TIP#237 IMPLEMENTATION

        [kennykb-numerics-branch] Resynchronized with the HEAD; at this
         checkpoint [-rkennykb-numerics-branch-20051008], the HEAD and
        kennykb-numerics-branch contain identical code.

1 files changed, 583 insertions, 3 deletions

diff --git a/ChangeLog b/ChangeLog
index 659563e..77e4faa 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,9 +1,54 @@
+2005-10-08  Don Porter  <dgp@users.sourceforge.net>
+
+        TIP#237 IMPLEMENTATION
+
+	[kennykb-numerics-branch] Resynchronized with the HEAD; at this
+	 checkpoint [-rkennykb-numerics-branch-20051008], the HEAD and
+	kennykb-numerics-branch contain identical code.
+
+	[kennykb-numerics-branch]	Merge updates from HEAD
+
+	* generic/tclExecute.c:	More performance macros and special
+	handling of the wide integer type for performance on 32-bit
+	systems.
+
+2005-10-07  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Macro GetNumberFromObj() is version of
+	TclGetNumberFromObj() that saves a function call for common uses.
+
+	* generic/tclInt.h:	Made #undef NO_WIDE_TYPE the default on
+	32-bit systems.  Being able to use 64-bit values without leaping
+	to mp_int should help with performance.
+	* generic/tclObj.c:	Bug fixes in the #undef NO_WIDE_TYPE
+	* generic/tclExecute.c:	configuration.
+
+	* generic/tclExecute.c:	Improved performance of comparison opcodes
+	and bitwise operations and removed yet more dead code.
+
 2005-10-07  Jeff Hobbs  <jeffh@ActiveState.com>
 
 	* unix/tclUnixFCmd.c (TraverseUnixTree): Adjust 2004-11-11 change to
 	* tests/fCmd.test (fCmd-20.2):           account for NFS special
 	files with a readdir rewind threshold. [Bug 1034337]
 
+2005-10-06  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Improved performance of INST_RSHIFT and
+	INST_LSHIFT.
+
+2005-10-05  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Improved performance of INST_MULT, INST_DIV,
+	INST_ADD, and INST_SUB and replaced a "goto... label" with a
+	"break from loop" in TclIncrObj() and removed some dead code.
+
 2005-10-05  Andreas Kupries <andreask@activestate.com>
 
 	* generic/tclPipe.c (TclCreatePipeline): Fixed [SF Tcl Bug
@@ -28,6 +73,18 @@
 	Correct the resizing of the environ array.  We assume that we are
 	in full ownership, but that's not correct.[Bug 979640]
 
+2005-10-04  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Updated TclIncrObj() to more efficiently
+	add native long integers.  Also updated IllegalExprOperandType
+	and the INST_UMINUS, INST_UPLUS, INST_BITNOT, and
+	INST_TRY_CVT_TO_NUMERIC sections for performance.
+
+	* generic/tclBasic.c:	Updated more callers to make use of
+	TclGetNumberFromObj.  Removed some dead code.
+
 2005-10-04  Jeff Hobbs  <jeffh@ActiveState.com>
 
 	* win/tclWinSerial.c (SerialSetOptionProc): free argv [Bug 1067708]
@@ -56,6 +113,17 @@
 	* generic/tclTimer.c (Tcl_DeleteTimerHandler): bail out early
 	if passed NULL argument.
 
+2005-10-03  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclBasic.c:	Re-implemented ExprRoundFunc and ExprEntierFunc
+	to use TclGetNumberFromObj.
+
+	* generic/tclInt.h:	Added new routine TclGetNumberFromObj to
+	* generic/tclObj.c:	provide efficient access to the actual
+	internal rep of a numeric Tcl_Obj without conversions.
+
 2005-10-03  Kevin Kenny  <kennykb@acm.org>
 
 	* tools/loadICU.tcl:  Changed the file names of message catalogs
@@ -71,12 +139,29 @@
 	* generic/tclMain.c: Separate encoding conversion of command line
 	arguments from list formatting.	 [Bug 1306162].
 
