(The non-ascii chars are “ and ”, versus ascii ".)

Florian, although you did not select a 'Type', selecting multiple versions implicitly claims that the current behavior is a bug.  I believe R.David has explained that it is not, even if sub-optimal.  Do you want to
A. Argue on the basis of some claim in the docs that this really is a bug.
B. Close this issue as 'Not a bug'.
C. Turn it into an enhancement issue for 3.10 by calling decode in the appropriate place.  Is so, you might first try making the change in your code after finding the appropriate place and see if the improvement is worth the change.