Diff of /sml/branches/primop-branch-3/NOTES/HISTORY

Parent Directory | Revision Log | Patch

	revision 627, Tue Apr 25 20:40:35 2000 UTC	revision 1220, Wed May 22 15:04:10 2002 UTC
#	Line 8	Line 8
8	The form of an entry should be:	The form of an entry should be:
9
10	Name:	Name:
11	Date:	Date: yyyy/mm/dd
12	Tag: <post-commit CVS tag>	Tag: <post-commit CVS tag>
13	Description:	Description:
14
15		----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/05/22 11:03:00 EDT
18		Tag: blume-20020522-shrink
19		Description:
20
21		John tweaked yesterday's fix for 1131 to handle an out-of-memory
22		situation that comes up when allocating huge arrays.
23
24		----------------------------------------------------------------------
25		Name: Matthias Blume
26		Date: 2002/05/21 16:00:00 EDT
27		Tag: Release_110_40
28		Description:
29
30		New working release (110.40).  New bootfiles.
31
32		[Also: John Reppy fixed GC bug 1131.]
33
34		----------------------------------------------------------------------
35		Name: Matthias Blume
36		Date: 2002/05/21 12:35:00 EDT
37		Tag: blume-20020521-cmdoc
38		Description:
39
40		CM documentation update.
41
42		----------------------------------------------------------------------
43		Name: Matthias Blume
44		Date: 2002/05/21 10:55:00 EDT
45		Tag: blume-20020521-misc
46		Description:
47
48		- John tweaked runtime to be silent on heap export (except when
49		GC messages are on).
50		- I added a few more things (cross-compiling versions of CMB) to
51		config/preloads (as suggestions).
52
53		----------------------------------------------------------------------
54		Name: Matthias Blume
55		Date: 2002/05/20 22:25:00 EDT
56		Tag: blume-20020520-controls
57		Description:
58
59		- Added ControlUtil structure to control-lib.cm.
60		- Use it throughout.
61		- Used Controls facility to define MLRISC controls (as opposed to
62		registering MLRISC control ref cells with Controls after the fact)
63		- Fixed messed-up controls priorities.
64
65		* Removed again all the stuff from config/preloads that one wouldn't
66		be able to preload at the time the initial heap image is built.
67		(Many libraries, e.g., CML, do not exist yet at this time.  The
68		only libraries that can be preloaded via config/preloads are those
69		that come bundled with the bootfiles.)
70
71		----------------------------------------------------------------------
72		Name: Matthias Blume
73		Date: 2002/05/20 10:59:00 EDT
74		Tag: blume-20020520-preloads
75		Description:
76
77		Added a lot of commented-out suggestions for things to be included
78		in config/preloads.
79
80		----------------------------------------------------------------------
81		Name: Allen Leung
82		Date: 2002/05/18 14:20:00 EDT
83		Tag: leunga-20020518-mdl
84		Description:
85
86		o Made the mdl tool stuff compile and run again.
87		o I've disabled all the stuff that depends on RTL specifications; they
88		are all badly broken anyway.
89
90		----------------------------------------------------------------------
91		Name: Matthias Blume
92		Date: 2002/05/17 16:49:00 EDT
93		Tag: blume-20020517-controls
94		Description:
95
96		0. John Reppy made several modifications to the SML/NJ library.
97		In particular, there is a shiny new controls-lib.cm.
98
99		1. Pushed new controls interface through compiler so that everything
100		compiles again.
101
102		2. Added FormatComb and FORMAT_COMB to the CML version of the
103		SML/NJ library (so that CML compiles again).
104
105		3. Modified init scripts because XXX_DEFAULT environment variables
106		are no longer with us.  (Boot-time initialization is now done
107		using the same environment variables that are also used for
108		startup-time initialization of controls.)
109
110		----------------------------------------------------------------------
111		Name: Lal George
112		Date: 2002/05/15 09:20:10 EDT
113		Tag: george-20020515-pseudo-op-decls
114		Description:
115
116		All pseudo-ops emitted before the first segment declaration
117		such as TEXT, DATA, and BSS directives are assumed to be global
118		declarations and are emitted first in the assembly file. This is
119		useful in a number of situations where one has pseudo-ops that are not
120		specific to any segment, and also works around the constraint that one
121		cannot have client pseudo-ops in the TEXT segment.
122
123		Because no segment is associated with these declarations it is
124		an error to allocate any space or objects before the first segment
125		directive and an exception will be raised. However, we cannot make
126		this check for client pseudo-ops.
127
128		These top level declarations are a field in the CFG graph_info.
129		In theory you can continue to add to this field after the CFG has been
130		built -- provided you know what you are doing;-)
131
132		----------------------------------------------------------------------
133		Name: Matthias Blume
134		Date: 2002/05/13 16:40:00 EDT
135		Tag: blume-20020513-pp-etc
136		Description:
137
138		A few minor bugfixes:
139
140		- Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
141		(Bogus printouts for redefined bindings still occur.  Compiler
142		bug should no longer occur now.  We need to redo the prettyprinter
143		from scratch.)
144
145		- CM pathname printer now also adds escape sequences for ( and )
146
147		- commend and docu fixes for ml-nlffi
148
149		----------------------------------------------------------------------
150		Name: Matthias Blume
151		Date: 2002/05/10 16:40:00 EDT
152		Tag: blume-20020510-erg-textio
153		Description:
154
155		Applied the following bugfix provided by Emden Gansner:
156
157		Output is corrupted when outputSubstr is used rather than output.
158
159		The problem occurs when a substring
160
161		ss = (s, dataStart, dataLen)
162
163		where dataStart > 0, fills a stream buffer with avail bytes left.
164		avail bytes of s, starting at index dataStart, are copied into the
165		buffer, the buffer is flushed, and then the remaining dataLen-avail
166		bytes of ss are copied into the beginning of the buffer. Instead of
167		starting this copy at index dataStart+avail in s, the current code
168		starts the copy at index avail.
169
170		Fix:
171		In text-io-fn.sml, change line 695 from
172		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
173		to
174		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
175
176		----------------------------------------------------------------------
177		Name: Matthias Blume
178		Date: 2002/04/12 13:55:00 EDT
179		Tag: blume-20020412-assyntax
180		Description:
181
182		1. Grabbed newer assyntax.h from the XFree86 project.
183		2. Fiddled with how to compile X86.prim.asm without warnings.
184		3. (Very) Minor cleanup in CM.
185
186		----------------------------------------------------------------------
187		Name: Matthias Blume
188		Date: 2002/04/01 (no joke!) 17:07:00 EST
189		Tag: blume-20020401-x86div
190		Description:
191
192		Added full support for div/mod/rem/quot on the x86, using the machine
193		instruction's two results (without clumsily recomputing the remainder)
194		directly where appropriate.
195
196		Some more extensive power-of-two support was added to the x86 instruction
197		selector (avoiding expensive divs, mods, and muls where they can be
198		replaced with cheaper shifts and masks).  However, this sort of thing
199		ought to be done earlier, e.g., within the CPS optimizer so that
200		all architectures benefit from it.
201
202		The compiler compiles to a fixed point, but changes might be somewhat
203		fragile nevertheless.  Please, report any strange things that you might
204		see wrt. div/mod/quot/rem...
205
206		----------------------------------------------------------------------
207		Name: Matthias Blume
208		Date: 2002/03/29 17:22:00
209		Tag: blume-20020329-div
210		Description:
211
212		Fixed my broken div/mod logic.  Unfortunately, this means that the
213		inline code for div/mod now has one more comparison than before.
214		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
215		The problem was with quotient = 0, because that alone does not tell
216		us which way the rounding went.  One then has to look at whether
217		remainder and divisor have the same sign...  :(
218
219		Anyway, I replaced the bootfiles with fresh ones...
220
221		----------------------------------------------------------------------
222		Name: Matthias Blume
223		Date: 2002/03/29 14:10:00 EST
224		Tag: blume-20020329-inlprims
225		Description:
226
227		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
228
229		Primops have changed. This means that the bin/boot-file formats have
230		changed as well.
231
232		To make sure that there is no confusion, I made a new version.
233
234
235		CHANGES:
236
237		* removed REMT from mltree (remainder should never overflow).
238
239		* added primops to deal with divisions of all flavors to the frontend
240
241		* handled these primops all the way through so they map to their respective
242		MLRISC support
243
244		* used these primops in the implementation of Int, Int32, Word, Word32
245
246		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
247
248		* parameterized INLMIN, INLMAX, and INLABS by a numkind
249
250		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
251		floating point
252
253		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
254		and Real (but Real.abs maps to a separate floating-point-only primop)
255
256
257		TODO items:
258
259		* Hacked Alpha32 instruction selection, disabling the selection of REMx
260		instructions because the machine instruction encoder cannot handle
261		them.  (Hppa, PPC, and Sparc instruction selection did not handle
262		REM in the first place, and REM is supported by the x86 machine coder.)
263
264		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
265		selection phase.  (The two can be streamlined because the hardware
266		delivers both quotient and remainder at the same time anyway.)
267
268		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
269		(Currently the behavior is inconsistent both across architectures and
270		wrt. the draft Basis spec.)
271
272		* Word8 should eventually be handled natively, too.
273
274		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
275		as if there currently is no execution path that could trigger it in
276		SML/NJ.  (The assumptions underlying functions arith and promotable do not
277		hold for things like multiplication and division.)
278
279		----------------------------------------------------------------------
280		Name: Matthias Blume
281		Date: 2002/03/27 16:27:00 EST
282		Tag: blume-20020327-mlrisc-divisions
283		Description:
284
285		Added support for all four division operations (ML's div, mod, quot,
286		and rem) to MLRISC.  In the course of doing so, I also rationalized
287		the naming (no more annoying switch-around of DIV and QUOT), by
288		parameterizing the operation by div_rounding_mode (which can be either
289		DIV_TO_ZERO or DIV_TO_NEGINF).
290
291		The generic MLTreeGen functor takes care of compiling all four
292		operations down to only round-to-zero div.
293
294		Missing pieces:
295
296		* Doing something smarter than relying on MLTreeGen on architectures
297		like, e.g., the x86 where hardware division delivers both quotient and
298		remainder at the same time.  With this, the implementation of the
299		round-to-neginf operations could be further streamlined.
300
301		* Remove inlining support for div/mod/rem from the frontend and replace it
302		with primops that get carried through to the backend.  Do this for all
303		int and word types.
304
305		----------------------------------------------------------------------
306		Name: Matthias Blume
307		Date: 2002/03/25 17:25:00 EST
308		Tag: blume-20020325-divmod
309		Description:
310
311		I improved (hopefully without breaking them) the implementation of Int.div,
312		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
313		advantage of the following observations:
314
315		Let  q = x quot y      r = x rem y
316		d = x div  y      m = x mod y
317
318		where "quot" is the round-to-zero version of integer division that
319		hardware usually provides.  Then we have:
320
321		r = x - q * y        where neither the * nor the - will overflow
322		d = if q >= 0 orelse x = q * y then q else q - 1
323		where neither the * nor the - will overflow
324		m = if q >= 0 orelse r = 0 then r else r + y
325		where the + will not overflow
326
327		This results in substantial simplification of the generated code.
328		The following table shows the number of CFG nodes and edges generated
329		for
330		fun f (x, y) = x OPER y
331		(* with OPER \in div, mod, quot, rem *)
332
333
334		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
335		--------------------------------------------------------
336		div |         24 |         39 |         12 |         16
337		mod |         41 |         71 |         12 |         16
338		quot |          8 |         10 |          8 |         10
339		rem |         10 |         14 |          8 |         10
340
341
342		----------------------------------------------------------------------
343		Name: Matthias Blume
344		Date: 2002/03/25 22:06:00 EST
345		Tag: blume-20020325-cprotobug
346		Description:
347
348		Fixed a bug in cproto (c prototype decoder).
349
350		----------------------------------------------------------------------
351		Name: Matthias Blume
352		Date: 2002/03/25 16:00:00 EST
353		Tag: blume-20020325-raw-primops
354		Description:
355
356		I did some cleanup to Allen's new primop code and
357		replaced yesterday's bootfiles with new ones.
358		(But they are stored in the same place.)
