Don't import Prelude in the generated Build_doctests modules.

[quasicomputational]

Prelude is a magical module name, and it may be useful for users
to define their own Prelude modules which may not coincide with
the Prelude from base.

[phadej]

What restricts people to redefine Data.Maybe?

I'm very skeptical about redefining Prelude, given where Build_doctest is intended to be used. I'm not against this change but I'm curious what's the setup where this is required?

[quasicomputational]

Data.Maybe isn't magical in the same way that Prelude is. Re-defining Prelude is the only way to bring things in scope without an import. This is less of a problem with doctests because $setup stanzas exist, but it's a configuration I can imagine existing.

[phadej]

@quasicomputational note that the Build_doctest is used to compile doctests test-suite executable, a doctest like executable. It's not the code which doctest ever sees & executes & tests.

You'd need to have contrived:

test-suite doctests
  type:                exitcode-stdio-1.0
  main-is:             doctests.hs
  build-depends:
    -- note: not `base`
    my-better-base,
    doctest >= 0.11.3 && <0.14,
    QuickCheck ==2.10.*,
    simple-example,
    template-haskell

  ghc-options:         -Wall -threaded
  hs-source-dirs:      tests
  default-language:    Haskell2010

to break in way your change is required. I don't want to encourage this use case. Generated code assumes dependency on base