Added missing </Para>. Ironic if you look at the last commit on this file.

Don't read TMPDIR from the environment under mingwin; avoids / being
turned into \ in the env var, which then causes problems when this value
is written to a shell file and then executed.

driver<->hsc deforestation continues; classifyOpts isn't needed any more

Spadework for getting ModuleName -> Module maps in the right places.
Also start to expand short names (PCS) to PersistentCompilerState, etc.

Back out changes to TcMonoType that didn't work right.

These changes are now done correctly on the before-ghci-branch,
and so will get merged in later.

Define HscLang, the output format of hsc.

Remove opt_OutputLanguage, opt_OutputFile, and a few other flags which
we're going to do through DynFlags or directly as arguments to hscLang now.

The driver, more or less verbatim copied from ghc/driver/Main.hs.

THE PLAN: the main body of the driver will eventually sit between the
compilation manager and HscMain, providing all the grotty services
like compilation of .hc files.

Some other parts of the driver, eg. flag parsing, will remain at the
front end.

Temporary file management, broken out of the driver.

Main is now HscMain: this will transform eventually into the
compiler's entry point.

CmdLineOpts now separates flags into static flags and dynamic flags;
dynamic flags will be passed around explicitly and can therefore
change from compilation to compilation.  All the dump flags and a few
language flags (like -fglasgow-exts) are currently DynFlags.

NOTE: nothing will now compile, for a long while until we get all this
connected up.  This commit is so that we start pulling in the same
direction.

export getImports; mkdependHS will need it.

Build main/Config.hs (cut-n-paste from driver/Makefile)

* Get more of the compilation manager working, using a dummy compiler
  and linker.
* When linking, figure out the set of packages which need to be linked.
* Redo package config plumbing to support the above.  Some stuff has
  moved from CmFind to CmStaticInfo.
* Get rid of FLAGS entirely.

Loop in fill_up_line_buffer if read returns EINTR.  The code
previously just returned in this case.

debugging output wibbles

Just for fun, perl v5.6.0 version info has a different syntax.

Added a tip about using <Para> inside <ListItem>s.

Many fixes to make it DocBook friendly.

Fill in some gaps; add Float# and Double# reps

fix heap check bogons in mci_make_constr*

Lots of changes for Xmlambda. all changes are between
#ifdef XMLAMBDA blocks.

- new bytecodes for Inj constructors and witnesses
- new primops for rows
- code for calling foreign functions. This only works with
'dynamic.c' but that is not checked in yet.

import Linker ( linkPrelude )

removed references to 'splim' in code for the INTERPRETER.

Re-add -fwarn-hi-shadowing now that we mostly remove duplicates from
the import path.

Link the compiled prelude in, for testing the interpreter.

Hack about in a major way, and get this thing linking interpreted code
to a compiled prelude.

- Add Linker.lhs, an interface to the underlying RTS object linker
- Split off the interpreter's abstract syntax into InterpSyn.lhs
- Some minor updates to the compilation manager stuff.

The GC needs to be cleverer about which address ranges are in
text/data space if we're dynamically loading modules.

(BTW, this linking stuff is mostly ported from code written by Julian
Seward for STG Hugs).

Initialise the object linker if we're in GHCi.

Add a debug flag for the object linker.

Extend the hash table implementation to support string-indexed dynamic
hash tables.

Object file linker for GHCi.

Implement a few more dull bits of code for the Compilation Manager.

Commit half-implemented CM, and supporting changes.

Mucho fixes for Windows MM timers.

Allow finite number of (!) fields in constructors :-)

Adding depend info

Fixing the order that labeled fields appear in datatypes. Currently
they are reversed, which breaks the prelude badly.

[From Haskell report:

The arguments to the positional constructor occur in the same order as the labeled fields. For example, the declaration

  data C = F { f1,f2 :: Int, f3 :: Bool}

defines a type and constructor identical to the one produced by

  data C = F Int Int Bool

]

wibble

Add test for generics

Add a test to reject things like:

	instance Ord a => Ord (forall s. T s a)

	g :: T s (forall b.b)

The for-alls are illegal in type arguments!