*A* developer, but not the developer of the package with the loose upper bound or the package that refused to compile with incomprehensible errors because of it, and generally not in a position to recognize the reason for the errors because they don't know the internals of the package they're trying to use. And I am certain of this because I'm sitting in #haskell fielding questions from them multiple times a day when some package gets broken by an overly lax or missing upper bound.

Also note that overly strict versioning certainly also leads to breakage --- but it's reported clearly by cabal as a version issue, not as ghc vomiting up unexpected errors from something that is presented as a curated package that should build without problems.

--