GitLab

Otherwise :print only works for local identifiers, not global ones.
In fact it was silently failing, so I fixed that too.

This patch changes the interface to lookupInstEnv, so that it
returns a pair (Instance, [Either TyVar Type])
rather than    (Inst,     TvSubst)

There is no functionality change, but the interface is tidier,
and closer to lookupFamInstEnv (when Manuel has changed that too).
The [Either TyVar Type] gives the type(s) at which the dfun should
be instantiated.  We need an Either because it might be instantiated
freely: see Note [InstTypes: instantiating types] in InstEnv.

(This might be a pattern we want to use elsewhere too.)

This patch simplifies the code slightly, and simultaneously improves
full laziness by floating allocations (lambdas, constructor apps) out
of loops.

See Note [Escaping a value lambda] in SetLevels, which explains.

There's a test that shows it up: simplrun009

This relevant to SpecConstr, because a call looks like
	f lvl
instead of
	f (\x. blah)
and the latter is easier to match in a robust way.

:hist <n> displays at most the last <n> steps (default 20).

We were freeing the hash table storage with exitHashTable() before
calling endProfiling(), which uses hash tables.

Seems to be a bug introduced by code to free the memory allocated by
the heap profiler.

- We need to keep pairs of (f, f_CC) in VectInfo as it is difficult
  to obtain Names from OccNames (of imported modules) in Core passes.
- There is a choice of keeping Names or Vars in VectInfo.  We go with Vars
  for now; mainly to avoid converting between Names and Vars repeatedly for
  the same VectInfo in other than one-shot mode.

  Again goes to the HEAD straight away to avoid conflicts down the road.

We had no way of getting the right value of HscTarget to use to
request object files as output.

This patch generalises the let-floating transformation in a way
suggested by Roman and Manuel when doing closure conversion.

There are extensive comments in Note [Floating and type abstraction],
which begins thus.  Consider this:
	x = /\a. C e1 e2
We'd like to float this to 
	y1 = /\a. e1
	y2 = /\a. e2
	x = /\a. C (y1 a) (y2 a)
for the usual reasons: we want to inline x rather vigorously.

(Further commennts follow in SimplUtils.)



The implementation is not hard; indeed it used to be in GHC years ago.
I removed it thinking that full laziness would achieve the same
effect, but I'm not sure it does; and in any case it seems more direct
to do it here.

The transformation should not make anything worse, so yell if 
you see anything unexpected happening.

This size-threshold flag is for both liberate-case and SpecConstr.

Replaces -flibereate-case-threshold.

For some ancient reason, TcType.isStringTy was treating newtypes as
transparent, which is quite consistent with isIntTy, isBoolTy etc.  
(I think the reason is that isStringTy was written to go with isFFIDotNetTy,
which deals in representation types.)

Anyway, this inconsistency is Utterly Wrong when called from 
Inst.shortCutStringLit, and that made tc224 fail.  I can't think how
it ever succeeded.  Maybe it never did!

Anyway this fixes it. It may be that .NET FFI calls are not quite as
permissive, but they are almost certainly broken in more serious ways,
so I'm going to jump that bridge if we come to it.

All primitive types were getting PrimAlts, where actually case
expressions on 'Any' should get a PolyAlt.  The result was that seq on
Any compiled into a no-op, which caused :force to go into an infinite
loop.

  I am putting this patch (as the previous VectInfo patch) straight away
  into the head to avoid the kind of merging disaster we had with the FC
  branch.  The patch does not interfere with any other functionality and
  hence should cause no harm in the head.

arrows is the only library in libraries/extra-packages not built when present

This should result in successful program termination. Not doing this
breaks checking the exit code of  runghc Setup.lhs build  for example.