GitLab

libraries built by stage2 need all their bits in the bindist too.
We were testing (stage == 1) rather than (stage /= 0).

It's now in the nofib repo.

Now configure looks for it as an installed program instead.

I'm less convinced this is a good idea now.  But it does avoid
rebuilding most of stage1 when you pull and reconfigure.  Better
solutions welcome.

we also have $1_$2_CONFIGURE_OPTS, which is used.

This fixes build problems of the form
  ghc-cabal: Version mismatch between ghc and ghc-pkg

A long-standing improvement to the error message for kinds. Now instead of
    Expected kind `* -> *', but `Int' has kind `*'
we get
    The first argument of `T' should have kind `* -> *',
    but `Int' has kind `*'

Ha!

Record updates are amazingly hard to typecheck right.  This is one place
where GHC's policy of typechecking the original source is much harder than
desugaring and typechecking that!

Anyway, the bug here is that to compute the 'fixed' type variables I was
only looking at one constructor rather than all the relevant_cons

Test is typecheck/should_compile/T3219, which GHC 6.10 barfs on (with Lint).

When using shared libs we should link each lib against its deps. This allows
the dynamic linker to chase dependencies and means we do not have to specify
all indirect dependencies (as we must do for static libs).