359
360		----------------------------------------------------------------------
361		Name: Matthias Blume
362		Date: 2002/03/24 22:40:00 EST
363		Tag: blume-20020324-bootfiles
364		Description:
365
366		Made the bootfiles that Allen asked for.
367
368		----------------------------------------------------------------------
369		Name: Allen Leung
370		Date: 2002/03/23 15:50:00 EST
371		Tag: leunga-20020323-flint-cps-rcc-primops
372		Description:
373
374		1. Changes to FLINT primops:
375
376		(* make a call to a C-function;
377		* The primop carries C function prototype information and specifies
378		* which of its (ML-) arguments are floating point. C prototype
379		* information is for use by the backend, ML information is for
380		* use by the CPS converter. *)
381		| RAW_CCALL of { c_proto: CTypes.c_proto,
382		ml_args: ccall_type list,
383		ml_res_opt: ccall_type option,
384		reentrant : bool
385		} option
386		(* Allocate uninitialized storage on the heap.
387		* The record is meant to hold short-lived C objects, i.e., they
388		* are not ML pointers.  With the tag, the representation is
389		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
390		*)
391		| RAW_RECORD of {tag:bool,sz:int}
392		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
393
394		2.  These CPS primops are now overloaded:
395
396		rawload of {kind:numkind}
397		rawstore of {kind:numkind}
398
399		The one argument form is:
400
401		rawload {kind} address
402
403		The two argument form is:
404
405		rawload {kind} [ml object, byte-offset]
406
407		3. RAW_CCALL/RCC now takes two extra arguments:
408
409		a. The first is whether the C call is reentrant, i.e., whether
410		ML state should be saved and restored.
411		b. The second argument is a string argument specifying the name of
412		library and the C function.
413
414		These things are currently not handled in the code generator, yet.
415
416		4. In CProto,
417
418		An encoding type of "bool" means "ml object" and is mapped into
419		C prototype of PTR.  Note that "bool" is different than "string",
420		even though "string" is also mapped into PTR, because "bool"
421		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
422
423		5. Pickler/unpicker
424
425		Changed to handle RAW_RECORD and newest RAW_CCALL
426
427		6. MLRiscGen,
428
429		1. Changed to handle the new rawload/rawstore/rawrecord operators.
430		2. Code for handling C Calls has been moved to a new module CPSCCalls,
431		in the file CodeGen/cpscompile/cps-c-calls.sml
432
433		7. Added the conditional move operator
434
435		condmove of branch
436
437		to cps.  Generation of this is still buggy so it is currently
438		disabled.
439
440		----------------------------------------------------------------------
441		Name: Lal George
442		Date: 2002/03/22 14:18:25 EST
443		Tag: george-20020322-cps-branch-prob
444		Description:
445
446		Implemented the Ball-Larus branch prediction-heuristics, and
447		incorporated graphical viewers for control flow graphs.
448
449		Ball-Larus Heuristics:
450		---------------------
451		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
452
453		By design it uses the Dempster-Shafer theory for combining
454		probabilities.  For example, in the function:
455
456		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
457
458		the ball-larus heuristics predicts that the n=0 is unlikely
459		(OH-heuristic), and the 'then' branch is unlikely because of the
460		RH-heuristic -- giving the 'then' branch an even lower combined
461		probability using the Dempster-Shafer theory.
462
463		Finally, John Reppy's loop analysis in MLRISC, further lowers the
464		probability of the 'then' branch because of the loop in the else
465		branch.
466
467
468		Graphical Viewing:
469		------------------
470		I merely plugged in Allen's graphical viewers into the compiler. The
471		additional code is not much. At the top level, saying:
472
473		Control.MLRISC.getFlag "cfg-graphical-view" := true;
474
475		will display the graphical view of the control flow graph just before
476		back-patching.  daVinci must be in your path for this to work. If
477		daVinci is not available, then the default viewer can be changed
478		using:
479
480		Control.MLRISC.getString "viewer"
481
482		which can be set to "dot" or "vcg" for the corresponding viewers. Of
483		course, these viewers must be in your path.
484
485		The above will display the compilation unit at the level of clusters,
486		many of which are small, boring, and un-interesting. Also setting:
487
488		Control.MLRISC.getInt "cfg-graphical-view_size"
489
490		will display clusters that are larger than the value set by the above.
491
492
493		----------------------------------------------------------------------
494		Name: Matthias Blume
495		Date: 2002/03/21 22:20:00 EST
496		Tag: blume-20020321-kmp-bugfix
497		Description:
498
499		Changed the interface to the KMP routine in PreString and fixed
500		a minor bug in one place where it was used.
501
502		----------------------------------------------------------------------
503		Name: Allen Leung
504		Date: 2002/03/21 20:30:00 EST
505		Tag: leunga-20020321-cfg
506		Description:
507
508		Fixed a potential problem in cfg edge splitting.
509
510		----------------------------------------------------------------------
511		Name: Allen Leung
512		Date: 2002/03/21 17:15:00 EST
513		Tag: leunga-20020321-x86-fp-cfg
514		Description:
515
516		1. Recoded the buggy parts of x86-fp.
517
518		a. All the block reordering code has been removed.
519		We now depend on the block placement phases to do this work.
520
521		b. Critical edge splitting code has been simplified and moved into the
522		CFG modules, as where they belong.
523
524		Both of these were quite buggy and complex.  The code is now much, much
525		simpler.
526
527		2. X86 backend.
528
529		a. Added instructions for 64-bit support.  Instruction selection for
530		64-bit has not been committed, however, since that
531		requires changes to MLTREE which haven't been approved by
532		Lal and John.
533
534		b. Added support for FUCOMI and FUCOMIP when generating code for
535		PentiumPro and above.  We only generate these instructions in
536		the fast-fp mode.
537
538		c. Added cases for JP and JNP in X86FreqProps.
539
540		3. CFG
541
542		CFG now has a bunch of methods for edge splitting and merging.
543
544		4. Machine description.
545
546		John's simplification of MLTREE_BASIS.fcond broke a few machine
547		description things:
548
549		rtl-build.{sig,sml} and hppa.mdl fixed.
550
551		NOTE: the machine description stuff in the repository is still broken.
552		Again, I can't put my fixes in because that involves
553		changes to MLTREE.
554
555		----------------------------------------------------------------------
556		Name: Matthias Blume
557		Date: 2002/03/20 15:55:00 EST
558		Tag: blume-20020320-kmp
559		Description:
560
561		Implemented Knuth-Morris-Pratt string matching in PreString and used
562		it for String.isSubstring, Substring.isSubstring, and
563		Substring.position.
564
565		(Might need some stress-testing.  Simple examples worked fine.)
566
567		----------------------------------------------------------------------
568		Name: Matthias Blume
569		Date: 2002/03/19 16:37:00 EST
570		Tag: blume-20020319-witnesses
571		Description:
572
573		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
574
575		This implements a generic mechanism for changing constness qualifiers
576		anywhere within big C types without resorting to outright "casts".
577		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
578		modify the constness at the outermost level.)
579		The implementation of "convert" is based on the idea of "witness"
580		values -- values that are not used by the operation but whose types
581		"testify" to their applicability.  On the implementation side, "convert"
582		is simply a projection (returning its second curried argument).  With
583		cross-module inlining, it should not result in any machine code being
584		generated.
585
586		----------------------------------------------------------------------
587		Name: Matthias Blume
588		Date: 2002/03/15 16:40:00 EST
589		Tag: blume-20020315-basis
590		Description:
591
592		Provided (preliminary?) implementations for
593
594		{String,Substring}.{concatWith,isSuffix,isSubstring}
595
596		and
597
598		Substring.full
599
600		Those are in the Basis spec but they were missing in SML/NJ.
601
602		----------------------------------------------------------------------
603		Name: Matthias Blume
604		Date: 2002/03/14 21:30:00 EST
605		Tag: blume-20020314-controls
606		Description:
607
608		Controls:
609		---------
610
611		1. Factored out the recently-added Controls : CONTROLS stuff and put
612		it into its own library $/controls-lib.cm.  The source tree for
613		this is under src/smlnj-lib/Controls.
614
615		2. Changed the names of types and functions in this interface, so they
616		make a bit more "sense":
617
618		module -> registry
619		'a registry -> 'a group
620
621		3. The interface now deals in ref cells only.  The getter/setter interface
622		is (mostly) gone.
623
624		4. Added a function that lets one register an already-existing ref cell.
625
626		5. Made the corresponding modifications to the rest of the code so that
627		everything compiles again.
628
629		6. Changed the implementation of Controls.MLRISC back to something closer
630		to the original.  In particular, this module (and therefore MLRISC)
631		does not depend on Controls.  There now is some link-time code in
632		int-sys.sml that registers the MLRISC controls with the Controls
633		module.
634
635		CM:
636		---
637
638		* One can now specify the lambda-split aggressiveness in init.cmi.
639
640		----------------------------------------------------------------------
641		Name: Allen Leung
642		Date: 2002/03/13 17:30:00 EST
643		Tag: leunga-20020313-x86-fp-unary
644		Description:
645
646		Bug fix for:
647
648		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
649		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
650		> - fun f(x,(y,z)) = Real.~ y;
651		> [autoloading]
652		> [autoloading done]
653		>       fchsl   (%eax), 184(%esp)
654		> Error: MLRisc bug: X86MCEmitter.emitInstr
655		>
656		> uncaught exception Error
657		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
658
659		The problem was that the code generator did not generate any fp registers
660		in this case, and the ra didn't know that it needed to run the X86FP phase to
661		translate the pseudo fp instruction.   This only happened with unary fp
662		operators in certain situations.
663
664		----------------------------------------------------------------------
665		Name: Matthias Blume
666		Date: 2002/03/13 14:00:00 EST
667		Tag: blume-20020313-overload-etc
668		Description:
669
670		1. Added _overload as a synonym for overload for backward compatibility.
671		(Control.overloadKW must be true for either version to be accepted.)
672
673		2. Fixed bug in install script that caused more things to be installed
674		than what was requested in config/targets.
675
676		3. Made CM aware of the (_)overload construct so that autoloading
677		works.
678
679		----------------------------------------------------------------------
680		Name: Matthias Blume
681		Date: 2002/03/12 22:03:00 EST
682		Tag: blume-20020312-url
683		Description:
684
685		Forgot to update BOOT and srcarchiveurl.
686
687		----------------------------------------------------------------------
688		Name: Matthias Blume
689		Date: 2002/03/12 17:30:00 EST
690		Tag: blume-20020312-version110392
691		Description:
692
693		Yet another version number bump (because of small changes to the
694		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
695
696		Changes:
697
698		The new pid generation scheme described a few weeks ago was overly
699		complicated.  I implemented a new mechanism that is simpler and
700		provides a bit more "stability":  Once CM has seen a compilation
701		unit, it keeps its identity constant (as long as you do not delete
702		those crucial CM/GUID/* files).  This means that when you change
703		an interface, compile, then go back to the old interface, and
704		compile again, you arrive at the original pid.
705
706		There now also is a mechanism that instructs CM to use the plain
707		environment hash as a module's pid (effectively making its GUID
708		the empty string).  For this, "noguid" must be specified as an
709		option to the .sml file in question within its .cm file.
710		This is most useful for code that is being generated by tools such
711		as ml-nlffigen (because during development programmers tend to
712		erase the tool's entire output directory tree including CM's cached
713		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
714		revert to the old, broken behavior of SML/NJ, but in specific cases
715		where there is no danger of interface confusion, its use is ok
716		(I think).
717
718		ml-nlffigen by default generates "noguid" annotations.  They can be
719		turned off by specifying -guid in its command line.
720
721		----------------------------------------------------------------------
722		Name: Lal George
723		Date: 2002/03/12 12 14:42:36 EST
724		Tag: george-20020312-frequency-computation
725		Description:
726
727		Integrated jump chaining and static block frequency into the
728		compiler. More details and numbers later.
729
730		----------------------------------------------------------------------
731		Name: Lal George
732		Date: 2002/03/11 11 22:38:53 EST
733		Tag: george-20020311-jump-chain-elim
734		Description:
735
736		Tested the jump chain elimination on all architectures (except the
737		hppa).  This is on by default right now and is profitable for the
738		alpha and x86, however, it may not be profitable for the sparc and ppc
739		when compiling the compiler.
740
741		The gc test will typically jump to a label at the end of the cluster,
742		where there is another jump to an external cluster containing the actual
743		code to invoke gc. This is to allow factoring of common gc invocation
744		sequences. That is to say, we generate:
745
746		f:
747		testgc
748		ja   L1      % jump if above to L1
749
750		L1:
751		jmp L2
752
753
754		After jump chain elimination the 'ja L1' instructions is converted to
755		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
756		up being implemented in their long form (if L2 is far away) using:
757
758		jbe     L3      % jump if below or equal to L3
759		jmp     L2
760		L3:
761		...
