GitLab

Eta reduction was wrongly transforming
	f = \x. f x
to
	f = f

Solution: don't trust f's arity information; instead look at its
unfolding.  See Note [Eta reduction conditions]

Almost all the new lines are comments!

	
(No need to merge to 6.8, but no harm if a subsequent patch needs it.)

The proximate cause for this patch is to improve the inlining for INLINE
things that are not functions; this came up in the NDP project.  See
Note [Lone variables] in CoreUnfold.

This caused some refactoring that actually made things simpler.  In 
particular, more of the inlining logic has moved from SimplUtils to 
CoreUnfold, where it belongs.

using the new block-inheriting forkIO (#1048)

Seems better than getting a confusing 'cannot find directory' exception.

I finally got around to investigating why the Simplifier was sometimes
iterating so often.  There's a nice example in Text.ParserCombinators.ReadPrec,
which produced:

NOTE: Simplifier still going after 3 iterations; bailing out.  Size = 339
NOTE: Simplifier still going after 3 iterations; bailing out.  Size = 339
NOTE: Simplifier still going after 3 iterations; bailing out.  Size = 339

No progress is being made.  It turned out that an interaction between
eta-expansion, casts, and eta reduction was responsible. The change is
small and simple, in SimplUtils.mkLam: do not require the body to be
a Lam when floating the cast outwards.  

I also discovered a missing side condition in the same equation, so fixing
that is good too.  Now there is no loop when compiling ReadPrec.

Should do a full nofib run though.

This was a bit tricky.  We had a "given" dict like (d7:Eq a); then it got
supplied to reduceImplication, which did some zonking, and emerged with
a "needed given" (d7:Eq Int). That got everything confused.

I found a way to simplify matters significantly.  Now reduceContext
	- first deals with methods/literals/dictionaries
	- then deals with implications
Separating things in this way not only made the bug go away, but
eliminated the need for the recently-added "needed-givens" results returned
by checkLoop.  Hurrah.

It's still a swamp.  But it's a bit better.

This bug only reports a problem with phantom types, but actually
there was quite a long-standing and significant omission in the
constraint generation for derived classes.  See
Note [Superclasses of derived instance] in TcDeriv.

The test deriving-1935 tests both cases.

This is a bit unpleasant, as "make binary-dist" really shouldn't actually
build anything, but it works.

It wasn't doing the right thing for bindists. Let's rethink...

See comment for details

Ian Lynagh <igloo@earth.li>**20071124231857
 It used to be a synonym for ":r" in 6.6.1, but this wasn't documented or
 known about by the developers. In 6.8.1 it was accidentally broken.
 This patch brings it back, but as "repeat the last command", similar to
 pressing enter in gdb. This is almost as good for people who want it to
 reload, and means that it can also be used to repeat commands like :step.

Ian Lynagh <igloo@earth.li>**20071124171220

It's far too large now, and no-one complained when 6.8.1 didn't have one.

See trac #1910.

configure will set INSTALL to ./install-sh if it can't find it in the path,
so we need to replace the . with the path to our root.