@cdorin, can you have a look and if you can reproduce, maybe assign it to @gabrieljenik directly since he is also involved at this project?

This is the code that checks if a SQ is unanswered.

////////////////////////////////////////////////////////////////////
            // WHICH RELEVANT, VISIBLE (SUB)-QUESTIONS HAVEN'T BEEN ANSWERED? //
            ////////////////////////////////////////////////////////////////////
            // check that all mandatories have been fully answered (but don't require answers for subquestions that are irrelevant
            $unansweredSQs = array();   // list of subquestions that weren't answered
            foreach ($relevantSQs as $sgqa)
            {
                if (($qInfo['type'] != Question::QT_ASTERISK_EQUATION) && (!isset($_SESSION[$LEM->sessid][$sgqa]) || ($_SESSION[$LEM->sessid][$sgqa] === '' || is_null($_SESSION[$LEM->sessid][$sgqa]))))
                {
                    // then a relevant, visible, mandatory question hasn't been answered
                    // Equations are ignored, since set automatically
                    $unansweredSQs[] = $sgqa;
                }
            }

A SQ is marked as unanswered if the question is null or blank.
As the SQ are checkboxes, if none is checked, all are saved as blank in the DB.

The question here is: blanks mean unanswered?
If so, I think there could be a lot to change as I think that is taken into consideration for many other question types.

Put in other words, @Mazi, when do you expect a question should be marked unanswered?
When there is no value or when it hasn't been processed? (unseen)

Thanks for debugging this, @gabrieljenik.

I think your question is difficult to answer. My first guess would be "a single answer at each row". But for other setup it could be for each column or for the whole question. Hard to tell.
Questions which were skipped should surely not be treated as unanswered since the user never had a chance to answer them.

Questions which were skipped should surely not be treated as unanswered since the user never had a chance to answer them.
You are talking about relevance, right?

Correct, relevance/conditions. Usually there should be a difference between NULL and empty values at the DB. One of them was for indicating skipped questions the other for unanswered questions.

I think that's being taken care: skipped questions shouldn't count as unanswered.
Still, by the situation you describe on the description of the bug, it seem your example expects that empty shouldn't count for unanswered.
Am I right?

Yes, I think assuming that for a multi choice question or even an array of checkboxes, we should not assume that every single checkbox has to be checked before we treat that question as answered. That would be a little too much.
@c_schmitz, what's your opinion on this?

My opinion :

skipped questions shouldn't count as unanswered.

+1

we should not assume that every single checkbox has to be checked before we treat that question as answered.

+1 for multiple choice and multiple choice with comment
Unsure for array or multiple text
Unsure for array of text + array of number

I think we need to go to the roots:

• What un-amswered should reflect?
• What's the feature for? How the user will take advantage from it?

From there:

• Mandatory questions not answered should count as aunanswered?
• Empty not mandatory answers?
  • Empty text boxes?
  • Non checked checkboxes?

From the code, I think "unaswered" means empty responses.
I am not sure how usefull is for participants.

I think we need to go to the roots:

Maybe : «if this question was mandatory it was invalid»

This mean : all array answered, checkbox count > 1 . I don't know for checkbox with comment.

Summing up yesterday's DEV meeting, most people recommend treating question as completed as follows:
a) multi choice: at least a single checkbox was checked
b) array of checkboxes: at least a single checkbox was checked at every row

@c_schmitz, what do you think? Should it be implemented it like this?

See https://github.com/LimeSurvey/LimeSurvey/blob/70f326fc531004d046a434b953d3040dc0f67a20/application/helpers/expressions/em_manager_helper.php#L6663

Not answered must follow mandatory rules …

@Mazi, I missed yeterday's debate. Sorry.
On the situation that you are saying:

• If there is a list of diseases where you need to check the ones that you had.
• If you are healthy and don't check any of those, the question will be marked as "unanswered" ?

I feel the current behaviour is OK as it is: Those checkboxes are marked as answered.

A new example:

• We replace the disease checkboxes by yes/no radio buttons with default in NO.
• A healthy person will still go through the question, will not change any default and click submit.
• That question will be treated as answered.

So, the dilema is: A non selected checkbox counts as answered? I think yes.

@gabrieljenik, as for the diseases example: That looks correct. If a checkbox was checked the question should be treated as answered, otherwise not.
Usually in this case one should have an "I am healthy" option as well which should also be marked as exclusive option, but that is another topic.

We should NOT add any code to replace a checkbox by radio buttons with yes/no labels. That would be a huge change.

Then : there are no issue finally ?

THere still is an issue. For the array with checkboxes the question onyl gets treated as completed if ALL (!) checkboxes are checked. That's surely wrong.

OK :)

Array number + options checkbox :

Specific in https://github.com/LimeSurvey/LimeSurvey/blob/70f326fc531004d046a434b953d3040dc0f67a20/application/helpers/expressions/em_manager_helper.php#L6728

That looks correct. If a checkbox was checked the question should be treated as answered, otherwise not.

Actually I meant the other way arround. If non checkbox is checked, still, that should be treated as answered.
That is, I believe, the current behaviour.

We should NOT add any code to replace a checkbox by radio buttons with yes/no labels. That would be a huge change.

Agree. It was an example of how checkbox compares to radio button.
Checkbox = Radio button with default answer.

For the array with checkboxes the question onyl gets treated as completed if ALL (!) checkboxes are checked

OK. Let me pickup on this specific part of the bug then.

@DenisChenu @Mazi What if we add a new case type in this line?
https://github.com/LimeSurvey/LimeSurvey/blob/70f326fc531004d046a434b953d3040dc0f67a20/application/helpers/expressions/em_manager_helper.php#L6816

For Numeric Array question types with Checkbox layout, if is enough for mandatory, we flag it as answered.
If not, we flag it as anunserwed.

What do you think?

I think this is the first part of a broader solution for other quesiton types, but let's start with this small change.
What do you think?

@gabrieljenik : it seems it does that for multiple checkbox, then : yes : seems a good idea :)

PR: https://github.com/LimeSurvey/LimeSurvey/pull/1652

I guess we all agree that code needs major refactroning, right? :)

Like WHOLE ExpressionManager ;)

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&id=30783