Merged kennykb-numerics-branch back to the head; TIPs 132 and 232

1 files changed, 343 insertions, 0 deletions

diff --git a/ChangeLog b/ChangeLog
index b0027c1..0f13031 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,8 +1,49 @@
+2005-05-10  Kevin Kenny  <kennykb@acm.org>
+
+	Merged all changes on kennykb-numerics-branch back into the
+	HEAD.  TIP's 132 and 232 are now Final.
+	
+2005-05-10  Kevin Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Merged changes from HEAD.
+	
 2005-05-10  Miguel Sofer <msofer@users.sf.net>
 
 	* generic/tclExecute.c (ExponLong, ExponWide): fixed special case
 	'i**0' for i>0 [Bug 1198892]
 
+2005-05-09  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	* win/tclWin32Dll.c (TclpCheckStackSpace, TclWinCPUID):
+	        Reworked structured event handling to function even
+		with -fomit-frame-pointers.
+	
+2005-05-08  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	* generic/tclStrToD.c: Made code more portable by finding a
+	                       workaround for MSVC's 'volatile' issue that
+			       does not require conditional compilation.
+	* win/tclWin32Dll.c (TclWinCPUID): Removed structured event
+	                                   handling from the GCC code
+					   since (a) bad code is generated
+					   by the instruction scheduling
+					   with -O2, and (b) it's not
+					   needed on any reasonably modern
+					   CPU.
+	
+2005-05-07  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	* generic/tclEvent.c:  Moved initialization of tclStrToD.c's
+	* generic/tclInt.h:    static constants into a procedure called
+	* generic/tclStrToD.c: from TclInitSubsystems to avoid double
+	                       checked locking protocol. Cleaned up
+			       an issue where MSVC ignored the 'volatile'
+			       specifier, causing incorrect comparison
+			       of an underflowed number against zero.
+	
 2005-05-06  Jeff Hobbs  <jeffh@ActiveState.com>
 
 	* unix/tcl.m4, unix/configure: correct Solaris 10 (5.10) check and
@@ -10,6 +51,10 @@
 
 2005-05-05  Kevin B. Kenny  <kennykb@acm.org>
 
+	[kennykb-numerics-branch] Merged with HEAD.
+	
+2005-05-05  Kevin B. Kenny  <kennykb@acm.org>
+
 	* win/tclWinThrd.c:  Corrected a compilation error on the
 	                     --enable-threads configuration.
 	
@@ -102,6 +147,33 @@
 	* doc/open.n:		"BINARY" in "access" argument to [open].
 	* tests/ioCmd.test:
 
+2005-04-26  Kevin B. Kenny  <kennykb@users.sourceforge.net>
+
+	* generic/tclBinary.c (FormatNumber):
+	Dredge the NaN out of the internal representation if
+	Tcl_GetDoubleFromObj returns TCL_ERROR on a NaN.
+	
+	* generic/tclObj.c (Tcl_GetDoubleFromObj): 
+	Restored silent overflow/underflow behaviour that the merge
+	of 2004-04-25 messed up.  Thanks to Don Porter for calling 
+	attention to this bug. Also removed an uninitialised memory
+	reference in this function that valgrind caught. Also changed
+	to return TCL_ERROR on a pure NaN.
+	
+	* generic/tclStrToD.c (RefineResult):
+	Added a test for the initial approximation being HUGE_VAL;
+	this test avoids EDOM being returned from ldexp on some platforms
+	on input values exceeding the floating point range.
+	
+	* tests/expr.test (expr-29.*, expr-30.*): 
+	Added further tests of overflow/underflow on input conversions.
+	
+2005-04-25  Kevin B. Kenny  <kennykb@users.sourceforge.net>
+
+	[kennykb-numerics-branch] Merged with HEAD.
+
+	* doc/CrtMathFunc.n:		Revised documentation for TIP 232
+	
 2005-04-25  Daniel Steffen  <das@users.sourceforge.net>
 
 	* compat/string.h: fixed memchr() protoype for __APPLE__ so that we
@@ -230,6 +302,12 @@
 	if this change causes trouble, but it seems more sensible to let
 	Tcl's "on-demand" shimmering rule, and not try to pre-guess things.
 
+2005-04-20  Kevin B. Kenny <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	* doc/expr.n:
+	* doc/mathfunc.n (new file):	Revised documentation for TIP 232
+	
 2005-04-20  Don Porter  <dgp@users.sourceforge.net>
 
 	* generic/tclGet.c (Tcl_GetInt):	Corrected error that did not
@@ -383,6 +461,15 @@
 	Tcl 8.5 (on which it was dependent anyway). Also added a [package
 	require] line to formalize the relationship.
 
