This is very odd---the example on the ticket does not use `Typeable` at all.

I looked through the diffs of the patch and I can't see anything obviously wrong.

Also, the `Typeable` code only touches modules that are in the front-end, and when I compile without optimizations, the example looks reasonable.

I wonder if the problem is related to some odd interaction between simplification rules and representational equities?

On Sat, Mar 21, 2015 at 4:35 AM, Herbert Valerio Riedel <hvriedel@gmail.com> wrote:

On 2015-03-21 at 08:21:20 +0100, Herbert Valerio Riedel wrote:
> On 2015-03-21 at 07:56:32 +0100, Neil Mitchell wrote:
>
> [...]
>
>> 3) I tested with GHC RC1 and GHC RC2, both of which were fine. The fact no
>> one else hit this with RC2 might just be because its a very recent
>> regression.
>
> We -- and by that I don't mean myself... :) -- could git-bisect between
> RC2 and RC3 here (semi-)automatically (i.e. maybe unattended if it's
> scriptable) if your test-case (even if it's not minimal) reliably
> triggers the bug...

I scripted up a test and git-bisected between RC2 and RC3, and the
following commit is the one where `shake-test oracle test` starts
failing

http://git.haskell.org/ghc.git/commitdiff/6f46fe15af397d448438c6b93babcdd68dd78df8

...which sadly is a rather large patch :-/

_______________________________________________
ghc-devs mailing list
ghc-devs@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs