stale hie files make weeder reporting inaccurate in certain renaming situations

[codygman]

In CI we cache ~/.cabal from main and each feature branch begins with a cache of the most recent main build. Then each successive CI build caches to a key with the feature branch name.

We also use require-hs-files

I either have a bug from:

• recompilation
• some other weeder bug
• false positive mentioned in Don't analyze HIE files when hs file doesn't exist #82 because of names that don't use package names in their module namespace like test/Spec.hs or executables/Main.hs

EDIT: Our module name's don't mention module namespace. What's weird is we had no issues until the improved recompliation avoidance in ghc 9.4 that IIRC uses timestamps.

If it's the last one, it might be fixed by re-implementing #82 with HieDB.Query.getUnreachable)

Otherwise to the best of my memory, here's the series of events that seems to lead to this (haven't reproduced yet, but these steps might):

• build project calling a function A.f in main
• copy the dist-newstyle directory to to simulate cache storing results of this commits build
• create a weed by removing reference to A.f
• remove A.f
• restore the dist-newstyle directory from above, simulating cache restore from previous commit
• build project
• run weeder --require-hs-files