762
763
764		For large compilation units L2  may be far away.
765
766
767		----------------------------------------------------------------------
768		Name: Matthias Blume
769		Date: 2002/03/11 13:30:00 EST
770		Tag: blume-20020311-mltreeeval
771		Description:
772
773		A functor parameter was missing.
774
775		----------------------------------------------------------------------
776		Name: Allen Leung
777		Date: 2002/03/11 10:30:00 EST
778		Tag: leunga-20020311-runtime-string0
779		Description:
780
781		The representation of the empty string now points to a
782		legal null terminated C string instead of unit.  It is now possible
783		to convert an ML string into C string with InlineT.CharVector.getData.
784		This compiles into one single machine instruction.
785
786		----------------------------------------------------------------------
787		Name: Allen Leung
788		Date: 2002/03/10 23:55:00 EST
789		Tag: leunga-20020310-x86-call
790		Description:
791
792		Added machine generation for CALL instruction (relative displacement mode)
793
794		----------------------------------------------------------------------
795		Name: Matthias Blume
796		Date: 2002/03/08 16:05:00
797		Tag: blume-20020308-entrypoints
798		Description:
799
800		Version number bumped to 110.39.1.  NEW BOOTFILES!
801
802		Entrypoints: non-zero offset into a code object where execution should begin.
803
804		- Added the notion of an entrypoint to CodeObj.
805		- Added reading/writing of entrypoint info to Binfile.
806		- Made runtime system bootloader aware of entrypoints.
807		- Use the address of the label of the first function given to mlriscGen
808		as the entrypoint.  This address is currently always 0, but it will
809		not be 0 once we turn on block placement.
810		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
811		with entry points) from mlriscGen.
812
813		----------------------------------------------------------------------
814		Name: Allen Leung
815		Date: 2002/03/07 20:45:00 EST
816		Tag: leunga-20020307-x86-cmov
817		Description:
818
819		Bug fixes for CMOVcc on x86.
820
821		1. Added machine code generation for CMOVcc
822		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
823		3. CMOVcc cannot have an immediate operand as argument.
824
825		----------------------------------------------------------------------
826		Name: Matthias Blume
827		Date: 2002/03/07 16:15:00 EST
828		Tag: blume-20020307-controls
829		Description:
830
831		This is a very large but mostly boring patch which makes (almost)
832		every tuneable compiler knob (i.e., pretty much everything under
833		Control.* plus a few other things) configurable via both the command
834		line and environment variables in the style CM did its configuration
835		until now.
836
837		Try starting sml with '-h' (or, if you are brave, '-H')
838
839		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
840		implements the underlying generic mechanism.
841
842		The interface to some of the existing such facilities has changed somewhat.
843		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
844		(The getFoo interface is still there for backward-compatibility, but its
845		use is deprecated.)
846
847		The ml-build script passes -Cxxx=yyy command-line arguments through so
848		that one can now twiddle the compiler settings when using this "batch"
849		compiler.
850
851		TODO items:
852
853		We should go through and throw out all controls that are no longer
854		connected to anything.  Moreover, we should go through and provide
855		meaningful (and correct!) documentation strings for those controls
856		that still are connected.
857
858		Currently, multiple calls to Controls.new are accepted (only the first
859		has any effect).  Eventually we should make sure that every control
860		is being made (via Controls.new) exactly once.  Future access can then
861		be done using Controls.acc.
862
863		Finally, it would probably be a good idea to use the getter-setter
864		interface to controls rather than ref cells.  For the time being, both
865		styles are provided by the Controls module, but getter-setter pairs are
866		better if thread-safety is of any concern because they can be wrapped.
867
868		*****************************************
869
870		One bug fix: The function blockPlacement in three of the MLRISC
871		backpatch files used to be hard-wired to one of two possibilities at
872		link time (according to the value of the placementFlag).  But (I
873		think) it should rather sense the flag every time.
874
875		*****************************************
876
877		Other assorted changes (by other people who did not supply a HISTORY entry):
878
879		1. the cross-module inliner now works much better (Monnier)
880		2. representation of weights, frequencies, and probabilities in MLRISC
881		changed in preparation of using those for weighted block placement
882		(Reppy, George)
883
884		----------------------------------------------------------------------
885		Name: Lal George
886		Date: 2002/03/07 14:44:24 EST 2002
887		Tag: george-20020307-weighted-block-placement
888
889		Tested the weighted block placement optimization on all architectures
890		(except the hppa) using AMPL to generate the block and edge frequencies.
891		Changes were required in the machine properties to correctly
892		categorize trap instructions. There is an MLRISC flag
893		"weighted-block-placement" that can be used to enable weighted block
894		placement, but this will be ineffective without block/edge
895		frequencies (coming soon).
896
897
898		----------------------------------------------------------------------
899		Name: Lal George
900		Date: 2002/03/05 17:24:48 EST
901		Tag: george-20020305-linkage-cluster
902
903		In order to support the block placement optimization, a new cluster
904		is generated as the very first cluster (called the linkage cluster).
905		It contains a single jump to the 'real' entry point for the compilation
906		unit. Block placement has no effect on the linkage cluster itself, but
907		all the other clusters  have full freedom in the manner in which they
908		reorder blocks or functions.
909
910		On the x86 the typical linkage code that is generated is:
911		----------------------
912		.align 2
913		L0:
914		addl    $L1-L0, 72(%esp)
915		jmp     L1
916
917
918		.align  2
919		L1:
920		----------------------
921
922		72(%esp) is the memory location for the stdlink register. This
923		must contain the address of the CPS function being called. In the
924		above example, it contains the address of  L0; before
925		calling L1 (the real entry point for the compilation unit), it
926		must contain the address for L1, and hence
927
928		addl $L1-L0, 72(%esp)
929
930		I have tested this on all architectures except the hppa.The increase
931		in code size is of course negligible
932
933		----------------------------------------------------------------------
934		Name: Allen Leung
935		Date: 2002/03/03 13:20:00 EST
936		Tag: leunga-20020303-mlrisc-tools
937
938		Added #[ ... ] expressions to mlrisc tools
939
940		----------------------------------------------------------------------
941		Name: Matthias Blume
942		Date: 2002/02/27 12:29:00 EST
943		Tag: blume-20020227-cdebug
944		Description:
945
946		- made types in structure C and C_Debug to be equal
947		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
948		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
949
950		----------------------------------------------------------------------
951		Name: Matthias Blume
952		Date: 2002/02/26 12:00:00 EST
953		Tag: blume-20020226-ffi
954		Description:
955
956		1. Fixed a minor bug in CM's "noweb" tool:
957		If numbering is turned off, then truly don't number (i.e., do not
958		supply the -L option to noweb).  The previous behavior was to supply
959		-L'' -- which caused noweb to use the "default" line numbering scheme.
960		Thanks to Chris Richards for pointing this out (and supplying the fix).
961
962		2. Once again, I reworked some aspects of the FFI:
963
964		A. The incomplete/complete type business:
965
966		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
967		gone!
968		- ML types representing an incomplete type are now *equal* to
969		ML types representing their corresponding complete types (just like
970		in C).  This is still safe because ml-nlffigen will not generate
971		RTTI for incomplete types, nor will it generate functions that
972		require access to such RTTI.   But when ML code generated from both
973		incomplete and complete versions of the C type meet, the ML types
974		are trivially interoperable.
975
976		NOTE:  These changes restore the full generality of the translation
977		(which was previously lost when I eliminated functorization)!
978
979		B. Enum types:
980
981		- Structure C now has a type constructor "enum" that is similar to
982		how the "su" constructor works.  However, "enum" is not a phantom
983		type because each "T enum" has values (and is isomorphic to
984		MLRep.Signed.int).
985		- There are generic access operations for enum objects (using
986		MLRep.Signed.int).
987		- ml-nlffigen will generate a structure E_foo for each "enum foo".
988		* The structure contains the definition of type "mlrep" (the ML-side
989		representation type of the enum).  Normally, mlrep is the same
990		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
991		then mlrep will be defined as a datatype -- thus facilitating
992		pattern matching on mlrep values.
993		("-ec" will be suppressed if there are duplicate values in an
994		enumeration.)
995		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
996		will be generated for each C enum constant xxx.
997		* Conversion functions m2i and i2m convert between mlrep and
998		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
999		* Coversion functions c and ml convert between mlrep and "tag enum".
1000		* Access functions (get/set) fetch and store mlrep values.
1001		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
1002		enumerations are merged into one single enumeration represented by
1003		structure E_'.
1004
1005		----------------------------------------------------------------------
1006		Name: Allen Leung
1007		Date: 2002/02/25 04:45:00 EST
1008		Tag: leunga-20020225-cps-spill
1009
1010		This is a new implementation of the CPS spill phase.
1011		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1012		In case of problems, replace it with the old file spill.sml
1013
1014		The current compiler runs into some serious performance problems when
1015		constructing a large record.  This can happen when we try to compile a
1016		structure with many items.  Even a very simple structure like the following
1017		makes the compiler slow down.
1018
1019		structure Foo = struct
1020		val x_1 = 0w1 : Word32.int
1021		val x_2 = 0w2 : Word32.int
1022		val x_3 = 0w3 : Word32.int
1023		...
1024		val x_N = 0wN : Word32.int
1025		end
1026
1027		The following table shows the compile time, from N=1000 to N=4000,
1028		with the old compiler:
1029
1030		N
1031		1000   CPS 100 spill                           0.04u  0.00s  0.00g
1032		MLRISC ra                               0.06u  0.00s  0.05g
1033		(spills = 0 reloads = 0)
1034		TOTAL                                   0.63u  0.07s  0.21g
1035
1036		1100   CPS 100 spill                           8.25u  0.32s  0.64g
1037		MLRISC ra                               5.68u  0.59s  3.93g
1038		(spills = 0 reloads = 0)
1039		TOTAL                                   14.71u  0.99s  4.81g
1040
1041		1500   CPS 100 spill                           58.55u  2.34s  1.74g
1042		MLRISC ra                               5.54u  0.65s  3.91g
1043		(spills = 543 reloads = 1082)
1044		TOTAL                                   65.40u  3.13s  6.00g
1045
1046		2000   CPS 100 spill                           126.69u  4.84s  3.08g
1047		MLRISC ra                               0.80u  0.10s  0.55g
1048		(spills = 42 reloads = 84)
1049		TOTAL                                   129.42u  5.10s  4.13g
1050
1051		3000   CPS 100 spill                           675.59u  19.03s  11.64g
1052		MLRISC ra                               2.69u  0.27s  1.38g
1053		(spills = 62 reloads = 124)
1054		TOTAL                                   682.48u  19.61s  13.99g
1055
1056		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1057		MLRISC ra                               4.96u  0.27s  2.72g
1058		(spills = 85 reloads = 170)
1059		TOTAL                                   2375.26u  57.21s  48.00g
1060
1061		As you can see the old cps spill module suffers from some serious
1062		performance problem.  But since I cannot decipher the old code fully,
1063		instead of patching the problems up, I'm reimplementing it
1064		with a different algorithm.  The new code is more modular,
1065		smaller when compiled, and substantially faster
1066		(O(n log n) time and O(n) space).  Timing of the new spill module:
1067
1068		4000  CPS 100 spill                           0.02u  0.00s  0.00g
1069		MLRISC ra                               0.25u  0.02s  0.15g
1070		(spills=1 reloads=3)
1071		TOTAL                                   7.74u  0.34s  1.62g
1072
1073		Implementation details:
1074
1075		As far as I can tell, the purpose of the CPS spill module is to make sure the
1076		number of live variables at any program point (the bandwidth)
1077		does not exceed a certain limit, which is determined by the
1078		size of the spill area.
1079
1080		When the bandwidth is too large, we decrease the register pressure by
1081		packing live variables into spill records.  How we achieve this is
1082		completely different than what we did in the old code.
1083
1084		First, there is something about the MLRiscGen code generator
1085		that we should be aware of:
1086
1087		o MLRiscGen performs code motion!
1088
1089		In particular, it will move floating point computations and
1090		address computations involving only the heap pointer to
1091		their use sites (if there is only a single use).
1092		What this means is that if we have a CPS record construction
1093		statement
1094
1095		RECORD(k,vl,w,e)
1096
1097		we should never count the new record address w as live if w
1098		has only one use (which is often the case).
1099
1100		We should do something similar to floating point, but the transformation
1101		there is much more complex, so I won't deal with that.