+2005-09-30  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclStringObj.c:	Bug fix: Missing cast to large enough
+	integral size before << operations led to broken [format %llx] results.
+	Thanks to Robert Henry for reporting the bug.
+
 2005-09-29  Jeff Hobbs  <jeffh@ActiveState.com>
 
 	* doc/mathfunc.n:   implementation for TIP #255, expr min/max
 	* library/init.tcl:
 	* tests/info.test, tests/expr-old.test: 
 
+2005-09-27  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tcl.h:	Changed name of the new Tcl_Obj intrep field
+	* generic/tclObj.c:	from "bignumValue" to "ptrAndLongRep" as
+	* generic/tclProc.c:	described in TIP 237, and more suitable for
+	other more general uses.
+
 2005-09-27  Donal K. Fellows  <donal.k.fellows@manchester.ac.uk>
 
 	* tests/binary.test (binary-14.18): Added test for [Bug 1116542]
@@ -84,11 +169,67 @@
 
 2005-09-26  Kevin Kenny  <kennykb@acm.org>
 
+	[kennykb-numerics-branch] Merge updates from HEAD.
+	
+2005-09-26  Kevin Kenny  <kennykb@acm.org>
+
 	* libtommath/:                   Updated to release 0.36.
 	* generic/tommath.h:             Regenerated.
 	* generic/tclTomMathInterface.h: Added ten missing aliases for mp_*
 	functions to avoid namespace pollution in Tcl's exported symbols.
 	[Bug 1263012]
+	
+2005-09-23  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* unix/Makefile.in:	Added -DMP_PREC=4 switch to all compiles so
+	* win/Makefile.in:	that minimum memory requirements of mp_int's
+	* win/makefile.vc:	will not be quite so large.  [Bug 1299153].
+
+	* generic/tclStrToD.c:	Fixed memory leak.  [Bug 1299803].
+	* generic/tclObj.c:	
+
+2005-09-20  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+	
+	* generic/tclExecute.c:	Revise TclIncrObj() to call
+	Tcl_GetBignumAndClearObj.
+
+	* generic/tcl.decls:	Add Tcl_GetBignumAndClearObj.
+	* generic/tclObj.c:
+
+	* generic/tclDecls.h:	make genstubs
+	* generic/tclStubInit.c:
+
+2005-09-16  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclInt.h:		Added TclBNInitBigNumFromWideInt()
+	* generic/tclTomMathInterface.c:	so that every caller isn't
+	required to duplicate the sign logic to use the unsigned interface.
+
+	* generic/tclBasic.c:	Reduce the number of places where Tcl
+	* generic/tclExecute.c:	intrudes into the internal format details
+	* generic/tclObj.c:	of the mp_int struct.
+	* generic/tclStrToD.c:
+	* generic/tcLStringObj.c:
+
+	* generic/tclTomMath.h:	Added mp_cmp_d to routines from
+	* unix/Makefile.in:	libtommath used by Tcl.
+	* win/Makefile.in:
+	* win/makefile.vc:
+
+	* libtommath/bn_mp_add_d.c:	Bug fix.  For mp_add_d(&a, d, &c),
+	when &a has the value -d, then the value &c computed should be zero,
+	but mp_add_d was producing an inconsistent zero value with a sign
+	field of MP_NEG, something like a value of -0, which other routines
+	in libtommath can't handle.
+
+	* generic/tclExecute.c:	Dropped all creation of "bigOne" values
+	and just use tommath routines that accept the value "1" directly.
 
 2005-09-15  Miguel Sofer <msofer@users.sf.net>
 
@@ -96,6 +237,8 @@
 
 2005-09-15  Don Porter  <dgp@users.sourceforge.net>
 
+	[kennykb-numerics-branch]	Merge updates from HEAD.
+
 	* generic/tclStringObj.c (TclAppendFormattedObjs):	Revision
 	to eliminate one round of string copying.
 
