New principle: only extend problematic and non-consensus features if the extension acts to fix the problems

[jyasskin]

This was brought up in our discussion of Protected Audiences Bidding and Auction Services API.

This might fit into https://w3ctag.github.io/design-principles/#leave-the-web-better.

The idea is that improving a problematic feature will tend to increase use of that feature, which increases any problems caused by use of the feature, so the improvements should compensate for the increase in use with a decrease in the problems-per-use.

Extensions shouldn't have to completely fix the problems in order to be a good idea, just have some indication that the total amount of problems will go down.

[martinthomson]

So the concern from our discussion was that an extension to something that has some sort of bad effect might make it more attractive to use the thing and thus have the bad effect occur more often. The suggestion was that maybe someone building an extension has an obligation to make the situation better (perhaps by working to mitigate the problem) in addition to building their extension. There isn't an expectation that someone completely solve a problem (that might be impossible), but incremental improvements might be OK. Circumstances will differ and we'll have to make a judgment call.

Protected Audience is not obviously a good example of this because the proponents of Protected Audience do not always agree that these are negative effects. A better example might be enhancements to window.alert(), which is acknowledged to be bad as a stop-the-world API. Those improvements wouldn't be acceptable without doing something to make alerting less bad.