1102
1103		Secondly, there are now two new cps primops at our disposal:
1104
1105		1. rawrecord of record_kind option
1106		This pure operator allocates some uninitialized storage from the heap.
1107		There are two forms:
1108
1109		rawrecord NONE [INT n]  allocates a tagless record of length n
1110		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1111		and initializes the tag.
1112
1113		2. rawupdate of cty
1114		rawupdate cty (v,i,x)
1115		Assigns to x to the ith component of record v.
1116		The storelist is not updated.
1117
1118		We use these new primops for both spilling and increment record construction.
1119
1120		1. Spilling.
1121
1122		This is implemented with a linear scan algorithm (but generalized
1123		to trees).  The algorithm will create a single spill record at the
1124		beginning of the cps function and use rawupdate to spill to it,
1125		and SELECT or SELp to reload from it.  So both spills and reloads
1126		are fine-grain operations.  In contrast, in the old algorithm
1127		"spills" have to be bundled together in records.
1128
1129		Ideally, we should sink the spill record construction to where
1130		it is needed.  We can even split the spill record into multiple ones
1131		at the places where they are needed.  But CPS is not a good
1132		representation for global code motion, so I'll keep it simple and
1133		am not attempting this.
1134
1135		2. Incremental record construction (aka record splitting).
1136
1137		Long records with many component values which are simulatenously live
1138		(recall that single use record addresses are not considered to
1139		be live) are constructed with rawrecord and rawupdate.
1140		We allocate space on the heap with rawrecord first, then gradually
1141		fill it in with rawupdate.  This is the technique suggested to me
1142		by Matthias.
1143
1144		Some restrictions on when this is applicable:
1145		1. It is not a VECTOR record.  The code generator currently does not handle
1146		this case. VECTOR record uses double indirection like arrays.
1147		2. All the record component values are defined in the same "basic block"
1148		as the record constructor.  This is to prevent speculative
1149		record construction.
1150
1151		----------------------------------------------------------------------
1152		Name: Allen Leung
1153		Date: 2002/02/22 01:02:00 EST
1154		Tag: leunga-20020222-mlrisc-tools
1155
1156		Minor bug fixes in the parser and rewriter
1157
1158		----------------------------------------------------------------------
1159		Name: Allen Leung
1160		Date: 2002/02/21 20:20:00 EST
1161		Tag: leunga-20020221-peephole
1162
1163		Regenerated the peephole files.  Some contained typos in the specification
1164		and some didn't compile because of pretty printing bugs in the old version
1165		of 'nowhere'.
1166
1167		----------------------------------------------------------------------
1168		Name: Allen Leung
1169		Date: 2002/02/19 20:20:00 EST
1170		Tag: leunga-20020219-mlrisc-tools
1171		Description:
1172
1173		Minor bug fixes to the mlrisc-tools library:
1174
1175		1.  Fixed up parsing colon suffixed keywords
1176		2.  Added the ability to shut the error messages up
1177		3.  Reimplemented the pretty printer and fixed up/improved
1178		the pretty printing of handle and -> types.
1179		4.  Fixed up generation of literal symbols in the nowhere tool.
1180		5.  Added some SML keywords to to sml.sty
1181
1182		----------------------------------------------------------------------
1183		Name: Matthias Blume
1184		Date: 2002/02/19 16:20:00 EST
1185		Tag: blume-20020219-cmffi
1186		Description:
1187
1188		A wild mix of changes, some minor, some major:
1189
1190		* All C FFI-related libraries are now anchored under $c:
1191		$/c.cm      --> $c/c.cm
1192		$/c-int.cm  --> $c/internals/c-int.cm
1193		$/memory.cm --> $c/memory/memory.cm
1194
1195		* "make" tool (in CM) now treats its argument pathname slightly
1196		differently:
1197		1. If the native expansion is an absolute name, then before invoking
1198		the "make" command on it, CM will apply OS.Path.mkRelative
1199		(with relativeTo = OS.FileSys.getDir()) to it.
1200		2. The argument will be passed through to subsequent phases of CM
1201		processing without "going native".  In particular, if the argument
1202		was an anchored path, then "make" will not lose track of that anchor.
1203
1204		* Compiler backends now "know" their respective C calling conventions
1205		instead of having to be told about it by ml-nlffigen.  This relieves
1206		ml-nlffigen from one of its burdens.
1207
1208		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1209
1210		* Export C_DEBUG and C_Debug from $c/c.cm.
1211
1212		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
1213		subtyping relationship between incomplete pointers and their complete
1214		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1215		with the convention of instantiating 'o with ('t, 'c) obj whenever
1216		the pointer target type is complete.  In the incomplete case, 'o
1217		will be instantiated with some "'c iobj" -- a type obtained by
1218		using one of the functors PointerToIncompleteType or PointerToCompleteType.
1219
1220		Operations that work on both incomplete and complete pointer types are
1221		typed as taking an 'o ptr while operations that require the target to
1222		be known are typed as taking some ('t, 'c) obj ptr.
1223
1224		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1225		where void is an eqtype without any values.  This makes it possible
1226		to work on voidptr values using functions meant to operate on light
1227		incomplete pointers.
1228
1229		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1230		been vastly simplified.
1231
1232		----------------------------------------------------------------------
1233		Name: Matthias Blume
1234		Date: 2002/02/19 10:48:00 EST
1235		Tag: blume-20020219-pqfix
1236		Description:
1237
1238		Applied Chris Okasaki's bug fix for priority queues.
1239
1240		----------------------------------------------------------------------
1241		Name: Matthias Blume
1242		Date: 2002/02/15 17:05:00
1243		Tag: Release_110_39
1244		Description:
1245
1246		Last-minute retagging is becoming a tradition... :-(
1247
1248		This is the working release 110.39.
1249
1250		----------------------------------------------------------------------
1251		Name: Matthias Blume
1252		Date: 2002/02/15 16:00:00 EST
1253		Tag: Release_110_39-orig
1254		Description:
1255
1256		Working release 110.39.  New bootfiles.
1257
1258		(Update: There was a small bug in the installer so it wouldn't work
1259		with all shells.  So I retagged. -Matthias)
1260
1261		----------------------------------------------------------------------
1262		Name: Matthias Blume
1263		Date: 2002/02/15 14:17:00 EST
1264		Tag: blume-20020215-showbindings
1265		Description:
1266
1267		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1268		the latter can be useful for exploring what bindings are available at
1269		the interactive prompt.  (The first function returns only the list
1270		of symbols that are really bound, the second prints those but also the
1271		ones that CM's autoloading mechanism knows about.)
1272
1273		----------------------------------------------------------------------
1274		Name: Matthias Blume
1275		Date: 2002/02/15 12:08:00 EST
1276		Tag: blume-20020215-iptrs
1277		Description:
1278
1279		Two improvements to ml-nlffigen:
1280
1281		1. Write files only if they do not exist or if their current contents
1282		do not coincide with what's being written.  (That is, avoid messing
1283		with the time stamps unless absolutely necessary.)
1284
1285		2. Implement a "repository" mechanism for generated files related
1286		to "incomplete pointer types".   See the README file for details.
1287
1288		----------------------------------------------------------------------
1289		Name: Matthias Blume
1290		Date: 2002/02/14 11:50:00 EST
1291		Tag: blume-20020214-quote
1292		Description:
1293
1294		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1295		because of the new and improved tag generation scheme.  (Thanks to Allen
1296		Leung for pointing it out.)
1297
1298		----------------------------------------------------------------------
1299		Name: Lal George
1300		Date: 2002/02/14 09:55:27 EST 2002
1301		Tag: george-20020214-isabelle-bug
1302		Description:
1303
1304		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1305		of Isabelle on the x86.
1306
1307		From Allen:
1308		-----------
1309		I've found the problem:
1310
1311		in ra-core.sml, I use the counter "blocked" to keep track of the
1312		true number of elements in the freeze queue.  When the counter goes
1313		to zero, I skip examining the queue.  But I've messed up the
1314		bookkeeping in combine():
1315
1316		else ();
1317		case !ucol of
1318		PSEUDO => (if !cntv > 0 then
1319		(if !cntu > 0 then blocked := !blocked - 1 else ();
1320		^^^^^^^^^^^^^^^^^^^^^^^
1321		moveu := mergeMoveList(!movev, !moveu)
1322		)
1323		else ();
1324
1325		combine() is called to coalesce two nodes u and v.
1326		I think I was thinking that if the move counts of u and v are both
1327		greater than zero then after they are coalesced then one node is
1328		removed from the freeze queue.  Apparently I was thinking that
1329		both u and v are of low degree, but that's clearly not necessarily true.
1330
1331
1332		02/12/2002:
1333		Here's the patch.  HOL now compiles.
1334
1335		I don't know how this impact on performance (compile
1336		time or runtime).  This bug caused the RA (especially on the x86)
1337		to go thru the potential spill phase when there are still nodes on the
1338		freeze queue.
1339
1340
1341
1342
1343		----------------------------------------------------------------------
1344		Name: Matthias Blume
1345		Date: 2002/02/13 22:40:00 EST
1346		Tag: blume-20020213-fptr-rtti
1347		Description:
1348
1349		Fixed a bug in ml-nlffigen that was introduced with one of the previous
1350		updates.
1351
1352		----------------------------------------------------------------------
1353		Name: Matthias Blume
1354		Date: 2002/02/13 16:41:00 EST
1355		Tag: blume-20020213-cmlpq
1356		Description:
1357
1358		Added new priority queue export symbols (which have just been added to
1359		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1360		not compile and the installer would choke.)
1361
1362		----------------------------------------------------------------------
1363		Name: Matthias Blume
1364		Date: 2002/02/13 16:15:00 EST
1365		Tag: blume-20020213-various
1366		Description:
1367
1368		1. More tweaks to ml-nlffigen:
1369
1370		- better internal datastructures (resulting in slight speedup)
1371		- "-match" option requires exact match
1372		- "localized" gensym counters (untagged structs/unions nested within
1373		other structs/unions or within typedefs get a fresh counter; their
1374		tag will be prefixed by a concatenation of their parents' tags)
1375		- bug fixes (related to calculation of transitive closure of types
1376		to be included in the output)
1377
1378		2. Minor Basis updates:
1379
1380		- added implementations for List.collate and Option.app
1381
1382		----------------------------------------------------------------------
1383		Name: Matthias Blume
1384		Date: 2002/02/11 15:55:00 EST
1385		Tag: blume-20020211-gensym
1386		Description:
1387
1388		Added a "-gensym" option to command line of ml-nlffigen.  This can be
1389		used to specify a "stem" -- a string that is inserted in all "gensym'd"
1390		names (ML structure names that correspond to unnamed C structs, unions,
1391		and enums), so that separate runs of ml-nlffigen do not clash.
1392
1393		----------------------------------------------------------------------
1394		Name: Matthias Blume
1395		Date: 2002/02/11 12:05:00 EST
1396		Tag: blume-20020211-gensml
1397		Description:
1398
1399		A quick fix for a problem with GenSML (in the pgraph-util library):
1400		Make generation of toplevel "local" optional.  (Strictly speaking,
1401		signature definitions within "local" are not legal SML.)
1402
1403		Other than that: updates to INSTALL and cm/TODO.
1404
1405		----------------------------------------------------------------------
1406		Name: Matthias Blume
1407		Date: 2002/02/08 15:00:00 EST
1408		Tag: blume-20020208-uniquepid
1409		Description:
1410
1411		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1412
1413		1. The installer (config/install.sh) has gotten smarter:
1414
1415		- Configuration options are a bit easier to specify now
1416		(in config/targets).
1417		- Bug in recognizing .tar.bz2 files fixed.
1418		- Installer automatically resolves dependencies between
1419		configuration options (e.g., if you ask for eXene, you will
1420		also get cml -- regardless whether you asked for it or not).
1421		- Installer can run in "quieter mode" by setting the environment
1422		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1423		"completely silent", though.
1424		- Build HashCons library as part of smlnj-lib.
1425
1426		2. A new scheme for assigning persistent identifiers to compilation
1427		units (and, by extension, to types etc.) has been put into place.
1428		This fixes a long-standing bug where types and even dynamic values
1429		can get internally confused, thereby compromising type safety
1430		(abstraction) and dynamic correctness.  See
1431
1432		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1433
1434		for an example of how things could go wrong until now.
1435
1436		The downside of the new scheme is that pids are not quite as
1437		persistent as they used to be: CM will generate a fresh pid
1438		for every compilation unit that it thinks it sees for the first
1439		time.  That means that if you compile starting from a clean, fresh
1440		source tree at two different times, you end up with different
1441		binaries.
