GitLab

Fixes #7780.

We float unsolved equalities from underneath a 'forall', to
help solve them, in TcSimplify.floatEqualities.

It's regrettably delicate though,as this bug shows. I'm not
happy with the new code; but there are copious notes; see
Note [Float equalities from under a skolem binding].

We care careful not to construct a canonical equality whose LHS and RHS
have incompatible kinds.  (This is one of the invariants of a canonical
equality.)  See Note [Equalities with incompatible kinds].  However,
what I had not dealt with is when LHS and RHS *look* as if they have
different kinds, but after zonking they become the same.  Bad!

(This led to an ASSERT failure in the test for Trac #7176.)

Although it's not technically needed, it's less confusing if the vanilla
way works when people use the sample build.mk.

The build system thought that the RTS built more library files than
it actually did, and installation failed when we tried to 'strip'
one of these non-existant files.

Signed-off-by: Austin Seipp <aseipp@pobox.com>

This reverts commit 0b72347d.

This reverts commit e7ca1324.

Signed-off-by: Austin Seipp <aseipp@pobox.com>

Signed-off-by: Austin Seipp <aseipp@pobox.com>

  * the new StgCmmArgRep module breaks a dependency cycle; I also
    untabified it, but made no real changes

  * updated the documentation in the wiki and change the user guide to
    point there

  * moved the allocation enters for ticky and CCS to after the heap check

    * I left LDV where it was, which was before the heap check at least
      once, since I have no idea what it is

  * standardized all (active?) ticky alloc totals to bytes

  * in order to avoid double counting StgCmmLayout.adjustHpBackwards
    no longer bumps ALLOC_HEAP_ctr

  * I resurrected the SLOW_CALL counters

    * the new module StgCmmArgRep breaks cyclic dependency between
      Layout and Ticky (which the SLOW_CALL counters cause)

    * renamed them SLOW_CALL_fast_<pattern> and VERY_SLOW_CALL

  * added ALLOC_RTS_ctr and _tot ticky counters

    * eg allocation by Storage.c:allocate or a BUILD_PAP in stg_ap_*_info

    * resurrected ticky counters for ALLOC_THK, ALLOC_PAP, and
      ALLOC_PRIM

    * ...

ForceSpecConstr will now only specialise recursive types a finite number of times.
There is a new option -fspec-constr-recursive, with a default value of 3.

Fixes a bug which meant that they weren't getting installed.

I think overall the new approach is simpler. Rather than unpacking
the libffi.a and putting the .o files into our libHSrts.a, we just
use the libffi.a.

This change also means that when compiling programs for the dyn
way, they get explicitly linked against libffi.so (rather than
relying on librts.so being linked against it). This might
fix a problem on FreeBSD, where programs cannot find libffi.so.

We no longer pass -hisuf/-osuf flags to "ghc -M". Doing so didn't really
make sense with the way the -dep-suffix flags now work.

There's no point splitting objects when we're going to be dynamically
linking. Plus it breaks compilation on OSX x86.

We were only setting an RPATH for the RTS DLL on Linux, but as far
as I can see we should be doing it for all ELF OSes. Hopefully this
will fix the problem where the installed ghc-pkg can't find libffi.dll
on FreeBSD.

Spotted by Bill Tutt

Pretty much everything does now, so list those that don't instead
instead of listing those that do.