@@ -165,6 +308,11 @@
 
 2005-09-12  Don Porter  <dgp@users.sourceforge.net>
 
+	[kennykb-numerics-branch]	Merge updates from HEAD.
+
+	* generic/tclCmdAH.c:		Added support for the "ll" width	
+	* generic/tclStringObj.c:	specifier to [format].
+
 	* generic/tclStringObj.c (TclAppendFormattedObjs):	Bug fix:
 	make sure %ld formats force the collection of a wide value, when
 	the value could be a different long.
@@ -176,6 +324,10 @@
 	  1286256]. The other warning can be removed only by removing the
 	  panic/return code.
 
+2005-09-09  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]	Merge updates from HEAD.
+
 2005-09-09  Kevin Kenny  <kennykb@acm.org>
 
 	* generic/tclStringObj.c:  Added two missing casts to silence
@@ -228,6 +380,17 @@
 	* generic/tclObj.c (pendingObjDataKey): Added missing 'static' to stop
 	symbol from leaking outside the Tcl library. [Bug 1263012]
 
+2005-09-02  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclScan.c:	Bug fix: The %o, %x, %i formats of [scan]
+	must not accept any 0b or 0o prefixes.  [scan $s %o] must continue
+	to work even with KILL_OCTAL enabled.
+
+	* generic/tclInt.h:	Added TCL_PARSE_SCAN_PREFIXES to the flags
+	* generic/tclStrToD.c:	accepted by TclParseNumber.
+
 2005-09-01  Andreas Kupries <andreask@activestate.com>
 
 	* unix/tclUnixSock.c (InitializeHostName): Synchronized use of
@@ -252,6 +415,22 @@
 	* generic/tclIORChan.c (RcNewHandle): Synchronized use of static
 	  modifier in declaration and definition of function.
 
+2005-09-01  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclObj.c:	TclParseNumber calls meant to parse an
+	integer value now pass the TCL_PARSE_INTEGER_ONLY flag.
+
+	* generic/tclScan.c:	Extended [scan] to accept the %lld,
+	%llo, %llx, and %lli formats.  Numeric scanning is now done
+	via TclParseNumber calls.
+
+	* generic/tclInt.h:	Extended TclParseNumber to accept new flag
+	* generic/tclStrToD.c:	values TCL_PARSE_INTEGER_ONLY,
+	TCL_PARSE_OCTAL_ONLY, and TCL_PARSE_HEXIDECIMAL_ONLY, to give caller
+	more control over the parsing rules.
+
 2005-08-31  Vince Darley <vincentdarley@sourceforge.net>
 
 	* doc/FileSystem.3:
@@ -261,12 +440,35 @@
 	allowed.  Tcl's core itself (tclEncoding.c:FillEncodingFileMap())
 	calls this with a NULL interpreter.
 
+2005-08-30  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclObj.c:	Extended bignum support to include bignums
+	so large they will not pack into a Tcl_Obj.  When they outgrow Tcl's
+	string rep length limits, a panic will result.
+
+	* generic/tclTomMath.h:	Added mp_sqrt to routines from
+	* unix/Makefile.in:	libtommath used by Tcl.
+	* win/Makefile.in:
+	* win/makefile.vc:
+
+	* generic/tclBasic.c:	Extended sqrt(.) so that range covers
+	the entire double range, accepting as many bignums in the domain
+	as that will allow.
+
 2005-08-29  Andreas Kupries <andreask@activestate.com>
 
 	* library/tm.tcl (::tcl::tm::roots): Accepted Don Porter's patch
 	  for [Tcl SF Bug 1189657]. Syncs the implementation to the
 	  specification (TIP #189).
 
+2005-08-29  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]	Merge updates from HEAD.
+
+	* generic/tclBasic.c:	Restored round(.) to the Tcl 8.4 rules.
+
 2005-08-29  Kevin Kenny  <kennykb@acm.org>
 
 	* generic/tclBasic.c (ExprMathFunc): Restored "round away from
@@ -302,6 +504,19 @@
 	duplicated object on the floor, which was a memory leak (and a wrong
 	result too). Thanks to Andreas Kupries for reporting this.
 
