Incorrect order of choices in Poll module.

@vitalsouto, assigning issues to yourself means that you will work on (and create a patch for) them :)

Anyway, after I struggled a bit with FAPI, I found out that the problem lies in our call to element_children() from theme_poll_choices().

I will re-roll http://drupal.org/node/939880#comment-4941348 because reversing keys in hook_validate() is not needed anymore after this patch.

Patch solved all above problems for me on Drupal 7.8. Thanks.

This is the correct fix... RTBC

element_children() defaults sorted to TRUE, so we need to explain why we're setting it to FALSE explicitly. Also this makes me wonder if there is somewhere earlier in the process it could be set to FALSE than the theme function? Why not in the form itself?

If that's already been rejected it'd be worth adding in a comment.

Lastly since we're using element_children() directly in the theme function, it'd be possible to use the $sort parameter to ensure these get sorted instead of the #sorted flag, don't have a preference though.

Should we be using element_children in theme_poll_choices() to sort the poll choices? Can't we just add them in correct order in our form definition? When creating a new poll, the 2 choice inputs that are displayed have the same weight value. I created a patch that changes this.

After a lot of thinking on this, I agree that increasing the weight for each choice is a good option as well, but should be done before adding the choice.

Our current tests don't catch this bug because in _pollGenerateEdit() we specifically assign weights to new choices, and it's obvious that will always work :)

The test-only patch attached removes the hardcoded weights and uncovers expected failures: the initial two choices are reversed. The commented test is intentional, because we're messing with weights in poll_form() and that brings more failures than needed for the purpose of this test-only patch. Anyway, It's untouched in the full patch and it passes on my local dev.

Tests behaved as expected, setting to RTBC based on the idea from #8 and because my patch only fixed the current tests.

Looks good. Committed/pushed to 8.x, moving back to 7.x for backport.

7.x patch.

Committed and pushed to 7.x, thanks! Looks like a nice test clean-up as a bonus. :)