>> But even though we are using GNU iconv, it turns out that in the old
>> code iconv_elides_question_marks was set to 0.
>Right. I just committed 8ed8876c6, which basically restores the old
>behavior of the test in that respect. It works on Mac OS X. On all
>platforms, it skips the tests that rely on iconv if it's not enabled.
Great! Okay, that passes all of the tests now, and I've cherry-picked the
relevant changes to the 1.7 release branch.
Should I just spin another 1.7.1 release candidate?