GitLab

gmp is now in a top-level directory and we only have the tarball in the
darcs repo. It gets untarred if it is needed.

Update functions in Compat.Directory from originals in System.Directory

A tricky bug to do with the way the implication constraints are
solved in TcSimplify.  See Note [Reducing implication constraints].

mapAccumL and mapAccumR are in Data.List now.
mapAccumB is unused.

Use standard isPrefixOf/isSuffixOf instead.

$(GHC) is always the installed compiler, $(HC) changes depending on
the stage.  This has the nice effect that we build stage2/ghc-inplace
using the stage1 compiler, and get to take advantage of bugfixes in
stage1 relative to $(GHC) (in particular I want the non-blocking stdin
fix for stage2, otherwise piping the testsuite output through tee
fails).

The SIGCONT handler was used to put stdin back in O_NONBLOCK mode,
after the process is resumed, but we don't want to do that now: the
standard handles are kept in blocking mode.

(This required a bit of refactoring of CmmInfo.)

This patch also fixes how the CPS handles functions with an empty body.
Factoring cmmToRawCmm out of the CPS pass made that fix easier.

These include:
 - Stack size detection now includes function arguments.
 - Stack size detection now avoids stack checks just because of
   the GC block.
 - A CmmCall followed by a CmmBranch will no longer generate an extra
   continuation consisting just of the brach.
 - Multiple CmmCall/CmmBranch pairs that all go to the same place
   will try to use the same continuation.  If they can't (because
   the return value signature is different), adaptor block are built.
 - Function entry statements are now in a separate block.
   (Fixed bug with branches to the entry block having unintended effects.)
 - Other changes that I can't recall right now.

This eliminates one of the panics introduced by 
the previous patch:
'First pass at implementing info tables for CPS'
  
The other panic introduced by that patch still remains.
It was due to the need to convert from a
  ContinuationInfo to a CmmInfo.
  (codeGen/CgInfoTbls.hs:emitClosureCodeAndInfoTable)
  (codeGen/CgInfoTbls.hs:emitReturnTarget)

This is a fairly complete implementation, however
two 'panic's have been placed in the critical path
where the implementation is still a bit lacking so
do not expect it to run quite yet.

One call to panic is because we still need to create
a GC block for procedures that don't have them yet.
(cmm/CmmCPS.hs:continuationToProc)

The other is due to the need to convert from a
ContinuationInfo to a CmmInfo.
(codeGen/CgInfoTbls.hs:emitClosureCodeAndInfoTable)
(codeGen/CgInfoTbls.hs:emitReturnTarget)