+2005-08-25  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch] Merge updates from HEAD
+
+	* generic/tclExecute.c: Bug fix.  INST_RSHIFT: shift of negative
+	values produced incorrect results.
+
+	* generic/tclExecute.c:	Bug fix.  INST_*SHIFT opcodes stack
+	management. [expr 0<<6] should be 0, not 6.
+
+	* generic/tclBasic.c:	Extended the domain of round(.) to all
+	non-Inf, non-NaN doubles, using bignums for the result as needed.
+
 2005-08-24  Andreas Kupries <andreask@activestate.com>
 
 	TIP#219 IMPLEMENTATION
@@ -337,6 +552,37 @@
 	point numbers with the largest and smallest possible significands,
 	and added test cases for them.
 
+2005-08-24  Kevin Kenny  <kennykb@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c: Corrected some TRACE bugs that prevented
+	compilation with --enable-symbols=all.
+	* generic/tclStrToD.c: Revised commentary to prepare for a
+	renaming of the file, removed some dead code, and fixed a bug
+	where TclBignumToDouble failed on huge negative numbers.
+	* tests/binary.test (binary-65.*): Added missing 'ieeeFloatingPoint' to
+	large/small significand tests.
+	* tests/expr.test (expr-45.*) Added missing braces around expressions.
+	
+2005-08-24  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclBasic.c:	Revised implementation of the ceil(.) and
+	* generic/tclInt.h:	floor(.) math functions in light of the
+	* generic/tclStrToD.c:	revised comparison operators, so that it
+	is always true that ($x <= ceil($x)) and ($x >= floor($x)).  The	
+	simple approach of "convert to double and call ceil() or floor()"
+	could not guarantee that.
+
+	* generic/tclExecute.c:	Bug fix: TclBignumToDouble return -Inf when
+	appropriate.  Removed declarations of removed routines.
+
+	* generic/tclExecute.c:	Revised the type promotion rules of the
+	comparison operators so that they form proper equivalence classes
+	over the set of numeric strings.
+
 2005-08-23  Mo DeJong  <mdejong@users.sourceforge.net>
 
 	* unix/configure.in:
@@ -344,11 +590,165 @@
 	* win/configure.in: Update minimum autoconf version
 	to 2.59.
 
+2005-08-23  Kevin Kenny  <kennykb@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+	
+	* generic/tclCmdMZ.c (Tcl_StringObjCmd):
+	* generic/tclInt.h:
+	* generic/tclObj.c (Tcl_GetBooleanFromObj, SetDoubleFromAny,
+	Tcl_GetLongFromObj, Tcl_GetWideIntFromObj, Tcl_GetBignumFromObj):
+	* generic/tclParseExpr.c (GetLexeme):
+	* generic/tclScan.c (Tcl_ScanObjCmd):
+	* generic/tclStrToD.c (TclParseNumber):
+	* tests/binary.test (binary-62.1-65.7):
+	* tests/expr.test (expr-40.1-42.1):
+	* scan.test (scan-14.1,14.2):
+	Modified Tcl_ParseNumber to accept an argument to force
+	interpretation as decimal, and modified [scan] to use it.
+	Corrected a bug where Not a Number with hexadecimal information
+	bits returned consistently incorrect values. #ifdef-ed out
+	some code that is needed only for IBM hexadecimal floating point.
+	Fixed bugs in code to handle the corner cases of smallest and
+	largest significands.  Added test cases to improve test coverage
+	in generic/tclStrToD.c.	Added test cases for 0b notation (TIP
+	#114).  Removed TclStrToD, and the static functions that it calls,
+	which are now dead code (TclParseNumber	now does all input 
+	floating-point conversions.)
+	
+2005-08-23  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclStrToD.c:	Bug fix: set shift magnitude properly whether
+	we're expanding to mp_int type or not.
+
+	* generic/tclExecute.c:	Bug fix: ACCEPT_NAN under INST_UMINUS.
+
+	* generic/tclStrToD.c:	New macros TIP_114_FORMATS and KILL_OCTAL to
+	configure acceptance of 0o and 0b numbers and rejection of "leading
+	zero as octal".
+
+	* generic/tclBasic.c:	Re-used the guts of int(.) and wide(.) math
+	functions to perform conversions in OldMathFuncProc.
+
+	* generic/tclBasic.c:	Support for ACCEPT_NAN.
+	* generic/tclExecute.c:
+
+	* generic/tclInt.decls:	Restored TclExprFloatError to internal stubs
+	* generic/tclBasic.c:	table, and moved definition back to tclExecute.c
+	* generic/tclExecute.c:	from tclBasic.c to handle #undef ACCEPT_NAN.
+
+	* generic/tclIntDecls.h:	make genstubs
+	* generic/tclStubInit.c:
+
+	* generic/tclInt.h:	New internal macros TclIsNaN and TclIsInfinite
+	* generic/tclBasic.c:	replace the IS_NAN and IS_INF macros scattered
+	* generic/tclExecute.c:	here and there.
+	* generic/tclObj.c:
+	* generic/tclStrToD.c:
+	* generic/tclUtil.c:
+
 2005-08-22  Daniel Steffen  <das@users.sourceforge.net>
 