1442
1443		Cutoff recompilation, however, has not been compromised because
1444		CM keeps pid information in special caches between runs.
1445
1446		----------------------------------------------------------------------
1447		Name: Lal George
1448		Date: 2002/02/07 15:34:13 EST 2002
1449		Tag: <none>
1450		Description:
1451
1452		Compilers that generate assembly code may produce  global labels
1453		whose value is resolved at link time. The various peephole optimization
1454		modules did not take this in account.
1455
1456		TODO. The Labels.addrOf function should really return an option
1457		type so that clients are forced to deal with this issue, rather
1458		than an exception being raised.
1459
1460		----------------------------------------------------------------------
1461		Name: Lal George
1462		Date: 2002/02/06 13:55:02 EST
1463		Tag: george-20020206-ra-breakup
1464		Description:
1465
1466		1. A bug fix from Allen.
1467
1468		A typo causes extra fstp %st(0)'s to be generated at compensation
1469		edges, which might cause stack underflow traps at runtime.  This
1470		occurs in fft where there are extraneous fstps right before the 'into'
1471		trap instruction (in this case they are harmless since none of the
1472		integers overflow.)
1473
1474		2. Pulled out various utility modules that were embedded in the modules
1475		of the register allocator. I need these modules for other purposes, but
1476		they are not complete enough to put into a library (just yet).
1477		----------------------------------------------------------------------
1478		Name: Matthias Blume
1479		Date: 2002/01/31 16:05:00 EST
1480		Tag: blume-20020131-sparc-ccalls
1481		Description:
1482
1483		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1484		of extra stack space by mistake.  Fixed.
1485
1486		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1487
1488		----------------------------------------------------------------------
1489		Name: Allen Leung
1490		Date: 2002/01/30
1491		Tag: leunga-20020130-nowhere-bug-fix
1492		Description:
1493
1494		MLRISC bug fixes:
1495		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1496		2. MachineInt.fromString was negating its value.
1497
1498		----------------------------------------------------------------------
1499		Name: Matthias Blume
1500		Date: 2002/01/29
1501		Tag: blume-20020129-INSTALL
1502		Description:
1503
1504		- Added somewhat detailed installation instructions (file INSTALL).
1505		- Fixed curl-detection bug in config/install.sh.
1506		- It is now possible to select the URL getter using the URLGETTER
1507		environment variable:
1508
1509		not set / "unknown"      --> automatic detection (script tries wget,
1510		curl, and lynx)
1511		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
1512		how to properly invoke them)
1513		other                    --> use $URLGETTER directly, it must take
1514		precisely two command-line arguments
1515		(source URL and destination file name)
1516
1517		----------------------------------------------------------------------
1518		Name: Matthias Blume
1519		Date: 2002/01/28
1520		Tag: blume-20020128-sparc-ccalls
1521		Description:
1522
1523		- Fixed problem with calculation of "used" registers in sparc-c-calls.
1524		- Make use of the allocParam argument in sparc-c-calls.
1525
1526		----------------------------------------------------------------------
1527		Name: Matthias Blume
1528		Date: 2002/01/28
1529		Tag: blume-20020128-allocParam
1530		Description:
1531
1532		John Reppy:  Changes c-calls API to accept client-callback for
1533		allocating extra stack space.
1534		me: Corresponding changes to mlriscGen (using a dummy argument that
1535		does not change the current behavior).
1536
1537		----------------------------------------------------------------------
1538		Name: Matthias Blume
1539		Date: 2002/01/28 12:00:00
1540		Tag: Release_110_38
1541		Description:
1542
1543		This time for real!!!
1544
1545		----------------------------------------------------------------------
1546		Name: Matthias Blume
1547		Date: 2002/01/28 10:56:00 EST
1548		Tag: blume-20020128-retraction
1549		Description:
1550
1551		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1552		with blume-Release_110_38-retracted.)
1553
1554		1. Fixed a problem with incorrect rounding modes in real64.sml.
1555		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1556
1557		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1558		and enums fixed.  The naming of corresponding ML identifiers should
1559		now be consistent again.
1560
1561		----------------------------------------------------------------------
1562		Name: Allen Leung
1563		Date: 2002/01/27
1564		Tag: leunga-20020127-nowhere
1565		Description:
1566
1567		Added a target called nowhere in the configuration scripts.
1568		Enabling this will build the MLRISC 'nowhere' tool (for translating
1569		programs with where-clauses into legal SML code) during installation.
1570
1571		----------------------------------------------------------------------
1572		Name: Matthias Blume
1573		Date: 2002/01/25 21:27:00 EST
1574		Tag: blume-Release_110_38-retracted
1575		Description:
1576
1577		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1578
1579		README will be added later.
1580
1581		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1582		(Release_110_38) removed.  Reason: Last-minute bug fixes.
1583
1584		----------------------------------------------------------------------
1585		Name: Matthias Blume
1586		Date: 2002/01/25
1587		Tag: blume-20020125-ffi
1588		Description:
1589
1590		A large number of tweaks and improvements to ml-nlffi-lib and
1591		ml-nlffigen:
1592
1593		- ML represenation types have been streamlined
1594		- getter and setter functions work with concrete values, not abstract
1595		ones where possible
1596		- ml-nlffigen command line more flexible (see README file there)
1597		- some bugs have been fixed (hopefully)
1598
1599		----------------------------------------------------------------------
1600		Name: Lal George
1601		Date: 2002/01/24
1602		Tag: george-20020124-risc-ra-interface
1603		Description:
1604
1605		There is a dramatic simplification in the interface to the
1606		register allocator for RISC architectures as a result of making
1607		parallel copy instructions explicit.
1608
1609		----------------------------------------------------------------------
1610		Name: Matthias Blume
1611		Date: 2002/01/22
1612		Tag: blume-20020122-x86-ccalls
1613		Description:
1614
1615		Bug fix for c-calls on x86 (having to do with how char- and
1616		short-arguments are being handled).
1617
1618		----------------------------------------------------------------------
1619		Name: Matthias Blume
1620		Date: 2002/01/21
1621		Tag: blume-20020121-ff
1622		Description:
1623
1624		Another day of fiddling with the FFI...
1625
1626		1. Bug fix/workaround:  CKIT does not complain about negative array
1627		dimensions, so ml-nlffigen has to guard itself against this possibility.
1628		(Otherwise a negative dimension would send it into an infinite loop.)
1629
1630		2. Some of the abstract types (light objects, light pointers, most "base"
1631		types) in structure C are now eqtypes.
1632
1633		3. Added constructors and test functions for NULL function pointers.
1634
1635		----------------------------------------------------------------------
1636		Name: Matthias Blume
1637		Date: 2002/01/18
1638		Tag: blume-20020118-ready-for-new-release
1639		Description:
1640
1641		Made config/srcarchiveurl point to a new place.  (Will provide boot
1642		files shortly.)
1643
1644		Maybe we christen this to be 110.38?
1645
1646		----------------------------------------------------------------------
1647		Name: Matthias Blume
1648		Date: 2002/01/18
1649		Tag: blume-20020118-more-ffifiddle
1650		Description:
1651
1652		Today's FFI fiddling:
1653
1654		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
1655		C.Set.* that use concrete (MLRep.*) arguments and results instead
1656		of abstract ones.
1657
1658		- Provided word-style bit operations etc. for "int" representation
1659		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1660		and Long.
1661
1662		----------------------------------------------------------------------
1663		Name: Matthias Blume
1664		Date: 2002/01/18
1665		Tag: blume-20020118-use-x86-fp
1666		Description:
1667
1668		Now that x86-fast-fp seems to be working, I turned it back on again
1669		by default.  (Seems to work fine now, even with the FFI.)
1670
1671		Other than that, I added some documentation about the FFI to
1672		src/ml-nlffigen/README and updated the FFI test examples in
1673		src/ml-nlffi-lib/Tests/*.
1674
1675		----------------------------------------------------------------------
1676		Name: Allen Leung
1677		Date: 2002/01/17
1678		Tag: leunga-20020117-x86-fast-fp-call
1679		Description:
1680
1681		1. Fixed a problem with handling return fp values when x86's fast fp
1682		mode is turned on.
1683
1684		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1685		of %f32.
1686
1687		3. Added a constructor INT32lit to the ast of MLRISC tools.
1688
1689		----------------------------------------------------------------------
1690		Name: Matthias Blume
1691		Date: 2002/01/16
1692		Tag: blume-20020116-ffifiddle
1693		Description:
1694
1695		More fiddling with the FFI interface:
1696
1697		- Make constness 'c instead of rw wherever possible.  This eliminates
1698		the need for certain explicit coercions.  (However, due to ML's
1699		value polymorphism, there will still be many cases where explicit
1700		coercions are necessary.  Phantom types are not the whole answer
1701		to modeling a subtyping relationship in ML.)
1702
1703		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1704
1705		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1706		clumsy workaround for varargs problem.)
1707
1708		----------------------------------------------------------------------
1709		Name: Lal George
1710		Date: 2002/01/15
1711		Tag: <none>
1712		Description:
1713
1714		1. Since COPY instructions are no longer native to the architecture,
1715		a generic functor can be used to implement the expandCopies function.
1716
1717		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1718		TEXT segment.
1719
1720		----------------------------------------------------------------------
1721		Name: Matthias Blume
1722		Date: 2002/01/15
1723		Tag: blume-20020115-ffiupdates
1724		Description:
1725
1726		1. Fix for bug resulting in single-precision float values being returned
1727		incorrectly from FFI calls.
1728
1729		2. Small modifications to C FFI API:
1730
1731		- memory-allocation routines return straight objects (no options)
1732		and raise an exception in out-of-memory situations
1733		- unsafe extensions to cast between function pointers and pointers
1734		from/to ints
1735		- added structure C_Debug as an alternative to structure C where
1736		pointer-dereferencing (|*| and |*!) always check for null-pointers
1737		- added open_lib' to DynLinkage;  open_lib' works like open_lib
1738		but also takes a (possibly empty) list of existing library handles
1739		that the current library depends on
1740
1741		----------------------------------------------------------------------
1742		Name: Matthias Blume
1743		Date: 2002/01/10
1744		Tag: blume-20020110-newffigen
1745		Description:
1746
1747		1. Updates to portable graph code.
1748
1749		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1750		more scalable now so that even huge interfaces such as the one
1751		for GTK compile in finite time and space. :-)
1752		See src/ml-nlffigen/README for details on what's new.
1753
1754		----------------------------------------------------------------------
1755		Name: Lal George
1756		Date: 2001/01/09 14:31:35 EST 2002
1757		Tag: george-20011206-rm-native-copy
1758		Description:
1759
1760		Removed the native COPY and FCOPY instructions
1761		from all the architectures and replaced it with the
1762		explicit COPY instruction from the previous commit.
1763
1764		It is now possible to simplify many of the optimizations
1765		modules that manipulate copies. This has not been
1766		done in this change.
1767
1768		----------------------------------------------------------------------
1769		Name: Lal George
1770		Date: 2001/12/06 16:50:13 EST 2001
1771		Tag: george-20011206-mlrisc-instruction
1772		Description:
1773
1774		Changed the representation of instructions from being fully abstract
1775		to being partially concrete. That is to say:
1776
1777		from
1778		type instruction
1779
1780		to
1781		type instr                              (* machine instruction *)
1782
1783		datatype instruction =
1784		LIVE of {regs: C.cellset, spilled: C.cellset}
1785		| KILL of {regs: C.cellset, spilled: C.cellset}
1786		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1787		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1788		| INSTR of instr
1789
1790		This makes the handling of certain special instructions that appear on
1791		all architectures easier and uniform.
1792
1793		LIVE and KILL say that a list of registers are live or killed at the
1794		program point where they appear. No spill code is generated when an
1795		element of the 'regs' field is spilled, but the register is moved to
1796		the 'spilled' (which is present, more for debugging than anything else).
1797
1798		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1799		We used to generate:
1800
1801		DEFFREG f1
1802		f1 := f2 + f3
1803		trapb
1804
1805		but now generate:
1806
1807		f1 := f2 + f3
1808		trapb
1809		LIVE {regs=[f1,f2,f3], spilled=[]}
1810
1811		Furthermore, the DEFFREG (hack) required that all floating point instruction
1812		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1813		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1814		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1815		this architecture).
1816
1817		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1818		all the architectures. This will result in further simplification of the
1819		register allocator that must be aware of them for coalescing purposes, and
1820		will also simplify certain aspects of the machine description that provides
1821		callbacks related to parallel copies.
