> Doesn't quickly adding a deprecation warning break building any of those 1000+ packages with -Werror?
It appears that
https://github.com/haskell/pvp/issues/12 is close to being resolved[1], and as hvr mentioned Hackage already prevents packages with -Werror from being uploaded, so this shouldn't be an issue for packages already uploaded. But yes, if someone is using unordered-containers and building locally with -Werror then the build would fail.
That being said, if the plan is to deprecate it then I don't think there's a better way than marking it DEPRECATED. We could go with a "soft" deprecation--remove function comments and replace with "This function is deprecated, use findWithDefault" and make some announcements--but I doubt many people would act on that so it would just be kicking the can down the road.
> I'd support adding the new function, but am very reluctant to force any quick changes.
Glad to hear! It would be nice if there was a way of getting information about how often this particular function was used so we could get a better estimate of how many packages would truly be affected. And I completely agreed that we should do everything we can to not break people's builds, but hopefully not at the expense of improving APIs.