please attach a sample survey (or group) so we can test.

In this sample file the question 201 and 202 (question group: Example 2) have an entry into the Randomization group name field ("1") - same behaviour: I see the welcome screen but can't proceed when I click on "Next"

yes, randomization groups are broken in _dev and _yii.

I'm still trying to understand how they worked in 1.91+.

buidsurveysession() does the equivalent of createFieldMap(), but then randomizes the questions within the groups. However, then createFieldMap() is called and it does the same, but without randomizing the questions. So, I'm not clear how 1.91+ avoids undoing the randomization done by buildsurveysession().

1.92 relies upon the contents of createFieldMap (rather than structures from buildsurveysession), which is not performing this randomization.

This may highlight a need to refactor 1.92 to eliminate buildsurveysession and createFieldMap, or at least avoid the duplication occuring (since buildsurveysession, createFieldMap, and EM all do large and very similar database reads when starting the survey).

survey 88881 is specifically designed to test randomization. It shows the same problem.

This is going to require some pretty major changes, so will have to wait for 2.0

Some notes:
(1) Need to re-order $fieldmap when there are randomization groups - DONE

(2) Need to change how EM figures out the gseq (group seq). Currently, uses $LEM->groupId2groupSeq array, but when randomize, must use random_gid field for an SGQA code, not the GID, to determine in which group it belongs.

(3) This also affects JavaScript - <div id='group-N'> uses GID coding, but that won't work for randomization - perhaps switch to using gseq?

Thomas, I suspect you misunderstand how the feature is supposed to work. At least comment 2 and 3 hint to that. Please get in touch before continuing.

Carsten-

To clarify those comment, here is how EM is supposed to work: find first group of questions and validate them. If none valid, find next group of questions until find at least one relevant and non-hidden question.

In general, it uses gseq to know the order of groups. Originally, it used gid, assuming that gids would be in the proper order from group_order. Now, it is clear it can't rely upon the gid variable to indicate boundaries between groups (since after randomization, questions from different groups can appear in the same randomized group). So, I need to change EM to exclusively use gseq to identify group boundaries and membership. Randomization functions can change gseq externally (e.g. create a mapping table of gseq to the qids they contain, and in the proper order.

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&amp;id=8006

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&amp;id=8007

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&amp;id=8008

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&amp;id=8011

Fixed in 1.92 revision 120323

1.92+ update released.

Fix committed to Yii branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&amp;id=8096