1822
1823		ANNOTATION should be obvious, and now INSTR represents the honest to God
1824		machine instruction set!
1825
1826		The <arch>/instructions/<arch>Instr.sml files define certain utility
1827		functions for making porting easier -- essentially converting upper case
1828		to lower case. All machine instructions (of type instr) are in upper case,
1829		and the lower case form generates an MLRISC instruction. For example on
1830		the alpha we have:
1831
1832		datatype instr =
1833		LDA of {r:cell, b:cell, d:operand}
1834		| ...
1835
1836		val lda : {r:cell, b:cell, d:operand} -> instruction
1837		...
1838
1839		where lda is just (INSTR o LDA), etc.
1840
1841		----------------------------------------------------------------------
1842		Name: Matthias Blume
1843		Date: 2001/11/22 21:40:00 EST
1844		Tag: Release_110_37
1845		Description:
1846
1847		Release 110.37.  This time for real.
1848
1849		----------------------------------------------------------------------
1850		Name: Matthias Blume
1851		Date: 2001/11/21 16:35:00 EST
1852		Tag: blume-20011121-foot-in-mouth
1853		Description:
1854
1855		Removed the "Release_110_37" tag because of a serious bug.
1856		This will be re-tagged once the bug is fixed.
1857
1858		----------------------------------------------------------------------
1859		Name: Matthias Blume
1860		Date: 2001/11/21 16:14:00 EST
1861		Tag: blume-20011121-forgottenfile
1862		Description:
1863
1864		Forgot to add a file.  (Just a .tex-file -- part of
1865		the CM manual source.)
1866
1867		----------------------------------------------------------------------
1868		Name: Matthias Blume
1869		Date: 2001/11/21 16:10:00 EST
1870		Tag: blume-20011121-invalid_110_37
1871		Description:
1872
1873		Note: I removed the original tag "Release_110_37" from this commit
1874		because we found a serious bug in all non-x86 backends.
1875		- Matthias
1876
1877		1. Modifications to the SML/NJ code generator and to the runtime system
1878		so that code object name strings are directly inserted into code
1879		objects at code generation time.  The only business the runtime system
1880		has with this is now to read the name strings on occasions.
1881		(The encoding of the name string has also changed somewhat.)
1882
1883		2. CM now implements a simple "set calculus" for specifying export lists.
1884		In particular, it is now possible to refer to the export lists of
1885		other libraries/groups/sources and form unions as well as differences.
1886		See the latest CM manual for details.
1887
1888		3. An separate notion of "proxy" libraries has again be eliminated from
1889		CM's model.  (Proxy libraries are now simply a special case of using
1890		the export list calculus.)
1891
1892		4. Some of the existing libraries now take advantage of the new set
1893		calculus.
1894		(Notice that not all libraries have been converted because some
1895		of the existing .cm-files are supposed to be backward compatible
1896		with 110.0.x.)
1897
1898		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1899		"General.exnMessage"!  The former relies on a certain hook to be
1900		initialized, and that often does not happen in the stand-alone case.)
1901
1902		----------------------------------------------------------------------
1903		Name: Lal George
1904		Date: 2001/11/21  13:56:18 EST
1905		Tag: george-2001121-pseudo-ops
1906		Description:
1907
1908		Implemented a complete redesign of MLRISC pseudo-ops. Now there
1909		ought to never be any question of incompatabilities with
1910		pseudo-op syntax expected by host assemblers.
1911
1912		For now, only modules supporting GAS syntax are implemented
1913		but more should follow, such as MASM, and vendor assembler
1914		syntax, e.g. IBM as, Sun as, etc.
1915
1916		----------------------------------------------------------------------
1917		Name: Matthias Blume
1918		Date: 2001/11/14 11:52:00 EST
1919		Tag: blume-20011114-srcname
1920		Description:
1921
1922		1. Routed the name of the current source file to mlriscgen where it
1923		should be directly emitted into the code object.  (This last part
1924		is yet to be done.)
1925
1926		2. Some cleanup of the pgraph code to make it match the proposal that
1927		I put out the other day.  (The proposal notwithstanding, things are
1928		still in flux here.)
1929
1930		----------------------------------------------------------------------
1931		Name: Lal George
1932		Date: 2001/11/14 09:44:04 EST
1933		Tag:
1934		Description:
1935
1936		Fix for a backpatching bug reported by Allen.
1937
1938		Because the boundary between short and long span-dependent
1939		instructions is +/- 128, there are an astounding number of
1940		span-dependent instructions whose size is over estimated.
1941
1942		Allen came up with the idea of letting the size of span
1943		dependent instructions be non-monotonic, for a maxIter
1944		number of times, after which the size must be monotonically
1945		increasing.
1946
1947		This table shows the number of span-dependent instructions
1948		whose size was over-estimated as a function of maxIter, for the
1949		file Parse/parse/ml.grm.sml:
1950
1951		maxIter            # of instructions:
1952		10                      687
1953		20                      438
1954		30                      198
1955		40                        0
1956
1957		In compiling the compiler, there is no significant difference in
1958		compilation speed between maxIter=10 and maxIter=40. Actually,
1959		my measurements showed that maxIter=40 was a tad faster than
1960		maxIter=10! Also 96% of the  files in the compiler reach a fix
1961		point within 13 iterations, so fixing maxIter at 40, while high,
1962		is okay.
1963
1964		----------------------------------------------------------------------
1965		Name: Matthias Blume
1966		Date: 2001/10/31 15:25:00 EST
1967		Tag: blume-20011031-pgraph
1968		Description:
1969
1970		CKIT:
1971		* Changed the "Function" constructor of type Ast.ctype to carry optional
1972		argument identifiers.
1973		* Changed the return type of TypeUtil.getFunction accordingly.
1974		* Type equality ignores the argument names.
1975		* TypeUtil.composite tries to preserve argument names but gives up quickly
1976		if there is a mismatch.
1977
1978		installation script:
1979		* attempts to use "curl" if available (unless "wget" is available as well)
1980
1981		CM:
1982		* has an experimental implementation of "portable graphs" which I will
1983		soon propose as an implementation-independent library format
1984		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1985
1986		NLFFI-LIB:
1987		* some cleanup (all cosmetic)
1988
1989		NLFFIGEN:
1990		* temporarily disabled the mechanism that suppresses ML output for
1991		C definitions whose identifiers start with an underscore character
1992		* generate val bindings for enum constants
1993		* user can request that only one style (light or heavy) is being used;
1994		default is to use both (command-line arguments: -heavy and -light)
1995		* fixed bug in handling of function types involving incomplete pointers
1996		* generate ML entry points that take record arguments (i.e., using
1997		named arguments) for C functions that have a prototype with named
1998		arguments
1999		(see changes to CKIT)
2000
2001		----------------------------------------------------------------------
2002		Name: Allen Leung
2003		Date: 2001/10/27 20:34:00 EDT
2004		Tag: leunga-20011027-x86-fast-fp-call
2005		Description:
2006
2007		Fixed the bug described in blume-20010920-slowfp.
2008
2009		The fix involves
2010		1. generating FCOPYs in FSTP in ia32-svid
2011		2. marking a CALL with the appropriate annotation
2012
2013		----------------------------------------------------------------------
2014		Name: Matthias Blume
2015		Date: 2001/10/16 11:32:00 EDT
2016		Tag: blume-20011016-netbsd
2017		Description:
2018
2019		Underscore patch from Chris Richards (fixing problem with compiling
2020		runtime system under recent NetBSD).
2021
2022		----------------------------------------------------------------------
2023		Name: Allen Leung
2024		Date: 2001/10/12 17:18:32 EDT 2001
2025		Tag: leung-20011012-x86-printflowgraph
2026		Description:
2027
2028		X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
2029
2030		----------------------------------------------------------------------
2031		Name: Lal George
2032		Date: 2001/10/11 23:51:34 EDT
2033		Tag: george-20011011-too-many-instrs
2034		Description:
2035
2036		The representation of a program point never expected to see more
2037		than 65536 instructions in a basic block!
2038
2039		----------------------------------------------------------------------
2040		Name: Lal George
2041		Date: 2001/10/09 09:41:37 EDT
2042		Tag: george-20011008-mlrisc-labels
2043		Description:
2044
2045		Changed the machine description files to support printing of
2046		local and global labels in assembly code, based on host assembler
2047		conventions.
2048
2049		----------------------------------------------------------------------
2050		Name: Matthias Blume
2051		Date: 2001/09/25 15:25:00 EDT
2052		Tag: blume-20010925-exninfo
2053		Description:
2054
2055		I provided a non-hook implementation of exnName (at the toplevel) and
2056		made the "dummy" implementation of exnMessage (at the toplevel) more
2057		useful: if nothing gets "hooked in", then at least you are going to
2058		see the exception name and a message indicating why you don't see more.
2059
2060		[For the time being, programs that need exnMessage and want to use
2061		ml-build should either use General.exnMessage (strongly recommended) or
2062		refer to structure General at some other point so that CM sees a
2063		static dependency.]
2064
2065		[Similar remarks go for "print" and "use":  If you want to use their
2066		functionality in stand-alone programs generated by ml-build, then use
2067		TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
2068
2069		----------------------------------------------------------------------
2070		Name: Matthias Blume
2071		Date: 2001/09/20 17:28:00 EDT
2072		Tag: blume-20010920-slowfp
2073		Description:
2074
2075		Allen says that x86-fast-fp is not safe yet, so I turned it off again...
2076
2077		----------------------------------------------------------------------
2078		Name: Matthias Blume
2079		Date: 2001/09/20 17:20:00 EDT
2080		Tag: blume-20010920-canonicalpaths
2081		Description:
2082
2083		0. Updated the BOOT file (something that I forgot to do earlier).
2084
2085		1. Small internal change to CM so that it avoids "/../" in filenames
2086		as much as possible (but only where it is safe).
2087
2088		2. Changed config/_run-sml (resulting in a changed bin/.run-sml) so
2089		that arguments that contain delimiters are passed through correctly.
2090		This change also means that all "special" arguments of the form
2091		@SMLxxx... must come first.
2092
2093		3. Changed install script to put relative anchor names for tool commands
2094		into pathconfig.
2095
2096		----------------------------------------------------------------------
2097		Name: Matthias Blume
2098		Date: 2001/09/18 15:35:00 EDT
2099		Tag: blume-20010918-readme11036
2100		Description:
2101
2102		Added README files.
2103
2104		----------------------------------------------------------------------
2105		Name: Matthias Blume
2106		Date: 2001/09/18 11:45:00 EDT
2107		Tag: Release_110_36 (retag)
2108		Description:
2109
2110		Fixed mistake in config/preloads. Retagged as 110.36.
2111
2112		----------------------------------------------------------------------
2113		Name: Matthias Blume
2114		Date: 2001/09/18 09:40:00 EDT
2115		Tag: Release_110_36_orig (tag changed)
2116		Description:
2117
2118		New version (110.36).  New bootfiles.
2119
2120		----------------------------------------------------------------------
2121		Name: Matthias Blume
2122		Date: 2001/09/14 16:15:00 EDT
2123		Tag: blume-20010914-x86fastfp
2124		Description:
2125
2126		John committed some changes that Allen made, in particular a (hopefully)
2127		correctly working version of the x86-fp module.
2128
2129		I changed the default setting of the Control.MLRISC.getFlag "x86-fast-fp"
2130		flag to "true".  Everything seems to compile to a fixpoint ok, and
2131		"mandelbrot" speeds up by about 15%.
2132
2133		----------------------------------------------------------------------
2134		Name: Matthias Blume
2135		Date: 2001/09/13 11:20:00 EDT
2136		Tag: blume-20010913-minimal
2137		Description:
2138
2139		1. Stefan Monnier's patch to fix a miscompilation problem that
2140		was brought to light by John Reppy's work on Moby.
2141
2142		2. Implemented a minimal "structure Compiler" that contains just
2143		"version" and "architecture".  The minimal version will be
2144		available when the full version is not.  This is for backward-
2145		compatibility with code that wants to test Compiler.version.
