GitLab

Trim imports

Further integration with the new package story.  GHC now supports
pretty much everything in the package proposal.

  - GHC now works in terms of PackageIds (<pkg>-<version>) rather than
    just package names.  You can still specify package names without
    versions on the command line, as long as the name is unambiguous.

  - GHC understands hidden/exposed modules in a package, and will refuse
    to import a hidden module.  Also, the hidden/eposed status of packages
    is taken into account.

  - I had to remove the old package syntax from ghc-pkg, backwards
    compatibility isn't really practical.

  - All the package.conf.in files have been rewritten in the new syntax,
    and contain a complete list of modules in the package.  I've set all
    the versions to 1.0 for now - please check your package(s) and fix the
    version number & other info appropriately.

  - New options:

	-hide-package P    sets the expose flag on package P to False
	-ignore-package P  unregisters P for this compilation

	For comparison, -package P sets the expose flag on package P
        to True, and also causes P to be linked in eagerly.

        -package-name is no longer officially supported.  Unofficially, it's
	a synonym for -ignore-package, which has more or less the same effect
	as -package-name used to.

	Note that a package may be hidden and yet still be linked into
	the program, by virtue of being a dependency of some other package.
	To completely remove a package from the compiler's internal database,
        use -ignore-package.

	The compiler will complain if any two packages in the
        transitive closure of exposed packages contain the same
        module.

	You *must* use -ignore-package P when compiling modules for
        package P, if package P (or an older version of P) is already
        registered.  The compiler will helpfully complain if you don't.
	The fptools build system does this.

   - Note: the Cabal library won't work yet.  It still thinks GHC uses
     the old package config syntax.

Internal changes/cleanups:

   - The ModuleName type has gone away.  Modules are now just (a
     newtype of) FastStrings, and don't contain any package information.
     All the package-related knowledge is in DynFlags, which is passed
     down to where it is needed.

   - DynFlags manipulation has been cleaned up somewhat: there are no
     global variables holding DynFlags any more, instead the DynFlags
     are passed around properly.

   - There are a few less global variables in GHC.  Lots more are
     scheduled for removal.

   - -i is now a dynamic flag, as are all the package-related flags (but
     using them in {-# OPTIONS #-} is Officially Not Recommended).

   - make -j now appears to work under fptools/libraries/.  Probably
     wouldn't take much to get it working for a whole build.

---------------------------------
	Fix a bug in usage recording
	---------------------------------

As a result of the new stuff on hi-boot-file consistency checking, I
accidentally caused Foo.hi to record a usage line for module Foo, and
this in turn caused rather nasty bad things to happen.  In particular,
there were occasional crashes of form

ghc-6.3: panic! (the `impossible' happened, GHC version 6.3.20041017):
        forkM Constructor Var.TcTyVar{d r1B9}

At least I think that's why the crash happened.

Anyway, it was certainly a bug, and this commit fixes it.  The main
payload of this fix is in Desugar.lhs;  the rest is comments and
tidying.

Wibbles to hi-boot files and newtypes

------------------------------------
	Simplify the treatment of newtypes
	Complete hi-boot file consistency checking
	------------------------------------

In the representation of types, newtypes used to have a special constructor
all to themselves, very like TyConApp, called NewTcApp.    The trouble is
that means we have to *know* when a newtype is a newtype, and in an hi-boot
context we may not -- the data type might be declared as
	data T
in the hi-boot file, but as
	newtype T = ...
in the source file.  In GHCi, which accumulates stuff from multiple compiles,
this makes a difference.

So I've nuked NewTcApp.  Newtypes are represented using TyConApps again. This
turned out to reduce the total amount of code, and simplify the Type data type,
which is all to the good.


This commit also fixes a few things in the hi-boot consistency checking
stuff.

------------------------------------
	Add Generalised Algebraic Data Types
	------------------------------------

This rather big commit adds support for GADTs.  For example,

    data Term a where
 	  Lit :: Int -> Term Int
	  App :: Term (a->b) -> Term a -> Term b
	  If  :: Term Bool -> Term a -> Term a
	  ..etc..

    eval :: Term a -> a
    eval (Lit i) = i
    eval (App a b) = eval a (eval b)
    eval (If p q r) | eval p    = eval q
    		    | otherwise = eval r


Lots and lots of of related changes throughout the compiler to make
this fit nicely.

One important change, only loosely related to GADTs, is that skolem
constants in the typechecker are genuinely immutable and constant, so
we often get better error messages from the type checker.  See
TcType.TcTyVarDetails.

There's a new module types/Unify.lhs, which has purely-functional
unification and matching for Type. This is used both in the typechecker
(for type refinement of GADTs) and in Core Lint (also for type refinement).

-------------------------------
	Use merge-sort not quicksort
	Nuke quicksort altogether
	-------------------------------

Quicksort has O(n**2) behaviour worst case, and this occasionally bites.
In particular, when compiling large files consisting only of static data,
we get loads of top-level delarations -- and that led to more than half the
total compile time being spent in the strongly connected component analysis
for the occurrence analyser.  Switching to merge sort completely solved the
problem.

