> Considering the huge popularity of these SO questions, I don't think this should be reverted [...]

As I understand it (and, again, I make no claim to be an argparse expert), there does not seem to be one absolutely correct answer here; there has to be a tradeoff.  If we revert the change in default as in PR 6919, users porting code from 2.7 will continue to run into the unfortunate change in behavior introduced in 3.3.  But, with the reversion, those users are no worse off than they were before: the existing workarounds, like those in the cited SO answers, still apply.  And it's a one-time annoyance for them, along with all the other changes they may need to make to port to a current Python 3.x.  Whereas, if the change is not reverted, then we introduce a new incompatibility to a new class of users, that is, those upgrading from Python 3.3 through 3.6 to 3.7, generating a new set of SO questions, etc.  That seems to be making a less-than-ideal situation worse.  So, as release manager, I continue to think that the reversion (PR 6919) should go in to 3.7.0.  (For 3.8 and beyond, it would be great to have at least one core developer take responsibility for argparse enhancements.)