2146
2147		----------------------------------------------------------------------
2148		Name: Matthias Blume
2149		Date: 2001/08/28 14:03:00 EDT
2150		Tag: blume-20010828-ml-lex
2151		Description:
2152
2153		Fix for bug 1581, received from Neophytos Michael.
2154
2155		----------------------------------------------------------------------
2156		Name: Matthias Blume
2157		Date: 2001/08/27 11:20:00 EDT
2158		Tag: blume-20010827-readme11035
2159		Description:
2160
2161		Fleshed out the README file for 110.35.
2162
2163		----------------------------------------------------------------------
2164		Name: Matthias Blume
2165		Date: 2001/08/24 17:10:00 EDT
2166		Tag: Release_110_35
2167		Description:
2168
2169		New version number (110.35).  New bootfiles.
2170
2171		----------------------------------------------------------------------
2172		Name: Lal George
2173		Date: 2001/08/24 13:47:18 EDT 2001
2174		Tag: george-20010824-MLRISC-graphs
2175		Description:
2176
2177		removed clusters from MLRISC completely and replaced with graphs.
2178
2179		----------------------------------------------------------------------
2180		Name: Matthias Blume
2181		Date: 2001/08/23 17:50:00 EDT
2182		Tag: blume-20010823-toplevel
2183		Description:
2184
2185		- some reorganization of the code that implements various kinds of
2186		environments in the compiler (static, dynamic, symbolic, combined)
2187		- re-implemented the EnvRef module so that evalStream works properly
2188		(if the stream contains references to "use", "CM.make", etc.)
2189		- cleaned up evalloop.sml and interact.sml (but they need more cleaning)
2190
2191		----------------------------------------------------------------------
2192		Name: Matthias Blume
2193		Date: 2001/08/20 15:50 EDT
2194		Tag: blume20010820-slipup
2195		Description:
2196
2197		I forgot to commit a few files.  Here they are...
2198
2199		----------------------------------------------------------------------
2200		Name: Matthias Blume
2201		Date: 2001/08/20 15:35:00 EDT
2202		Tag: blume-20010820-debugprof
2203		Description:
2204
2205		!!!! NEW BOOTFILES !!!!
2206
2207		This is another round of reorganizing the compiler sources.  This
2208		time the main goal was to factor out all the "instrumentation"
2209		passes (for profiling and backtracing) into their own library.
2210		The difficulty was to do it in such a way that it does not depend
2211		on elaborate.cm but only on elabdata.cm.
2212
2213		Therefore there have been further changes to both elaborate.cm and
2214		elabdata.cm -- more "generic" things have been moved from the former
2215		to the latter.  As a result, I was forced to split the assignment
2216		of numbers indicating "primtyc"s into two portions: SML-generic and
2217		SML/NJ-specific.  Since it would have been awkward to maintain,
2218		I bit the bullet and actually _changed_ the mapping between these
2219		numbers and primtycs.  The bottom line of this is that you need
2220		a new set of bin- and bootfiles.
2221
2222		I have built new bootfiles for all architectures, so doing a fresh
2223		checkout and config/install.sh should be all you need.
2224
2225		The newly created library's name is
2226
2227		$smlnj/viscomp/debugprof.cm
2228
2229		and its sources live under
2230
2231		src/compiler/DebugProf
2232
2233		----------------------------------------------------------------------
2234		Name: Matthias Blume
2235		Date: 2001/08/15 17:15:00 EDT
2236		Tag: blume-20010815-compreorg
2237		Description:
2238
2239		This is a first cut at reorganizing the CM libraries that make up the
2240		core of the compiler.  The idea is to separate out pieces that could
2241		be used independently by tools, e.g., the parser, the typechecker, etc.
2242
2243		The current status is a step in this direction, but it is not quite
2244		satisfactory yet.  Expect more changes in the future.
2245
2246		Here is the current (new) organization...
2247
2248		What used to be $smlnj/viscomp/core.cm is now divided into
2249		six CM libraries:
2250
2251		$smlnj/viscomp/basics.cm
2252		/parser.cm
2253		/elabdata.cm
2254		/elaborate.cm
2255		/execute.cm
2256		/core.cm
2257
2258		The CM files for these libraries live under src/system/smlnj/viscomp.
2259		All these libraries are proxy libraries that contain precisely
2260		one CM library component.  Here are the locations of the components
2261		(all within the src/compiler tree):
2262
2263		Basics/basics.cm
2264		Parse/parser.cm
2265		ElabData/elabdata.cm
2266		Elaborator/elaborate.cm
2267		Execution/execute.cm
2268		core.cm
2269
2270		[This organization is the same that has been used already
2271		for a while for the architecture-specific parts of the visible
2272		compiler and for the old version of core.cm.]
2273
2274		As you will notice, many source files have been moved from their
2275		respective original locations to a new home in one of the above
2276		subtrees.
2277
2278		The division of labor between the new libraries is the following:
2279
2280		basics.cm:
2281		- Simple, basic definitions that pertain to many (or all) of
2282		the other libraries.
2283		parser.cm:
2284		- The SML parser, producing output of type Ast.dec.
2285		- The type family for Ast is also defined and exported here.
2286		elabdata.cm:
2287		- The datatypes that describe input and output of the elaborator.
2288		This includes types, absyn, and static environments.
2289		elaborator.cm:
2290		- The SML/NJ type checker and elaborator.
2291		This maps an Ast.dec (with a given static environment) to
2292		an Absyn.dec (with a new static environment).
2293		- This libraries implements certain modules that used to be
2294		structures as functors (to remove dependencies on FLINT).
2295		execute.cm:
2296		- Everything having to do with executing binary code objects.
2297		- Dynamic environments.
2298		core.cm:
2299		- SML/NJ-specific instantiations of the elaborator and MLRISC.
2300		- Top-level modules.
2301		- FLINT (this should eventually become its own library)
2302
2303		Notes:
2304
2305		I am not 100% happy with the way I separated the elaborator (and its
2306		data structures) from FLINT.  Two instances of the same problem:
2307
2308		1. Data structures contain certain fields that carry FLINT-specific
2309		information.  I hacked around this using exn and the property list
2310		module from smlnj-lib.  But the fact that there are middle-end
2311		specific fields around at all is a bit annoying.
2312
2313		2. The elaborator calculates certain FLINT-related information.  I tried
2314		to make this as abstract as I could using functorization, but, again,
2315		the fact that the elaborator has to perform calculations on behalf
2316		of the middle-end at all is not nice.
2317
2318		3. Having to used exn and property lists is unfortunate because it
2319		weakens type checking.  The other alternative (parameterizing
2320		nearly *everything*) is not appealing, though.
2321
2322		I removed the "rebinding =" warning hack because due to the new organization
2323		it was awkward to maintain it.  As a result, the compiler now issues some of
2324		these warnings when compiling init.cmi during bootstrap compilation. On
2325		the plus side, you also get a warning when you do, for example:
2326		val op = = Int32.+
2327		which was not the case up to now.
2328
2329		I placed "assign" and "deref" into the _Core structure so that the
2330		code that deals with the "lazy" keyword can find them there.  This
2331		removes the need for having access to the primitive environment
2332		during elaboration.
2333
2334		----------------------------------------------------------------------
2335		Name: Matthias Blume
2336		Date: 2001/08/13
2337		Tag: blume-20010813-closures
2338		Description:
2339
2340		This fix was sent to us by Zhong Shao.  It is supposed to improve the
2341		performance of certain loops by avoiding needless closure allocation.
2342
2343		----------------------------------------------------------------------
2344		Name: Lal George
2345		Date: 2001/07/31 10:03:23 EDT 2001
2346		Tag: george-20010731-x86-fmalloc
2347		Description: Fixed bug in x86 calls
2348
2349		There was a bug where call instructions would mysteriously
2350		vanish. The call instruction had to be one that returned
2351		a floating point value.
2352
2353		----------------------------------------------------------------------
2354		Name: Lal George
2355		Date: 2001/07/19 16:36:29 EDT 2001
2356		Tag: george-20010719-simple-cells
2357		Description:
2358
2359		I have dramatically simplified the interface for CELLS in MLRISC.
2360
2361		In summary, the cells interface is broken up into three parts:
2362
2363		1. CellsBasis : CELLS_BASIS
2364
2365		CellsBasis is a top level structure and common for all
2366		architectures.  it contains the definitions of basic datatypes
2367		and utility  functions over these types.
2368
2369		2. functor Cells() : CELLS
2370
2371		Cells generates an interface for CELLS that incorporates the
2372		specific resources on the target architecture, such as the
2373		presence of special register classes, their number and size,
2374		and various useful substructures.
2375
2376		3. <ARCH>CELLS
2377
2378		e.g. SparcCells: SPARCCELLS
2379
2380		<ARCH>CELLS usually contains additional bindings for special
2381		registers  on the architecture, such as:
2382
2383		val r0 : cell           (* register zero *)
2384		val y : cell            (* Y register *)
2385		val psr : cell          (* processor status register *)
2386		...
2387
2388		The structure returned by applying the Cells functor is opened
2389		in this interface.
2390
2391		The main implication of all this is that the datatypes for cells is
2392		split between CellsBasis and CELLS -- a fairly simple change for user
2393		code.
2394
2395		In the old scheme the CELLS interface had a definitional binding of
2396		the form:
2397
2398		signature CELLS = sig
2399
2400		structure CellsBasis = CellsBasis
2401
2402		...
2403
2404		end
2405
2406		With all the sharing constraints that goes on in MLRISC, this old
2407		design  quickly leads to errors such as:
2408
2409		"structure definition spec inside of sharing ... "
2410
2411
2412		and appears to require an unacceptable amount of sharing and where
2413		constraint hackery.
2414
2415		I think this error message (the interaction of definitional specs and
2416		sharing) requires more explanation on our web page.
2417
2418		----------------------------------------------------------------------
2419		Name: Matthias Blume
2420		Date: 2001/07/19 15:00:00 EDT
2421		Tag: blume-20010719-libreorg
2422		Description:
2423
2424		This update puts together a fairly extensive but straightforward change
2425		to the way the libraries that implement the interactive system are
2426		organized:
2427
2428		The biggest change is the elimination of structure Compiler.  As a
2429		replacement for this structure, there is now a CM library
2430		(known as $smlnj/compiler.cm or $smlnj/compiler/current.cm)
2431		that exports all the substructures of the original structure Compiler
2432		directly.  So instead of saying Compiler.Foo.bar one now simply
2433		says Foo.bar.  (The CM libraries actually export a collection of
2434		structures that is richer than the collection of substructures of
2435		structure Compiler.)
2436
2437		To make the transition smooth, there is a separate library called
2438		$smlnj/compiler/compiler.cm which puts together and exports the
2439		original structure Compiler (or at least something very close to it).
2440
2441		There are five members of the original structure Compiler
2442		that are not exported directly but which instead became members
2443		of a new structure Backend (described by signature BACKEND).  These are:
2444		structure Profile (: PROFILE), structure Compile (: COMPILE), structure
2445		Interact (: INTERACT), structure Machine (: MACHINE), and val
2446		architecture (: string).
2447
2448		Structure Compiler.Version has become structure CompilerVersion.
2449
2450		Cross-compilers for alpha32, hppa, ppc, sparc, and x86 are provided
2451		by $smlnj/compiler/<arch>.cm where <arch> is alpha32, hppa, ppc, sparc,
2452		or x86, respectively.
2453		Each of these exports the same frontend structures that
2454		$smlnj/compiler.cm exports.  But they do not have a structure Backend
2455		and instead export some structure <Arch>Backend where <Arch> is Alpha32,
2456		Hppa, PPC, Sparc, or X86, respectively.
2457
2458		Library $smlnj/compiler/all.cm exports the union of the exports of
2459		$smlnj/compiler/<arch>.cm
2460
2461		There are no structures <Arch>Compiler anymore, use
2462		$smlnj/compiler/<arch>.cm instead.
2463
2464		Library host-compiler-0.cm is gone.  Instead, the internal library
2465		that instantiates CM is now called cm0.cm.  Selection of the host
2466		compiler (backend) is no longer done here but. (Responsibility for it
2467		now lies with $smlnj/compiler/current.cm.  This seems to be more
2468		logical.)
2469
2470		Many individual files have been moved or renamed.  Some files have
2471		been split into multiple files, and some "dead" files have been deleted.
2472
2473		Aside from these changes to library organization, there are also changes
2474		to the way the code itself is organized:
2475
2476		Structure Binfile has been re-implemented in such a way that it no
2477		longer needs any knowledge of the compiler.  It exclusively deals
2478		with the details of binfile layout.  It no longer invokes the
2479		compiler (for the purpose of creating new prospective binfile
2480		content), and it no longer has any knowledge of how to interpret
2481		pickles.
2482
2483		Structure Compile (: COMPILE) has been stripped down to the bare
2484		essentials of compilation.  It no longer deals with linking/execution.