+2005-04-11  Kevin Kenny <kennykb@users.sf.net>
+
+        [kennykb-numerics-branch] Merged with HEAD. Updated to libtommath 0.35.
+
+	* generic/tclBasic.c: Attempted to repeat changes that applied
+	to tclExecute.c in Miguel Sofer's commit of 2005-04-01, together
+	with (possibly) a few more uses of his new object creation macros.
+	Also plugged a memory leak in TclObjInvoke. [Bug 1180368]
+	
 2005-04-10  Kevin Kenny  <kennykb@acm.org>
 
 	* library/tzdata/America/Montevideo:
@@ -559,6 +646,29 @@
 	* unix/tcl.m4:    Updated the OpenBSD configuration and regenerated
 	* unix/configure: the configure script.
 	
+2005-03-15  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Merged with HEAD.
+	
+	* generic/tclBasic.c (many):
+	* generic/tclCompExpr.c (CompileMathFuncCall):
+	* generic/tclCompile.h:
+	* generic/tclExecute.c (many):
+	* generic/tclParseExpr.c (ParsePrimaryExpr):
+	* tests/compExpr-old.test:
+	* tests/compExpr.test:
+	* tests/compile.test:
+	* tests/expr-old.test:
+	* tests/expr.test:
+	* tests/for.test:
+	* tests/parseExpr.test:
+	Initial implementation of TIP #232.
+
+	* generic/tclObj.c (Tcl_DbNewBignumObj): Fixed typo that broke
+	--enable-symbols=mem build
+	* tests/binary.test (binary-40.3, binary-40.6): Corrected tests
+	to allow NaN(7ffffffffffff).
+	
 2005-03-14  Miguel Sofer <msofer@users.sf.net>
 
 	* generic/tclExecute.c: fixed INST_PUSH1's debugging code (wrong
@@ -583,6 +693,30 @@
 	into private.  Should be used only by internal workings of
 	execution traces.
 
+2005-03-09  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Merged from HEAD.
+
+	* doc/PrintDbl.3:
+	* doc/tclVars.n: Documented new semantics for tcl_precision.
+	* generic/tclExecute.c (Tcl_ExecuteByteCode): Removed the check
+	for division-by-zero on IEEE-754 machines.
+	* generic/tclUtil.c (Tcl_PrintDouble): Corrected bug where numbers
+	in the range [1e-4 .. 1.) were printed incorrectly.
+	* tests/compExpr-old.test (compExpr-old-11.13): Revised test
+	case for division by zero
+	* tests/expr-old.test (expr-34.11, expr-34.12): Revised test
+	cases for overflow in pow() to deal with infinities.
+	* tests/expr.test (expr-11.13, expr-29.1, expr-29.2): Revised
+	test case for division by zero and for underflow on input
+	conversions.
+	* tests/parseExpr.test (parseExpr-16.11): Revised test case for
+	overflow on input conversion.
+	* tests/string.test (string-6.38 deleted): Removed test case
+	for underflow on input conversion, which is no longer an error.
+	* tests/util.test (util-10.*): Added test case for the bug in
+	tclUtil.c.
+	
 2005-03-08  Jeff Hobbs  <jeffh@ActiveState.com>
 
 	* win/makefile.vc: clarify necessary defined vars that can come
@@ -598,6 +732,82 @@
 	fixed the peephole opt in INST_POP so that it is not used when
 	TCL_COMPILE_DEBUG is defined.
 
