GitLab

Its really a testsuite-framework bug that these mentions of 'opt' have
lingered so long.  They should have been rejected as a non-existent way.

Now instead of saying, e.g.
    namebase_if_compiler_lt('ghc','6.9', 'foo-6.8')
you say
    if_compiler_lt('ghc','6.9', namebase('foo-6.8'))

* gadt/lazypatok fails
* tcfail167's error message gets worse; ie, no more
  "Inaccessible case alternative: Can't match types `Char' and `Float'"

Both of these are minor regressions that will be addressed in due time.

Revert to getting an error from the renamer

* The error message has become less informative, but I don't see how to avoid
  this easily.
* The error message is now also different to that of 6.8.  I am not sure how
  to handle this case.  Can we have different stderr files for different
  compiler versions?  Or shall we simply merge the corresponding changes
  into 6.8?

The rationale for these new behaviors is as follows:

We now allow existentials to be unpacked in local (but not top-level!)
ValBinds, so some pattern signatures that used to fail in the renamer
now fail in the type checker (if the type of the body mentions the
unpacked type var):
   tc141(normal)

We don't always print HsWrappers in debug mode:
   TH_exn1(normal)
   TH_exn2(normal)

Sections are parsed (and therefore printed) differently:
   rnfail019(normal)
   read023(normal)

Character numbers in pattern printing are more precise:
   tcfail067(normal)
   tcfail124(normal)
   tcfail125(normal)