 	* unix/tclConfig.h.in: autoheader-2.59.
 
-cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
+2005-08-22  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclInt.h:	New ACCEPT_NAN macro to mark code that supports
+	* generic/tclCmdAH.c:	or disables accepting of the NaN value at 
+	* generic/tclExecute.c:	various points.
+	* generic/tclLink.c:
+
+	* generic/tclStrToD.c:	Bug fix.  Parsing of +/- Infinity was reversed.
+
+	* generic/tclTestObj.c:	Disabled unused [testconvertobj] command.
+
+	* generic/tclBasic:	Added [expr {entier(.)}].  Rewrote int(.)
+	and wide(.) to use the same guts, accepting all non-Inf doubles as
+	arguments.
+
+	* generic/tclInt.h:	New routine TclInitBignumFromDouble.
+	* generic/tclStrToD.c:	Modified to return code and write error message.
+
+	* generic/tclInt.h:	TCL_WIDE_INT_IS_LONG implies NO_WIDE_TYPE.
+	* generic/tclObj.c:	Removed now unnecessary tests of the
+	* generic/tclStrToD.c:	TCL_WIDE_INT_IS_LONG definition.
+
+	* generic/tclInt.h:	New internal routine TclSetBignumIntRep
+	* generic/tclObj.c:	consolidates packing of bignum value into
+	* generic/tclStrToD.c:	a Tcl_Obj within one source code file.
+
+	* tests/expr.test:	Corrected the wideIs64bit constraint.
+	* tests/format.test:
+	* tests/scan.test:
+
+2005-08-21  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclInt.h:		Moved TclParseInteger to tclUtil.c
+	* generic/tclParseExpr.c:	and made it static.
+	* generic/tclUtil.c:
+
+	* generic/tclInt.decls:	Moved TclExprFloatError to tclBasic.c and
+	* generic/tclBasic.c:	made it static.
+	* generic/tclExecute.c:
+
+	* generitc/tclIntDecls.h:	make genstubs
+	* generic/tclStubInit.c:
+
+	* generic/tclExecute.c:	errno, IS_NAN, IS_INF, LLD no longer called in
+	this file; dropped/disabled support for them.
+
+	* generic/tclCompExpr.c:	errno no longer used in these files;
+	* generic/tclParseExpr.c:	dropped support "hack" for it.
+
+	* generic/tclStrToD.c:	Disabled out of date support "hack" for errno.
+
+	* generic/tclBasic.c:	Eliminated VerifyExprObjType.  Initialize
+	errno to zero in OldMathFuncProc.
+
+2005-08-19  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclBasic.c:	Updated OldMathFuncProc and ExprAbsFunc to
+	do less invasion into numeric Tcl_Obj internals.  Made ExprDoubleFunc,
+	ExprIntFunc, ExprWideFunc, and ExprRoundFunc bignum-aware.  Revised
+	ExprSrandFunc error message.
+
+	* generic/tclProc.c:	Wrapped a few tclWideIntType uses in 
+	* generic/tclCmdMZ.c:	#ifndef NO_WIDE_TYPE.
+
+	* generic/tclInt.h:	#define'd NO_WIDE_TYPE.
+
+	* generic/tclVar.c:	Replaced TclPtrIncrVar and TclPtrIncrWideVar
+	* generic/tclInt.h:	with TclPtrIncrObjVar and replaced TclIncrVar2
+	* generic/tclInt.decls:	and TclIncrWideVar2 with TclIncrObjVar2.  New
+	routines call on TclIncrObj to do the work.
+
+	* generic/tclIntDecls.h:	make genstubs
+	* generic/tclStubInit.c:
+
+	* generic/tclCmdIL.c:	Rework Tcl_IncrObjCmd and the INST_*INCR*
+	* generic/tclExecute.c:	opcodes to use the new routines.
+
+2005-08-18  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:		Fixed string rep invalidation bug in
+	* tests/dict.test (dict-11.17):	INST_DICT_INCR_IMM rewrite.
+
+	* generic/tclDictObj.c:	DictIncrCmd rewrite to use TclIncrObj.
+
+	* generic/tclInt.h:	TclIncrObj static -> internal
+	* generic/tclExecute.c:
+
+2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 
 	* generic/tclBasic.c: eliminate a namespace clash caused by
 	BuiltinFuncTable not being static.
@@ -365,6 +765,106 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	clock-50.1 into two tests, with a more permissive check on the
 	error message for an out-of-range value.
 