+2005-03-04  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	
+	* generic/tclCmdMZ.c: Changed [scan] to treat out-of-range
+	floating point values as infinities and zeroes.
+	* generic/tclExecute.c: Changed [expr] to be permissive about
+	infinities, allowing them to propagate.
+	* generic/tclGet.c: Changed Tcl_GetDouble to be permissive about
+	over/underflow.
+	* generic/tclObj.c: Changed SetDoubleFromAny to be permissive
+	about over/underflow.
+	* generic/tclParseExpr.c: Made [expr] permissive about input
+	numbers out of range.
+	
+2005-03-03  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+
+	* generic/tclInt.h:
+	* generic/tclStrToD.c (Tcl_DoubleDigits, TclFormatNaN):
+	* generic/tclUtil.c (Tcl_PrintDouble):   
+		Changed the signature of TclDoubleDigits so that it
+		accepts a pointer to the signum of the argument, and
+		returns the signum via that pointer.  Added very
+		hacky code to handle IEEE signed zeroes in Tcl_DoubleDigits.
+		(It can't be done other than as a hack until C9x;
+		C89 simply doesn't deal with the concept of -0.0).
+		Added output conversion of tagged NaN values.
+	* generic/tclBinary.c (FormatNumber):
+		Changed to allow [binary format] to handle NaN.
+	* tests/binary.test (binary-60.1):
+		Added a quick-n-dirty test to make sure that NaN's
+		can be scanned and formatted.
+	* generic/tclParseExpr.c (GetLexeme, ParseMaxDoubleLength):
+		Modified so that tagged NaN (e.g., NaN(DEADBEEF)) can
+		be recognized.
+	
+2005-03-02  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Merged with HEAD as of 2005-02-23.
+	
+	* generic/tclExecute.c:
+		Broadened test for NaN to work on Windows.
+	* generic/tclInt.h:		  
+	* generic/tclStrToD.c (Tcl_DoubleDigits):
+	* generic/tclUtil.c (Tcl_PrintDouble, TclPrecTraceProc):
+		Added Tcl_DoubleDigits to format 'double' numbers
+	        with the minimum number of significant digits to
+	        yield correct rounding.  Modified tcl_precision to
+	        accept 0 as a precision (meaning "minimum digits"), and
+	        made 0 the default.  [TIP #132]
+	* generic/tclObj.c:
+		Made NaN's throw an error in Tcl_GetDoubleFromObj.
+	* unix/Makefile.in:
+	* win/Makefile.in:
+	* win/makefile.vc:
+		Added libtommath/bn_mp_init_set.c to the build.
+	* libtommath/tommath.h (mp_iseven): 
+		Fixed a bug that caused zero to test 'odd'.
+	* generic/tommath.h: 
+		Regenerated.
+	* tests/binary.test: 
+	* tests/expr-old.test:
+	* tests/expr.test:
+	* tests/scan.test:
+		Corrected a number of tests that depended on
+		tcl_precision, and removed the {eformat} condition
+		from tests that no longer require it.
+	* tests/util.test: 
+		Corrected a number of tests that depended on
+		tcl_precision, and removed the {eformat} condition
+		from tests that no longer require it.  Added a series
+		of tests for correct rounding in Tcl_PrintDouble. [TIP
+		#132].
+	
 2005-03-01  David N. Welton  <davidw@dedasys.com>
 
 	* doc/CrtSlave.3: Changed to Tcl_Object to Tcl_Obj in the man
@@ -659,6 +869,17 @@
 
 	* doc/CrtChannel.3: Typo: return->returns.
 
+2005-02-06  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	
+	* generic/tclStrToD.c (TclStrToD, SafeLdExp):
+	    Added code to manage the FPU precision on gcc+x86.
+	    Enabled fast conversion of floats with small exponents
+	    now that precision is correct.
+	* tests/expr.test: Corrected test for the smallest representible
+	    value to the right IEEE values.
+	
 2005-02-06  David N. Welton  <davidw@dedasys.com>
 
 	* doc/Thread.3: One-word grammar fix.
@@ -673,6 +894,18 @@
 
 	* generic/tclPathObj.c: Cleaned up typo in comment.
 
+2005-02-03  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+	
+	* generic/tclStrToD.c (TclStrToD, RefineResult, SafeLdExp):
+	    Added code to ensure that 'ldexp' is never called with
+	    a value that will underflow.
+	* tests/expr.test: Added tests for the smallest representible
+	    value, and rounding between it and zero. (The tests reflect
+	    current behaviour; plan is to change the specification of
+	    Tcl so that input conversion of doubles underflows silently.)
+	
 2005-02-02  Mo DeJong  <mdejong@users.sourceforge.net>
 
 	* generic/tclProc.c (TclInitCompiledLocals):
@@ -684,6 +917,35 @@
 	down a crash in Itcl, that crash is fixed by
 	Itcl patch 1115085.
 
