* opt/abcopt.sml: New file.
* main/flintcomp.sml: Change `Compiler' to `FLINT' or `CPS' in the
name of phases printed by Compiler.Stats.
(abcopt): New phase.
* viscomp-core.cm: Add FLINT.opt/abcopt.sml.

* opt/switchoff.sml: Remove.
* reps/equal.sml: Rewritten to generate more CPS-like code.
* opt/fcontract.sml (fcLet): re-introduced the branch/switch code,
but without the deadly multiple traversals.
* */control.sml: Less flintopt and more cpsopt.

*** empty log message ***

The compiler now compiles to a fix point but "split" still
doesn't work because of a problem with the pickler.

somewhat functional

merged in 110.19 and 110.20.
Conflicts resolved, but it probably doesn't compile yet.

turned the CPS optimizer back on and tamed the FLINT inlining

First try at a merge of 110.16

* opt/split.sml (sexp): don't split HANDLE (it's incorrect).
(funeffect): embryo to detect side-effect free APPs (non-functional).
(splitThreshold): to put a cap on inlining.
(stfn): don't bother splitting inlinable TFNs.
* opt/fixfix.sml (curry): fixed bug when uncurrying cooked functions.
* opt/fcontract.sml (fcFun):  fix bug when undertaking mut-rec functions.
* main/flintcomp.sml:  added `recover' to help debugging.
updated the fold to allow extraction of Fi and return it at the end.
* lsplit/ls-inline.sml (oneBranch): fixed the wrapper function.
* flint/flintutil.sml (freevars):  forgot to count the arg of SWITCH.
* flint/flint.sig:  added a tfkind to TFN (only inlining for now).
* main/control.sml: new file. Moved from TopLevel/viscomp/control.sml
(splitThreshold): new var.
* TopLevel/viscomp/control.sig (FLINT.printFctTypes): to reduce clutter.
(splitThreshold): to control splitting agressiveness.
* TopLevel/viscomp/control.sml:  moved substructs outside so that clients
  can refer to them directly (rather than through Control.Foo) to reduce
  spurious dependencies.
* TopLevel/main/{codes,compile}: call `split' from flintcomp, not compile.
* kernel/ltyextern.sml (tnarrow), reps/{reify,rttype,typeoper}.sml:
  flatten arguments when reifying them since the pretty-printer doesn't
  know how to deal with flattened reified TFNs.

* opt/split.sml: seriously reworked.  Now splits TFNs as well.
* kernel/primop.sml (effect): removed DEREF and MAKEREF from the pure
  primops.
* flint/flintutil.sml (copy): sort the type-map since {lt,tc}_subst
  actually expects it to be sorted.
* kernel/ltyextern.{sml,sig} (lt_nvpoly): new function (digged from
  recover.sml to build a LT_POLY in the case of named tvars.
* opt/recover.sml: removed unused depth arguments left behind from before
  we switched to named tvars.
(addLty): added for the benefit of opt/split.sml
(lt_nvpoly): moved to ltyextern.sml

merged in 110.10

Added IH_MAYBE to allow more selective inlining.

*** empty log message ***

Merged in Bratin's typelifting and Chris' named tvars.
FLINT is now using tvars from the first lcontract all the way to cps.

* reps/rttype.sml: changes from Bratin.
(rtLexp): define the NVAR case (return the tvar as an lvar).
* opt/[fl]contract.sml: removed the restriction on inlining at the same
  deBruijn depth.
* opt/lift.sml: new file for typelifting.
* flint/flintutil.sml (copy): added tvar renaming.
* flint/tvarcvt.sml: new file for named tvar <-> deBruijn conversions.

* opt/loopify.sml: new file
* cps/convert.sml: added code to take advantage of LK_WHILE loops.

*** empty log message ***

* opt/collect.sml: switched back to a read-only collect phase.  Also try
  to export the info to allow fcontract to circumvent some of the get calls.

* opt/fcontract.sml: take advantage of single-arm SWITCH (aka decon)
(SWITCH(CON)): just compare the conreps because types may differ (?)
(inlineWitness): ugly hack used to figure out if inlining happened,
  in order to decide whether to keep the inline bit or not (in cfun).

* added basic unrolling support
* changed fkind to have most annotations valid on
  functors as well.  It also adds a loopkind annotation as well as extends
  the inline boolean into a three-way alternative.
* switched to a continuation passing style to implement the
  let-associativity rule in a better way.

* opt/fcontract.sml (dropcstargs): replace constant args by the constant.

* opt/fixfix.sml (curry): correctly handle "imposible" cases.

* opt/collect.sml: completely revisited:  no distinction between
  internal/external uses, dead-code elimination done right away.  This
  makes fcontract a little simpler.

added dropping of dead-arguments

*** empty log message ***

*** empty log message ***

addition of fcontract and fixfix and Control.FLINT

merged in 110.7 changes

merged in 110.6 changes

merged in 110.5.1 from Zhong

merged with 110.5

*** empty log message ***

merged in Zhong's 110.4

dummy

Renaming of Control.CG flags related to type-checking.  Now we have
`checkFlint', `checkKinds', and `checkDatatypes'.

split printLambda into printLambda and printFlint

This commit was generated by cvs2svn to compensate for changes in r45,
which included commits to RCS files with non-trunk default branches.

*** empty log message ***

This commit was manufactured by cvs2svn to create branch 'SMLNJ'.

Initial revision