+2005-08-17  Kevin Kenny  <kennykb@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclBasic.c (Tcl_Expr{Long,Double}{,Obj}): Updated to
+	* generic/tclTest.c:                                deal with
+	* tests/expr-old.test:                              bignums (well, 
+	* tests/expr.test:                                  mostly).
+	Added a missing "errno=0;" in ExprUnaryFunc so that spurious
+	error returns aren't detected. 
+	Added test cases for Tcl_Expr* and Tcl_Expr*Obj because
+	there was very poor test coverage in those areas.
+	* generic/tclParseExpr.c: Reworked parsing of numbers to call
+	TclParseNumber rather than trying to do things locally.
+	* generic/tclStrToD.c: Corrected a comment. Changed so that
+	*endPtrPtr does not include any trailing whitespace.
+	
+2005-08-17  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	New routine TclIncrObj to centralize the
+	increment operation needed in many places.  Updated
+	INST_DICT_INCR_IMM to make use of it.
+
+2005-08-16  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Made bit shifting opcodes and INST_MOD
+	bignum-aware.
+
+	* tests/scan.test:	Making << bignum-aware means that repeated
+	* tests/string.test:	left shifting cannot turn a positive into
+	a negative.  Revised [int_range] and [largest_int] utility commands
+	in the test suite that relied on that happening.  Without revision
+	they became infinite loops.
+
+	* generic/tclExecute.c:	Made binary bitwise opcodes bignum-aware.
+
+	* generic/tclTomMath.h:	Added mp_or and mp_xor to routines from
+	* unix/Makefile.in:	libtommath used by Tcl.
+	* win/Makefile.in:
+	* win/makefile.vc:
+
+2005-08-15  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]	Updates from HEAD.
+	* generic/tclExecute.c:	More revisions to IllegalExprOperandType.
+	Merged INST_BITNOT with INST_UMINUS and make it bignum-aware
+	according to the rule: ~a = -a - 1.  Disabled unused code and
+	noted more TODOs.
+
+	* generic/tclInt.decls: Disabled TclLooksLikeInt() and all callers.
+	* generic/tclUtil.c:
+	* generic/tclCompCmds.c:
+
+	* generic/tclBasic.c:	Rewrite of VerifyExprObjType().
+
+	* generic/tclIntDecls.h:	make genstubs
+	* generic/tclStubInit.c:
+
+	* generic/tclExecute.c:	Updated execution of comparison bytecodes
+	to be bignum-aware, routing string compares through INST_STR_CMP.
+
+2005-08-14  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Updated execution of arithmetic bytecodes
+	to be bignum-aware, and to allow calculations on NaN to produce
+	a NaN result.  INST_UMINUS updated to call mp_neg.
+
+	* generic/tclTomMath.h:	Added mp_and, mp_expt_d, and mp_neg to
+	* unix/Makefile.in:	routines from libtommath used by Tcl.
+	* win/Makefile.in:
+	* win/makefile.vc:
+
+2005-08-13  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclObj.c:	Extended Bignum auto-narrowing to auto-narrow
+	to tclWideIntType when appropriate; this helps keep things working as
+	the bytecode execution code is migrated to supporting bignums.
+
+	* generic/tclExecute.c:	Major overhaul of IllegalExprOperandType.
+	Changed several TclNewFooObj() calls to more logically appropriate
+	ones.  Added several TODO comments marking opportunies for future
+	work.  Made more use of the eePtr->constants.  Made INST_UMINUS
+	bignum aware.
+
+2005-08-12  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Simplify doCondJump.  Use eePtr->constants
+	as result of INST_DICT_NEXT, INST_LAND, and INST_LOR.  Separate
+	INST_LNOT from INST_UMINUS and simplify.
+
 2004-08-12  Kevin Kenny  <kennykb@acm.org>
 
