binding expert, could you weigh in on this idea? https://pulp.plan.io/issues/6496#note-18
<dalley> will that cause problems?
<dkliban> i don't quite follow
<dkliban> is the suggestion to add a field that's called deleted-resources?
<dkliban> or is the suggestion to remove old tasks?
<dalley> right now the deleted "created resources" become "null"
<dkliban> yep
<dalley> ggainey is saying its difficult to actually get rid of them, but we could replace "null" with "deleted resource"
<dkliban> oh
<ggainey> dkliban: yeah - the prob is when the task lives, but the thing(s) it created are deleted
<dkliban> would we use a database trigger or soemthing?
<ggainey> and right now, we just say 'null' for a missing resource
<ggainey> dkliban: it would have to be on every table
<dkliban> dalley: it wouldn't really be a problem ... the created resources field is a list of strings
<dalley> I don't think we should be messing w/ that, especially in combination with generic foreign key where it's all quite abstracted to begin with
<ggainey> dkliban: the change I had in mind is output-only - the table and model stay as they are, the task-output just says "<deleted resource>" where it currently says "null"
<dkliban> ggainey: cool 
<dkliban> that makes more sense than what i was saying
<dkliban> the bindings should have no issue with this
<ggainey> dalley: dkliban : tell you what - I'm'a submit a PR, so we have the conrete code in front of us, and if we decide to do someting else (or nothing!), I can just close it
<dalley> does katello try to do anything w/ created_resources that needs to take this into account?
<dkliban> i don't know
<ggainey> same
<ggainey> it's def an obscure-ish edge case
<ggainey> (which migration steps into all the time, of course :) )
<jsherrill> we do record created resources from some pulp tasks
<jsherrill> like after creating a new version
<dkliban> ggainey: earlier today in #pulp, wibbit was asking about such a scenario. however, it was not about the migration plugin but the repository version 'base_version' field
<jsherrill> so we could add some logic to check for nil values rather than an empty array
<ggainey> dkliban: yeah, I took a stab at answering him, but it was pre-coffee, yours was better
<dalley> it's definitely easier to process null than a random string
<jsherrill> yeah, i'd rather not have a random string
<ggainey> jsherrill: yeah, that's the crux of it - right now, you get an array which may have null entries, which is not create for humans reading it but better for code to rely on
<jsherrill> especially when an href is also a string
<ggainey> yeah
<jsherrill> its not possible to remove those nil's from the response?
<jsherrill> or is that not desired
<ggainey> ^not create^not great
<ggainey> jsherrill: that's exactly the problem
<dalley> I kinda think we should just leave it as-is until we can fix this API in an eventual Pulp 4
<ggainey> the CreatedResource modle-entries *still exist* - it's the things they point to that are gone
<ggainey> dalley: I am def leaning that way myself, aye
<jsherrill> ggainey: yeah i guess i'm asking if its not possible to easily strip/ignore ones in that situation
<jsherrill> and its fine if its not, i don't know the code ;)
<ggainey> jsherrill: as it turns out, the answer is "argh" :)
<jsherrill> hehe
<ggainey> I spent some time inside serializers trying to figure out the right thing, and if there is a way, it's def nonobvious
<jsherrill> gotcha
<ggainey> the conversation is essentially "Hey, how many CreatedResource entities do we have here?" "One." "Great! Give me the Thing it points to!"  "It's gone."  <sadness ensues>
<jsherrill> well nil is def. preferred to random string from me!
<ggainey> ok
<ggainey> dalley: close this issue as WONTFIX?
<ggainey> I mean, I will, I'm asking if you're ok w/that
<dalley> ggainey, if there's no way to filter out the dead entries, then yes I think so
<ggainey> not that I could find, the CreatedResource serializer is called per-row, and whether it returns something or not, "an entry" is going to show up in the many-to-many list because the actual model does in fact have a row.
<ggainey> I can/we do change what the row displays-as, but it's too late to hide that it exists
<ggainey> dalley: anyway - I will close as WONTFIX. I think I'll throw this irc discussion in there as well, so someone else doesn't have to go thru the same confusion I did trying to figure out what was up