GitLab

Remove CrossStage from Convert

Fix an obscure record-selector-in-newtype bug

Added support for negation to THSyntax and DsMeta.repE.

Oops... add missing free-vars

err msg wibble

Add a note about where the time spent in foreign code is attributed.

Cleaned up `repE'.  Reordered to match order of cases in HsExpr and made an
effort to catch all cases.

Wibble to lint-ing unfoldings

Added a case for HsPar to repE.  Also completed the cases in repE to at
least panic for the missing syntactic forms of HsExpr.

Is there any good reason for the semi-random order of the cases in repE?
Using the same order as in the data declaration for HsExpr would make it
easier to spot missing cases.

Slight tidy up

Make imports work for pre-504

Remove unnecessary compiler and linker flags for Mingw32.

Track locations for binders better (fixes several rn and mod failures)

Use getExecDir, just like in ghc-pkg and SysTools

Fix bogon in linker

Fix a type and a bug for PowerPC.

For Mac OS X, use the underlying Mach Microkernel calls instead of mmap.
Darwin's mmap doesn't honor the passed-in address without MAP_FIXED,
and MAP_FIXED replaces all existing mappings, so it can't be used.

Remove Win32 library dependency.

Put findLinkable in Finder.lhs, where it can be found in non-ghci stage1 compilers!

Import wibbles

Import wibbles

Reduce exports, and add comments

------------------------
	More dependency fiddling
	------------------------

	WARNING: Interface file format has changed (again)
		 You need to 'make clean' in all library code

* Orphan modules are now kept separately
  Home-package dependencies now contain only home-package dependencies!
  See HscTypes.Dependencies

* Linker now uses the dependencies to do dynamic linking

Result: Template Haskell should work even without --make (not yet tested)

We shouldn't be using MAP_FIXED on solaris (fixes last night's crash).

Unfortunately on Solaris the heuristics don't work too well.  Solaris
seems to ignore the addr argument to mmap when MAP_FIXED is not
specified, and starts handing out memory from the top of the address
space.  So we allocate 2M each time and munmap() ends to leave an
aligned 1M chunk.  This will eventually leave the address space
completely full of 1M holes.  I don't have a good solution to this at
the moment.

MERGE TO STABLE

* typo (dl -> dnl)

Wibble; cures failure in stage2 build

In eval_thunk_selector(), don't follow IND_STATICs because they might
lead us into to-space.  Fixes a case of "EVACUATED object entered!".

Also, add an assertion to catch this bug earlier.

MERGE TO STABLE

Uh oh; got the versioning stuff a bit wrong in the last commit

------------------------------------------
	1. New try and module and package dependencies
	2. OrigNameCache always contains final info
	------------------------------------------

These things nearly complete sorting out the incremental
linking problem that started us off!

1. This commit separates two kinds of information:

  (a) HscTypes.Dependencies:
	What (i)  home-package modules, and
	     (ii) other packages
      this module depends on, transitively.

      That is, to link the module, it should be enough
      to link the dependent modules and packages (plus
      any C stubs etc).

      Along with this info we record whether the dependent module
      is (a) a boot interface or (b) an orphan module.  So in
      fact (i) can contain non-home-package modules, namely the
      orphan ones in other packages (sigh).

  (b) HscTypes.Usage:
      What version of imported things were used to
      actually compile the module.  This info is used for
      recompilation control only.



2. The Finder now returns a correct Module (incl package indicator)
first time, so we can install the absolutely final Name in the
OrigNameCache when we first come across an occurrence of that name,
even if it's only an occurrence in an unfolding in some other interface
file.  This is much tidier.

As a result Module.lhs is much cleaner
	No DunnoYet
	No mkVanillaModule
ALl very joyful stuff.

In --make mode, consider everything that doesn't contain a '.' to be a
compilation manager input.  This fixes cases like

  ghc --make test37

where test37.hs exists.  I broke this in the previous commit, causing
a number of tests to break.

Document implicit parameter bindings

------------------------------------------------
	Allow implicit-parameter bindings anywhere that
		a normal binding group is allowed.
	------------------------------------------------

That is, you can have implicit parameters

	* in a let binding
	* in a where clause (but then you can't have non-implicit
	  ones as well)
	* in a let group in a list comprehension or monad do-notation

The implementation is simple: just add IPBinds to the allowable forms of HsBinds,
and remove the HsWith expression form altogether.   (It now comes in via the
HsLet form.)

It'a a nice generalisation really.  Needs a bit of documentation, which I'll do next.

Declare loop index i.

Add atomicModifyIORefzh_fast symbol (should have been done with the
rest of the atomicModifyIORef# commit, thanks to Manuel Chakravarty
for pointing out the problem).

Added an import to make TH compile.

Fix recent FreeBSD breakage in the mangler: the "Prologue Junk" test
was a little bit too strict about whitespace.

change the types of cmp_thread, rts_getThreadId, and labelThread to
take StgPtr rather than StgTSO *, since the compiler now has no
distinction between these two types in the back end.

I also noticed that labelThread need not be a primitive: it could just
as well be a normal C function called by the FFI, but I haven't made
that change.

Import wibbles

Turn an instance of 0xfff00000 into ~MBLOCK_MASK in the Win32 code.

Bite the bullet and generalise the central memory allocation scheme.
Previously we tried to allocate memory starting from a fixed address,
which was set for each architecture (0x5000000 was a common one), and
to decide whether a particular address was in the heap or not we would
do a simple comparison against this address.

This doesn't work too well, because:

 - if we dynamically-load some objects above the boundary, the
   heap-allocated test becomes invalid

 - on windows we have less control, and the heap might be
   split into multiple sections

 - it turns out that on some Linux kernels we don't get memory where
   we asked for it.  This might be a bug in those kernels, but it
   exposes the fragility of our allocation scheme.

The solution is to bite the bullet and maintain a table mapping
addresses to a value indicating whether that address is in the heap or
not.  Since we normally allocate heap in chunks of 1Mb, the table is
quite small: 4k on a 32-bit machine, using one byte for each 1Mb
block.  Testing an address for heap residency now involves a memory
access, but the table is normally cache-resident.  I didn't manage to
measure any slowdown after making the change.

On a 64-bit machine, we'll need to use a 2-level table; I haven't
implemented that yet.

Now we can generalise the procedure used to grab memory from the OS.
In the general case, we allocate one megablock more than we need to,
and trim off the slop around the allocation to leave an aligned chunk.
The next time around, however, we try to allocate memory right after
the last chunk allocated, on the grounds that it is aligned and
probably free: if this doesn't work, we have to back off to the
general mechanism (it seems to work most of the time).

This cleans up the Windows story too: is_heap_alloced() has gone, and
we should be able to handle more than 256M of memory (or whatever the
arbitrary limit was before).

MERGE TO STABLE (after lots of testing)