I've nuked quicksort altogether to make sure this does not happen again.

-------------------------------
	Add instance information to :i
 	Get rid of the DeclPool
	-------------------------------


1.  Add instance information to :info command.  GHCi now prints out
    which instances a type or class belongs to, when you use :i

2.  Tidy up printing of unqualified names in user output.
    Previously Outputable.PrintUnqualified was
	type PrintUnqualified = Name -> Bool
    but it's now
	type PrintUnqualified = ModuleName -> OccName -> Bool
    This turns out to be tidier even for Names, and it's now also usable
    when printing IfaceSyn stuff in GHCi, eliminating a grevious hack.

3.  On the way to doing this, Simon M had the great idea that we could
    get rid of the DeclPool holding pen, which held declarations read from
    interface files but not yet type-checked.   We do this by eagerly
    populating the TypeEnv with thunks what, when poked, do the type
    checking.   This is just a logical continuation of lazy import
    mechanism we've now had for some while.

The InstPool and RulePool still exist, but I plan to get rid of them in
the same way.  The new scheme does mean that more rules get sucked in than
before, because previously the TypeEnv was used to mean "this thing was needed"
and hence to control which rules were sucked in.  But now the TypeEnv is
populated more eagerly => more rules get sucked in.  However this problem
will go away when I get rid of the Inst and Rule pools.

I should have kept these changes separate, but I didn't.  Change (1)
affects mainly
	TcRnDriver, HscMain, CompMan, InteractiveUI
whereas change (3) is more wide ranging.

-----------------------------------------------
       Record whether data constructors are declared infix
	-----------------------------------------------

This allows us to generate the InfixC form in Template Hasekll.
And for 'deriving' Read and Show, we now read and parse the infix
form iff the constructor was declared infix, rather than just if
it does not have the default fixity (as before).

IfaceSyn changes slightly, so that IfaceConDecl can record their
fixity, so there are trivial changes scattered about, and
you'll need to recompile everything.

In TysWiredIn I took the opportunity to simplify pcDataCon slightly,
by eliminating the unused Theta argument.

-----------------------------------------------
	Improve location info on unused-import warnings
	-----------------------------------------------

Improving the location involves plumbing the location of the import a bit
more assiduously -- hence change to imp_mods in TcRnTypes

------------------------
	More newtype clearing up
	------------------------

* Change the representation of TyCons so that it accurately reflects
	* data     (0 or more constrs)
	* newtype  (1 constr)
	* abstract (unknown)
  Replaces DataConDetails and AlgTyConFlavour with AlgTyConRhs

* Add IfaceSyn.IfaceConDecls, a kind of stripped-down analogue
  of AlgTyConRhs

* Move NewOrData from BasicTypes to HsDecl (it's now an HsSyn thing)

* Arrange that Type.newTypeRep and splitRecNewType_maybe unwrap just
  one layer of new-type-ness, leaving the caller to recurse.

  This still leaves typeRep and repType in Type.lhs; these functions
  are still vaguely disturbing and probably should get some attention.

Lots of knock-on changes.  Fixes bug in ds054.

Yet another fix to the -Onot optimisation that hides data type
representations in .hi files.

1.  Expose the representation if any fields are exposed

2.  Don't expose newtypes whose data-cons are abstract, unless the
    rep type is a FFI type.  (Previously we were more conservative
    and always exposed newtypes, just in case of a foreign decl.)

Always expose newtypes, at least for now (see comments)

Wibbles to exporting types abstractly

---------------------------------------
	  Don't expose constructors as vigorously
	  ---------------------------------------

GHC used to expose the constructors of a data type in the interface file,
even if (a) we were not optimising, and (b) the constructors are not exported.

In practice this isn't really necessary, and it's bad because it forces too
much recompilation.  I've been meaning to fix this for some while.

Now the data cons are hidden, even in the interface file, if both (a) and (b)
are true.  That means less interface file wobbling.

Mind you, the interface file still changes, because the to/from functions for
generic type classes change their types.  But provided you don't use them, you'll
get "compilation not required".

We could play the same game for classes (by hiding their class ops) but that'd
mean we'd have to change the data type for IfaceClassDecl, and I can't be
bothered to do that today.  It's unusual to have a class which exports none
of its methods anyway.



On the way, I changed the representation of tcg_exports and mg_exports (from
Avails to NameSet), but that should be externally invisible.

Remove unused parameter to groupAvails

Make 'ghc --show-iface' give a reasonable error message on old
interface files again.  We previously disabled the version check for
--show-iface so that you could run --show-iface on a profiled
interface file, but this disabled too much error checking.  Really we
just want to disable the 'way' check, not the whole version check.

HEADS UP: interface format changed.  Recompile libraries.

Print info about orphan rules and instances

Deal corectly with rules for Ids defined in this module,
even when they are imported (as orphans) from other modules.

The epicentre for this stuff is SimplCore.

Arrange that loadImportedRules can see the module dependencies
of this module, and hence know whether or not to load an hi-boot
interface.

Oops; forgot to add this entire directory!