 	* generic/tclClock.c (MktimeObjCmd):
@@ -375,8 +875,25 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	Tcl's time can track system time on Linux even if TZ is not set.
 	Changed ::tcl::clock::Mktime to check for failure, and added a
 	test case that mimics failure but is really success.
+	
+2005-08-11  Don Porter  <dgp@users.sourceforge.net>
 
-2005-08-11  Kevin Kenny  <kennykb@acm.org>
+	[kennykb-numerics-branch]
+
+	* generic/tclExecute.c:	Rewrite of INST_LAND/INST_LOR to take
+	advantage of loss of "pure double" issues.  Merged INST_UPLUS
+	with INST_TRY_CVT_TO_NUMERIC and updated to use improved rules
+	for impure "double"s as well.
+
+	* generic/tclStrToD.c:	Restored conditional generation of
+	tclWideIntType values by TclParseNumber so that Tcl's not
+	completely broken while bignum calculation support is incomplete.
+	The NO_WIDE_TYPE macro can be used to disable this.
+
+	* generic/tclBasic.c (ExprAbsFunc):	First pass making [expr abs(.)]
+	bignum-aware.
+
+2004-08-11  Kevin Kenny  <kennykb@acm.org>
 
 	* generic/tclEvent.c:          Eliminated the USE_THREAD_STORAGE
 	* generic/tclInt.h:            option (which is on in every build
@@ -391,6 +908,24 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	* win/rules.vc:                just a little bit cleaner.)
 	* win/tcl.m4:
 	* win/tclWinThrd.c:
+	
+2005-08-10  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclTomMath.h:	Added mp_shrink, mp_to_unsigned_bin,
+	* unix/Makefile.in:	mp_to_unsigned_bin_n, and mp_unsigned_bin_size
+	* win/Makefile.in:	to routines from libtommath used by Tcl.
+	* win/makefile.vc:
+
+	* generic/tommath.h:	make gentommath_h
+
+	* generic/tclObj.c:	Substantial rewrite to make all number
+	parsing flow through TclParseNumber().  Also established the
+	NO_WIDE_TYPE and BIGNUM_AUTO_NARROW #ifdef's to help track the
+	assumptions of different portions of the code.
+
+	* generic/tclInt.h:	Added NO_WIDE_TYPE #ifdefs
 
 2005-08-10  Kevin Kenny  <kennykb@acm.org>
 
@@ -410,12 +945,24 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	* generic/tclPathObj.c: fix to [file rootname] bug in optimized
 	code path reported on comp.lang.tcl.
 
