Import unused sample

[matiasgarciaisaia]

Let users import into a survey all the uncontacted respondents from another finished one.

The source survey must be already terminated, and has to belong to the same project as the target one.

surveda-import-sample.webm

Error handling:

Fixes #2389

[ggiraldez]

Looks good as far as I can evaluate with minimal Elixir and completely rusted redux/react knowledge. Left a small suggestion on some query, but it's probably not too important.

[ggiraldez]

I suppose this could also be a select ..., count(*) as respondents from ... where r.disposition = 'registered' which may be more efficient. I'm not sure about the Ecto syntax though.

[matiasgarciaisaia]

The Ecto syntax is not a problem:

        from s in Survey,
          left_join: r in Respondent,
          on: r.survey_id == s.id,
          where: s.project_id == ^project.id and s.state == :terminated and r.disposition == :registered,
          select: %{survey_id: s.id, name: s.name, ended_at: s.ended_at, respondents: count(s.id)},
          group_by: [s.id],
          order_by: [desc: count(s.id)]

But then you don't get the surveys that have 0 available respondents listed. I'd rather explicitely show that there are no available respondents rather than ignore the survey at all.

I could, though, add a comment on the code explaining this.

There's probably another query that can be written for that, but not sure it'd be better.

[ggiraldez]

Good point. You can use a left join for that, and count(r.id) or any other field that is NULL when joining an empty set.

[matiasgarciaisaia]

Ohhhh! The key is to add the r.disposition = 'terminated' condition to the JOIN instead of the WHERE (third query in the screenshot), so the join is made with an empty half-row. When the condition is on the WHERE (second query), you first match surveys and respondents, but then discard the joined row altogether for not satisfying the condition.

[ggiraldez]

This may merit a comment. Why is adding a respondent group to the survey updating an attribute of the survey?

[matiasgarciaisaia]

That's a business rule - if we're importing contacts from a deanonimized survey, this one is now also deanonimized.

I guess we can't compute that value from the respondent groups "at runtime" in case one can "game" the system by adding/removing respondent groups somehow - but I'm not really sure about that. I mostly followed what we're doing when adding respondents from CSVs.

[ggiraldez]

That's fair. I guess what threw me off was mostly that the updating function is in the respondent group file, but it updates the survey. I see that the module is RespondentGroupAction, but I still find it a bit confusing. Also, how are incentives related to anonymization? Is that legacy naming?

[matiasgarciaisaia]

No, it's not legacy. Incentives distribution is made by having the user download a list of phone numbers, and that's the only way for a user to get phone numbers out of Surveda. Those numbers already responded surveys, so there's potential to link responses to phone numbers - breaking the anonymity.

There might be a better way to model this "tainting" of the respondent group, but I'm not sure which one that'd be.