accordingly ASSIGN_INSTR into ASSIGN_STMT. The original situation was 
particularly confusing as BLOCK_INSTR did not point to anything of
node set Instr. 
I hope this is a small contribution to making the AST better digestible
to newcomers and godfathers alike.

The original name of the was once motivated by the coexistance
of a FUNDEF_ISWHILEFUN flag in the dark ages. For many years we
convert while-loops into do-loops before even thinking about
lac2fun transformation. The new name better reflects the special
character of these functions.

has two child nodes: Vardecs (instead of Vardec) and Assigns
(instead of Instr).

and
          GDBwhatAre( node *nm, node *fundef)
to print the PRF_ARGS' values of nm. 

E.g., given this:

 _awlfi_6426__awlfi_6407 = _noteintersect_( _flat_7, _pinl_3451__flat_19, 
_pinl_3482__wlbsc_3315_sc_bound, _ivexp_6288, _flat_2, 
_pinl_7226_p, _ivexp_6288, _ivexp_6289); 

call GDBwhatAre("_awlfi_6426__awlfi_6407", arg_info->fundef)
( edited for space)
_flat_7 = _add_SxS_( _ivexi_6270_ext, _wlpg_3285_elems); 
_pinl_3451__flat_19 = [ 0 ]; 
_pinl_3482__wlbsc_3315_sc_bound = [ 30 ]; 
_ivexp_6288 = [ 2 ]; 
_flat_2 = [ 20 ]; 
_pinl_7226_p = [ false ]; 
_ivexp_6288 = [ 2 ]; 
_ivexp_6289 = [ 21 ]; 

are now handled properly and the corresponding unit test is passed
successfully.

AL now recognises loop-invariant variables in loops and with-loops
and reorganises multi-operand expressions such that loop invariant
variables are grouped together to enable loop invariant removal.

to the corresponding recursive application N_ap node within the function
body. Adapted a number of modules to establish and maintain this
additional attribute as well as to check it. This new attribute will
later be used to facilitate implementation of a new feature in AL that
reorganises multi-operand expression to enable loop invariant removal
and with-loop invariant removal.

No commit message

This set of changes claim to make no changes
to operation, unless you use "-doivecyc" in the sac2c
command line. These changes are intended to 
provide a more general and elegant solution to 
Bug #747.

Other changes here include added tree_compound utility
functions, a new primitive for sacprelude's use: _mask_VxVxS_(),
various bugfixes for AWLF and friends.

There is a fair bit of development remaining before these
changes are of use in most applications, as enabling 
-doivecyc, at present, inhibits the operation of WLF,
AWLF, CWLE, SSAWLUR, SCCF, and perhaps a few others.

No commit message

No commit message

any single-line comment.

assign for an N_id arg_node.

functions that were erroneously recognized as expression-modifying
functions.

such that it is decoupled from the identification of local scopes.
This change was motivated by bug #855.

properly reflect its purpose. Furthermore, with-loop induction
variables are now treated correctly. While this has not been relevant
so far, the planned improvements to AL require this as a preparatory
step.

          sacprelude::intersectMin and sacprelude::intersectMax
  so they can be used by AWLF.

2. Fix several AWLF bugs.

3. Tidy up cubeslicer a wee bit.

This version of AWLF appears to more or less work.
It suffers from the AL/AS/DL bug and the VPid bug
(aka IVE vs. LIR in Bodo's PM code), so we await
fixes for those. However, if you have an AKD problem 
you want to try it out on, do this:

  sac2c -O3 -doawlf foo.sac

is still broken, especially when axis confluence is
required. This version is somewhat better, but
we're not there yet. 

on the unit tests. We're not all there yet, but 
this is good stuff.

that the computation has to be made on denormalized 
bounds, i.e., GENERATOR_BOUND2-1 or things get very
wrong.

making bison and Clemens happy.

The conflict is caused by the introduced support of return 
statement without brackets. For example:

return 5;

is a legal construction now. However, this construction causes
the following conflict:

return (5);

is it a function call with an argument '5', or it is a return
statement with an expression (5).

This theoretically (aside from at least one
memory-eating bug) completes the implementation of AWLF
for unit stride (STEP) and unit WIDTH. 

Several unit tests remain broken, notably:
prdreverse.sac
prdreverse2.sac
prdreverseAKD.sac
prdreverseAKDNoStdlib.sac

and others that are broken, pending repair
of Bug #851, Bug #855, and Bug #853.

   in AWLF. By holes, I mean index expressions of the form:
     _sel_VxA_( [2, i, j], producerWL) 
   appearing in consumerWLs.

2. Add comment to SCCF re mask() optimization.

This leaves AWLF mostly implemented. The missing 
pieces are:

  a. support for non-unit STEP and WIDTH in WLs.
     This is not very high on my list, but it should
     be straightforward to implement.

  b. Support for * and / in index vector computations in WLs.
     This is ISMOP.

  c. Support for axis confluence in consumerWL axis computations.
     E.g., an expression of the form:

     _sel_VxA_( [ iv, iv + 2], producerWL);

    in which cube slicing is required. This is the most
    egregious fault in AWLF code, at present, I believe.

statements.

Correct spelling error in diagnostic.

Sort of works, but it is not ready for prime time yet.

New mutc rc method
Fix bug in make spawn functions

This is a synchpoint for rbe. Don't take it too
seriously...

Omitting brackets in the return statement.

No commit message

AWLF looking fairly good, but three problems remain:

 - axis confluence of indices (e.g., ProducerlWL[ iv, iv + 1])
   not coded yet.

 - Holes in axis sets (e.g., ProducerWL[ 2, iv])
   not coded yet.

 - NEEDCOUNT for WLs is wrong when hypercube slicing (CUBSL) is
   used. Somewhere, I think, I have left hanging PART_CODE
   blocks that are contributing to needcounts, and making
   AWLF reject the fold. No idea where they are coming from.