+2005-08-08  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclObj.c:	Replaced some goto's with loops and started
+	use of BIGNUM_AUTO_NARROW and NO_WIDE_TYPE.
+
 2005-08-06  Donal K. Fellows  <donal.k.fellows@man.ac.uk>
 
 	* generic/tclThreadStorage.c: Stop exposing the guts of the thread
 	storage system through the internal stubs table. Client code
 	should always use the standard API.
 
+2005-08-05  Don Porter  <dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch]
+	* generic/tclObj.c:	Rewrote Tcl_GetDoubleFromObj().
+
 2005-08-05  Donal K. Fellows  <dkf@users.sf.net>
 
 	* unix/tclUnixInit.c (localeTable): Solaris uses a non-standard
@@ -444,6 +991,12 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 
 2005-08-04  Don Porter  <dgp@users.sourceforge.net>
 
+	[kennykb-numerics-branch] Updated from HEAD
+
+	* generic/tclObj.c:	Rewrote Tcl_GetBooleanFromObj() and supporting
+	routines to make use of TclParseNumber.  This reduces the potential
+	number of times a string value must be scanned.
+
 	* generic/tclObj.c:	Simplified routines that manage the typeTable.
 	Deleted the UpdateStringOfBoolean() routine, that can never be called.
 
@@ -454,6 +1007,10 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	* generic/tclInt.h:		[Bug 1251399]
 	* generic/tclObj.c:
 
+2005-08-02  Don Porter	<dgp@users.sourceforge.net>
+
+	[kennykb-numerics-branch] Updated from HEAD
+
 2005-07-30  Daniel Steffen  <das@users.sourceforge.net>
 
 	* unix/tclLoadDyld.c (TclpDlopen, TclpLoadMemory): workarounds
@@ -627,6 +1184,23 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 	on Windows to favor GetComputerName() over gethostname() as
 	a source of the information.
 
+2005-07-12  Kevin Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Updated from HEAD
+
+	* generic/tclCmdMZ.c (Tcl_StringObjCmd):
+	* generic/tclInt.h:
+	* generic/tclObj.c (Tcl_GetDoubleFromObj, SetDoubleFromAny,
+	                    Tcl_GetIntFromObj, SetIntOrWideFromAny):
+	* generic/tclStrToD.c (TclParseNumber, etc.):
+	* tclTomMathInterface.c (TclBNInitBignumFromWideUInt):
+	* tests/obj.test (obj-1.1, obj-2.2, obj-3.1, obj-3.2):
+	
+	Initial attempt at an implementation of TIP #249, comprising
+	a unified parser and modifications to the Tcl_Get*FromObj
+	routines to use it.  Further integration of the parser is
+	necessary and planned.
+	
 2005-07-12  Donal K. Fellows  <dkf@users.sf.net>
 
 	* doc/lsearch.n: Clarify documentation of -exact option; wording was
@@ -1180,8 +1754,14 @@ cvs2005-08-17  George Peter Staplin <GeorgePS@XMission.com>
 2005-05-12  Kevin Kenny	 <kennykb@acm.org>
 
 	* doc/mathfunc.n: Changed NAME line to match the name of the page.
+	
+2005-05-11  Kevin Kenny <kennykb@acm.org>
 
-2005-05-11  Kevin Kenny	 <kennykb@acm.org>
+	[kennykb-numerics-branch] Resynchronized with the HEAD; at this
+	checkpoint [-rkennykb-numerics-branch-20050511], the HEAD and
+	kennykb-numerics-branch contain identical code.
+	
+2005-05-11  Kevin Kenny  <kennykb@acm.org>
 
 	* generic/tclStrToD.c (TclStrToD, RefineResult, ParseNaN):
 	Changed the code to cast 'char' to UCHAR explicitly when