GitLab

See Note [SPECIALISE pragmas] in TcBinds.  Fixes Trac #444.

This patch fixes two bugs in the Mach-O linker and adds debugging statements
to the same. The bugs:

1. The test for symbol->n_value == 0 is removed and replaced by a test of the
flag field.  Checking the n_value field was just wrong; the value of a
symbol should only be examined when allocating space for a common block,
in which case the n_value field gives the size of the block.  This bug
led to an infrequently occuring linker crash.

I believe the behavior of the linker now agrees with the intent of the
sketchy Apple documentation.

2. Jump islands were being filled with garbage instead of the the location
of the referenced symbol. This caused relocations of type X86_64_RELOC_GOT and
X86_64_RELOC_GOT_LOAD to eventually lead to crashes.  The fix is simply to
look up the symbol.

Enough debug statements have been added to follow the operation of the Mach-O
linker while it loads modules.  They are not yet as informative and well
organized as for ELF.  Improving the debug statements will require some
reorganization of the code -- the Mach-O linker seems basically sound, but
is crying out for some refactoring and commenting.
    

findDupsEq is quadratic, whereas removeDups is n log n.
Fixes T1969 regression.

We were failing to float out a binding that could be floated,
because of a confusion in the Lam case of floatExpr.

In investigating this I also discoverd that there is really
no point at all in giving a different level to variables in
a binding group, so I've now given them all the same (in 
SetLevels.lvlLamBndrs

The overall difference is quite minor in a nofib run:

        Program           Size    Allocs   Runtime   Elapsed
-------------------------------------------------------------
            Min          +0.0%     -8.5%    -28.4%    -28.7%
            Max          +0.0%     +0.7%     -0.7%     -1.1%
 Geometric Mean          +0.0%     -0.0%    -11.6%    -11.8%

I don't trust those runtimes, but smaller is good!  The 8.5% 
improvement in allocation in fulsom, and seems real.  The 
0.7% allocation increase only happens in programs with
very small allocation.  I tracked one down to a call of this form

  GHC.IO.Handle.Internals.mkDuplexHandle5
    = \ args -> GHC.IO.Handle.Internals.openTextEncoding1
                  mb_codec ha_type
                  (\mb_encoder mb_decoder -> blah)

With the new floater the argument of openTextEncoding1 becomes

     (let lvl = .. in \mb_encoder mb_decoder -> blah)

And rightly so.  However in fact this argument is a continuation
and hence is called once, so the floating is fruitless.

Roll on one-shot-function analysis (which I know how to do
but fail to get to!).

Previously, the demand analyser assumed that every binder 
starts off with no strictness info.  But now that we are
preserving strictness on nesting bindings in interface files,
that assumption is no longer correct, because an inlined function
might have a nested binding with strictness set.

So we need to know when we are in the initial sweep, so that we can
set the strictness to 'bottom'. 

See Note [Initialising strictness]

By not taking account of SCC notes we were inadvertently
discarding some, which led to mis-counted entries with -O

Should fix (part of) Trac #4414

At last. A bug that is really easy to fix.

As Roman reported in #4428, nested let-bindings weren't
being recorded with their unfoldings.  Needless to say,
fixing this had more knock-on effects than I expected.

"Short-cut" substitution means "do nothing if the substitution
is empty". We *never* want do to that in the simplifier because
even though the substitution is empty, the in-scope set has
useful information:

 * We get up-to-date unfoldings; and that in turn may
   reduce the number of iterations of the simplifier

 * We avoid space leaks, because failing to substitute may
   hang on to old Ids from a previous iteration

(This is what was causing the late inlining of foo in
Trac #4428.)

There is an accompanying patch for libraries
    base
    template-haskell

This has been wrong for a long time.  Fixes Trac #4431.

dist-install/build/ffi.h should have a dependency on ffitarget.h as *_stub.c requires it during the stage-2 build.

You can't use TH with a profiling compiler.

Major changes are:
* Version number now includes full GHC version and arch
* Uninstaller copes with multiple versions

Optimised the representation of Inert Sets to use Maps to get to the relevant inert constraint faster.

There are two main changes

 * New LANGUAGE option RebindableSyntax, which implies NoImplicitPrelude

 * if-the-else becomes rebindable, with function name "ifThenElse"
   (but case expressions are unaffected)

Thanks to Sam Anklesaria for doing most of the work here

This would have produced a more civilised error for Trac #4396

- add link to inno setup docs, so readers don't have to guess what
  ghc.iss might be

- add Task section, and associated most Registry actions with separate
  (sub)tasks, so that file associations, default action, and PATH
  setting can be optional

- copy license file into doc directory

- install icon file again, so that DefaultIcon is no longer a dangling
  pointer (#4352)

- only delete ghc_haskell key if empty (in case there were other tools
  using it)

- add versioned GHCi to right-click menu (to allow for multiple tool installs)

See Note [Localise pattern binders]