Test in ci unwanted changes over hackage index GPD check

[jneira]

• From Enable hackage tests for windows #8055 (comment)
• We can get how any Cabal version affects GPD checks with cabal run Cabal-tests:hackage-tests -- check:

D:\ws\haskell\cabal>cabal run Cabal-tests:hackage-tests -- check
Up to date
Reading index from: D:\cabal\packages\hackage.haskell.org\01-index.tar
157946 files processed
42067 files have lexer/parser warnings
115687 files have check warnings
1118 build impossible
7980 build warning
51952 build dist suspicious
36949 build dist suspicious warning
17688 build dist inexcusable

• Afaics those numbers are not checked in ci and it would have revealed regressions as the caused one by handle conditionals in duplicate module checks #7616
• We could leverage it:
  • Regenerating the stats in a known file using the last Cabal master version and a fixed hackage index. Automatically via makefile or using another tool.
  • Making Cabal-tests:hackage-tests -- check generate stats with HEAD and the same hackage index and failing if they differ in any way with the saved one (so we would check if we are being too restrictive or too loose)
  • Forcing the developer update numbers on purpose to make the test pass (or revise their code)

[phadej]

Recall that Hackage index evolves. New upload may produce new check warnings.

[jneira]

Recall that Hackage index evolves. New upload may produce new check warnings.

Yeah, thanks for pointing it out, i forgot to note:

• Regenerating the stats in a known file using the last Cabal master version and a fixed hackage index

which then would be aligned with

• Making Cabal-tests:hackage-tests -- check generate stats with HEAD and the same hackage index

(Updating the description)