You can turn it into an optimization problem instead if you try to minimize the wasted votes for the entire state. This ensures that you don't get states with districts that all have huge -- but equal! -- efficiency gaps.

In the same way, however, nearly any law creates winners and losers compared to a slightly different, hypothetical version of the law.

I think the intention behind creating geographical districting was to allow geographically dispersed interests to gain political power that would otherwise end up concentrated in population centers.

This is the same basic idea as the Electoral College -- create a definition of consensus that requires not just population-dominated consensus, but geographically diverse consensus.

The tension about what constitutes legitimate consensus mirrors the sort of debate in the cryptocurrency community about proof of work vs proof of stake vs other methods of weighting the consensus algorithm.

Redistricting is thus somewhat analogous to developing an ASIC that exploits a quirk of the consensus algorithm. Of course interest groups are going to want to exploit whatever loopholes they can find that are within the rules.

But in the case of redistricting we can easily see the folly of it. The impact of minority votes all being lumped into one district is that all the politicians in the other districts can safely ignore minority concerns, whereas if each of the other districts had 10% minority voters, those concerns would not be something all politicians could ignore.

But on the other hand, with a single minority-dominated district, we're much more likely to get minority politicians winning at least one seat, whereas with minority votes spread thinly over all the other districts, this would be less likely.

We can impose whatever restrictions on shapes we want, but ultimately we are just shifting the balance of power between geographically-general interests and geographically specific ones.

Since redistricting is so embarrassing, though, I'd support limiting district shapes to regular polygons whose height and width are within 10% of one another.

In fact, given modern technology, you could easily draw the boundaries after the election. If you did, I bet they'd look a lot more "natural" than the results of current GOP gerrymandering.

They can cover the entire state's interest through compromises as oppose to infighting for district vs. district within a state.

The universal rule of single member districts is relatively recent, and it was a blunt (though effective) tool to deal with the threat of a form of super-gerrymandering by which states would assign some or all of their representatives to state-wide at-large seats so that the statewide majority would elect them, as a means of preventing the existence of majority-black districts in majority-rule states.

Multimember districts with a proportional (even candidate-centered, like STV) electoral system rather than plurality or majority-runoff deals with both that problem and the problem with single-member FPTP districts.