+2005-02-01  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]  Merged with HEAD as of today.
+	
+	* generic/tclInt.decls: 
+	    Changed numbers of new stubs to resolve a conflict.
+	* generic/tclInt.h:
+	    Added new TclStrToD routine that replaces the native
+	    'strtod' thro
+	ughout Tcl.
+	* generic/tclCmdMZ (Tcl_StringObjCmd):
+	* generic/tclGet.c (Tcl_GetDouble):
+	* generic/tclObj.c (SetBooleanFromAny, SetDoubleFromAny):
+	* generic/tclParseExpr.c (GetLexeme):
+	* generic/tclScan.c (Tcl_ScanObjCmd):
+	    Replaced all uses of the native 'strtod' with a TclStrToD
+	    routine that performs correct rounding and handles denormals.
+	* generic/tclStrToD.c: (new file)
+	    New scanning function for extracting 'double' from a string
+	    that rounds correctly, and handles denormals and infinities.
+	* unix/Makefile.in:
+	* win/Makefile.in:
+	* win/makefile.vc:
+	    Added tclStrToD.c and the tommath routines that support it.
+
+	These changes represent a partial implementation of TIP #132.
+	Output conversion of floating point numbers, and proper handling
+	of infinities within expressions, still need to be addressed.
+	
 2005-02-01  Don Porter  <dgp@users.sourceforge.net>
 
 	* generic/tclExecute.c (TclCompEvalObj): Removed stray statement
@@ -780,6 +1042,87 @@
 	* doc/FileSystem.3: Add missing ARGUMENTS section definitions for
 	arguments to Tcl_FSLink. [Bug 1106272]
 
+2005-01-21  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch]
+
+	* unix/Makefile.in:     Updated Makefile to build libtommath on
+				Unix as well as Windows. [Bug 1106865]
+
+	* generic/tclTestObj.c (TestbignumobjCmd): 
+	Silenced a compiler warning about a mismatched 'const'.
+	
+2005-01-20  Kevin B. Kenny  <kennykb@acm.org>
+
+	[kennykb-numerics-branch] Development checkpoint.
+	
+	* compat/strtoll.c: 	Reverted to HEAD.
+	* compat/strtoull.c:
+	* doc/Ensemble.3:
+	* generic/tclBasic.c:
+	* generic/tclCmdIL.c:
+	* generic/tclNamesp.c:
+	* generic/tclPathObj.c:
+	* generic/tclPort.h:
+	* unix/configure:
+	* unix/configure.in:
+	* unix/tcl.m4:
+	* win/configure:
+	* win/configure.in:
+	* win/rules.vc:
+	* win/tcl.m4:
+
+	* generic/tcl.h: Added declarations for bignum types, and
+	                 for a 'bignumValue' in the Tcl_Obj structure.
+	* generic/tclInt.h: Added declarations of interface procedures
+	                    for memory allocation in libtommath.
+	
+	* generic/tcl.decls: Added new interface to bignum objects.
+	* generic/tclInt.decls: Added internal stubs for bignum routines
+	                        used by the test code in tclTestObj.c.
+	
+	* generic/tclDecls/h: 		Regen.
+	* generic/tclIntDecls.h:
+	* generic/tclStubInit.h:
+
+	* tools/fix_tommath_h.tcl: (New file) Script to edit 
+				   libtommath/tommath.h and produce
+				   generic/tommath.h so that storage
+				   classes, allocation routines, and
+	                           data types conform to Tcl's
+				   conventions.
+	* generic/tommath.h: (New file) Generated by the above.
+
+	* generic/tclTomMath.h: (New file) Additional declarations
+				to be included in tommath.h when building
+				Tcl.
+
+	* generic/tclTomMathInterface.c: (New file) Small 'glue' routines
+					 adapting tommath's API to Tcl.
+
+	* libtommath/bn_fast_s_mp_mul_digs.c:
+	* libtommath/bn_mp_mul_d.c:
+	* libtommath/bn_mp_read_radix.c:
+	* libtommath/tommath.h: Applied suggested changes from Tom St
+	Denis that correct an off-by-one error in single-digit
+	multiplication (leading to a pointer smash if uncorrected) and
+	change the string argument to 'mp_read_radix' from 'char*' to
+	'const char*'.
+
+	* libtommath/bn_mp_radix_size.c:
+	Local patch to ensure that sufficient memory is requested
+	even if the number has a single digit.
+	
+	* libtommath/bn_mp_read_radix.c:
+	Local patch to return MP_VAL if the input string contains
+	an invalid character.
+
+	* generic/tclObj.c: Added accessor functions for bignums.
+	* generic/tclTestObj.c: Added a 'testbignumobj' command to
+	exercise the accessor functions for bignums.
+
+	* win/Makefile.in: Added rules for making libtommath.
+	
 2005-01-19  Donal K. Fellows  <donal.k.fellows@man.ac.uk>
 
 	TIP#235 IMPLEMENTATION