GitLab

So we can use this abstraction elsewhere in the RTS

Patch from sthibaul. Fixes trac #2883.

We used to generated things like:
    extern StgWordArray (newCAF) __attribute__((aligned (8)));
    ((void (*)(void *))(W_)&newCAF)((void *)R1.w);
(which is to say, pretend that newCAF is some data, then cast it to a
function and call it).
This goes wrong on at least IA64, where:
    A function pointer on the ia64 does not point to the first byte of
    code. Intsead, it points to a structure that describes the function.
    The first quadword in the structure is the address of the first byte
    of code
so we end up dereferencing function pointers one time too many, and
segfaulting.

 * Also fix a nasty bug when creating fixup code that has a cyclic
   register movement graph.

 - In the GNU assembler they mean the same thing
 - The Solaris assembler only has .skip

Following Trac #2905, we now require -XGADTs for *pattern matches* on
GADTs, not just on *definitions*.

Also I found that -XGADTs wasn't being checked when declaring type families,
so I fixed that too.

I followed the suggestion in Trac #2994, which took longer than I
expected.  As usual I did a bit of tidying up at the same time,
and improved a few other error reports.

it shouldn't be there, and might hide bugs

A bug in the constraint simplifier means that an equality can be solved
twice.  It's harmless, and will go away with the new constraint simplifier.
Hence warning, to avoid unnecessary outright failure on eg Trac #2999.

If we float a binding out of a *value* lambda, the fixing-up of IdInfo
is a bit more complicated than before.  Since in principle FloatOut
can do this (and thus can do full lambda lifting), it's imporrtant
that transferPolyIdInfo does the Right Thing.

This doensn't matter unless you use FloatOut's abilty to lambda-lift, 
which GHC mostly doesn't, yet.  But Max used it and tripped over this bug.