2485		The interface has been cleaned up considerably.
2486
2487		Utility routines for dealing with linking and execution have been
2488		moved into their own substructures.
2489
2490		(The ultimate goal of these changes is to provide a light-weight
2491		binfile loader/linker (at least for, e.g., stable libraries) that
2492		does not require CM or the compiler to be present.)
2493
2494		CM documentation has been updated to reflect the changes to library
2495		organization.
2496
2497		----------------------------------------------------------------------
2498		Name: Matthias Blume
2499		Date: 2001/07/10 17:30:00 EDT
2500		Tag: Release_110_34
2501		Description:
2502
2503		Minor tweak to 110.34 (re-tagged):
2504
2505		- README.html file added to CVS repository
2506		- runtime compiles properly under FreeBSD 3.X and 4.X
2507
2508		----------------------------------------------------------------------
2509		Name: Matthias Blume
2510		Date: 2001/07/10 17:30:00 EDT
2511		Tag: Release_110_34
2512		Description:
2513
2514		New version number (110.34). New bootfiles.
2515
2516		----------------------------------------------------------------------
2517		Name: Matthias Blume
2518		Date: 2001/07/09 16:00:00 EDT
2519		Tag: blume-20010709-more-varargs
2520		Description:
2521
2522		I changed the handling of varargs in ml-nlffigen again:
2523		The ellipsis ... will now simply be ignored (with an accompanying warning).
2524
2525		The immediate effect is that you can actually call a varargs function
2526		from ML -- but you can't actually supply any arguments beyond the ones
2527		specified explicitly.  (For example, you can call printf with its format
2528		string, but you cannot pass additional arguments.)
2529
2530		This behavior is only marginally more useful than the one before, but
2531		it has the advantage that a function or, more importantly, a function
2532		type never gets dropped on the floor, thus avoiding follow-up problems with
2533		other types that refer to the offending one.
2534
2535		----------------------------------------------------------------------
2536		Name: Matthias Blume
2537		Date: 2001/07/09 11:25:00 EDT
2538		Tag: blume-20010709-varargs
2539		Description:
2540
2541		1. ckit-lib.cm now exports structure Error
2542		2. ml-nlffigen reports occurences of "..." (i.e., varargs function types)
2543		with a warning accompanied by a source location.  Moreover, it
2544		merely skips the offending function or type and proceeds with the
2545		rest of its work.u  As a result, one can safely feed C code containing
2546		"..." to ml-nlffigen.
2547		3. There are some internal improvements to CM, providing slightly
2548		more general string substitutions in the tools subsystem.
2549
2550		----------------------------------------------------------------------
2551		Name: Matthias Blume
2552		Date: 2001/06/27 15:10:00 EDT
2553		Tag: blume-20010627-concur
2554		Description:
2555
2556		Fixed a small bug in CM's handling of parallel compilation.
2557		(You could observe the bug by Control-C-interrupting an ordinary
2558		CMB.make or CM.stabilize and then attaching some compile servers.
2559		The result was that all of a sudden the previously interrupted
2560		compilation would continue on its own.  This was because of
2561		an over-optimization: CM did not bother to clean out certain queues
2562		when no servers were attached "anyway", resulting in the contents
2563		of these queues to grab control when new servers did get attached.)
2564
2565		There is also another minor update to the CM manual.
2566
2567		----------------------------------------------------------------------
2568		Name: Matthias Blume
2569		Date: 2001/06/26 16:15:00 EDT
2570		Tag: blume-20010626-cmdoc
2571		Description:
2572
2573		Minor typo fixed in CM manual (syntax diagram for libraries).
2574
2575		----------------------------------------------------------------------
2576		Name: Matthias Blume
2577		Date: 2001/06/25 22:55:00 EDT
2578		Tag: blume-20010625-x86pc
2579		Description:
2580
2581		Fixed a nasty bug in the X86 assembly code that caused signal
2582		handlers to fail (crash) randomly.
2583
2584		----------------------------------------------------------------------
2585		Name: Matthias Blume
2586		Date: 2001/06/25 12:05:00 EDT
2587		Tag: blume-20010625-nlffigen
2588		Description:
2589
2590		This update fixes a number of minor bugs in ml-nlffigen as reported by
2591		Nick Carter <nbc@andrew.cmu.edu>.
2592
2593		1. Silly but ok typedefs of the form "typedef void myvoid;" are now accepted.
2594		2. Default names for generated files are now derived from the name of
2595		the C file *without its directory*.  In particular, this causes generated
2596		files to be placed locally even if the C file is in some system directory.
2597		3. Default names for generated signatures and structures are also derived
2598		from the C file name without its directory.  This avoids silly things
2599		like "structure GL/GL".
2600		(Other silly names are still possible because ml-nlffigen does not do
2601		a thorough check of whether generated names are legal ML identifiers.
2602		When in doubt, use command line arguments to force particular names.)
2603
2604		----------------------------------------------------------------------
2605		Name: Matthias Blume
2606		Date: 2001/06/21 12:25:00 EDT
2607		Tag: blume-20010621-eXene
2608		Description:
2609
2610		eXene now compiles and (sort of) works again.
2611
2612		The library name (for version > 110.33) is $/eXene.cm.
2613
2614		I also added an new example in src/eXene/examples/nbody.  See the
2615		README file there for details.
2616
2617		----------------------------------------------------------------------
2618		Name: Matthias Blume
2619		Date: 2001/06/20 16:40:00 EDT
2620		Tag: blume-20010620-cml
2621		Description:
2622
2623		CML now compiles and works again.
2624
2625		Libraries (for version > 110.33):
2626
2627		$cml/cml.cm            Main CML library.
2628		$cml/basis.cm          CML's version of $/basis.cm.
2629		$cml/cml-internal.cm   Internal helper library.
2630		$cml/core-cml.cm       Internal helper library.
2631		$cml-lib/trace-cml.cm  Tracing facility.
2632		$cml-lib/smlnj-lib.cm  CML's version of $/smlnj-lib.cm
2633
2634		The installer (config/install.sh) has been taught how to properly
2635		install this stuff.
2636
2637		----------------------------------------------------------------------
2638		Name: Matthias Blume
2639		Date: 2001/06/19 17:55:00 EDT
2640		Tag: blume-20010619-instantiate
2641		Description:
2642
2643		This un-breaks the fix for bug 1432.
2644		(The bug was originally fixed in 110.9 but I broke it again some
2645		time after that.)
2646
2647		----------------------------------------------------------------------
2648		Name: Matthias Blume
2649		Date: 2001/06/19 17:25:00 EDT
2650		Tag: blume-20010619-signals
2651		Description:
2652
2653		This should (hopefully) fix the long-standing signal handling bug.
2654		(The runtime system was constructing a continuation record with an
2655		incorrect descriptor which would cause the GC to drop data on the floor...)
2656
2657		----------------------------------------------------------------------
2658		Name: Matthias Blume
2659		Date: 2001/06/15 15:05:00 EDT
2660		Tag: blume-20010615-moresparc
2661		Description:
2662
2663		Here is a short late-hour update related to Sparc c-calls:
2664
2665		-- made handling of double-word arguments a bit smarter
2666
2667		-- instruction selection phase tries to collapse certain clumsily
2668		constructed ML-Trees; typical example:
2669
2670		ADD(ty,ADD(_,e,LI d1),LI d2)  ->  ADD(ty,e,LI(d1+d2))
2671
2672		This currently has no further impact on SML/NJ since mlriscGen does
2673		not seem to generate such patterns in the first place, and c-calls
2674		(which did generate them in the beginning) has meanwhile been fixed
2675		so as to avoid them as well.
2676
2677		----------------------------------------------------------------------
2678		Name: Matthias Blume
2679		Date: 2001/06/15 15:05:00 EDT
2680		Tag: blume-20010615-sparc
2681		Description:
2682
2683		The purpose of this update is to provide an implementation of NLFFI
2684		on Sparc machines.
2685
2686		Here are the changes in detail:
2687
2688		* src/MLRISC/sparc/c-calls/sparc-c-calls.sml is a new file containing
2689		the Sparc implementation of the c-calls API.
2690		* The Sparc backend of SML/NJ has been modified to uniformely use %fp
2691		for accessing the ML frame.  Thus, we have a real frame pointer and
2692		can freely modify %sp without need for an omit-frame-ptr phase.
2693		The vfp logic in src/compiler/CodeGen/* has been changed to accomodate
2694		this case.
2695		* ml-nlffigen has been taught to produce code for different architectures
2696		and calling conventions.
2697		* In a way similar to what was done in the x86 case, the Sparc
2698		backend uses its own specific extension to mltree.  (For example,
2699		it needs to be able to generate UNIMP instructions which are part
2700		of the calling convention.)
2701		* ml-nlffi-lib was reorganized to make it more modular (in particular,
2702		to make it easier to plug in new machine- and os-dependent parts).
2703
2704		There are some other fairly unrelated bug fixes and cleanups as well:
2705
2706		* I further hacked the .cm files for MLRISC tools (like MDLGen) so
2707		that they properly share their libraries with existing SML/NJ libraries.
2708		* I fixed a minor cosmetic bug in CM, supressing certain spurious
2709		follow-up error messages.
2710		* Updates to CM/CMB documentation.
2711
2712		TODO items:
2713
2714		* MLRISC should use a different register as its asmTemp on the Sparc.
2715		(The current %o2 is a really bad choice because it is part of the
2716		calling conventions, so things might interfere in unexpected ways.)
2717
2718		----------------------------------------------------------------------
2719		Name: Matthias Blume
2720		Date: 2001/06/07
2721		Tag: blume-20010607-calls
2722		Description:
2723
2724		A number of internal changes related to C calls and calling conventions:
2725
2726		1. ML-Tree CALL statements now carry a "pops" field.  It indicates the
2727		number of bytes popped implicitly (by the callee).  In most cases
2728		this field is 0 but on x86/win32 it is some non-zero value.  This
2729		is information provided for the benefit of the "omit-frameptr" pass.
2730		2. The CALL instruction on the x86 carries a similar "pops" field.
2731		The instruction selection phase copies its value from the ML-Tree
2732		CALL statement.
2733		3. On all other architectures, the instruction selection phase checks
2734		whether "pops=0" and complains if not.
2735		4. The c-calls implementation for x86 now accepts two calling conventions:
2736		"ccall" and "stdcall".  When "ccall" is selected, the caller cleans
2737		up after the call and pops is set to 0.  For "stdcall", the caller
2738		does nothing, leaving the cleanup to the callee; pops is set to
2739		the number of bytes that were pushed onto the stack.
2740		5. The cproto decoder (compiler/Semant/types/cproto.sml) now can
2741		distinguish between "ccall" and "stdcall".
2742		6. The UNIMP instruction has been added to the supported Sparc instruction
2743		set. (This is needed for implementing the official C calling convention
2744		on this architecture.)
2745		7. I fixed some of the .cm files under src/MLRISC/Tools to make them
2746		work with the latest CM.
2747
2748		----------------------------------------------------------------------
2749		Name: Matthias Blume
2750		Date: 2001/06/05 15:10:00 EDT
2751		Tag: blume-20010605-cm-index
2752		Description:
2753
2754		0. The "lambdasplit" parameter for class "sml" in CM has been documented.
2755
2756		1. CM can now generate "index files".  These are human-readable files
2757		that list on a per-.cm-file basis each toplevel symbol defined or
2758		imported.  The location of the index file for
2759		<p>/<d>.cm is <p>/CM/INDEX/<d>.cm.
2760		To enable index-file generation, set CM.Control.generate_index to true
2761		or export an environment-symbol: export CM_GENERATE_INDEX=true.
2762
2763		The CM manual has been updated accordingly.
2764
2765		2. I made some slight modifications to the c-calls API in MLRISC.
2766
2767		a) There is now a callback to support saving/restoring of
2768		dedicated but caller-save registers around the actual call
2769		instruction.
2770		b) One can optionally specify a comment-annotation for the
2771		call instruction.
2772
2773		3. SML/NJ (mlriscGen.sml) uses this new API for the rawccall primop.
2774		(For example, the comment annotation shows the C prototype of
2775		the function being called.)
2776
2777		----------------------------------------------------------------------
2778		Name: Matthias Blume
2779		Date: 2001/06/01 13:30:00 EDT
2780		Tag: blume-20010601-nlffi-cleanup
2781		Description:
2782
2783		This is mostly a cleanup of MLFFI stuff:
2784
2785		- some signature files have been put into a more exposed place
2786		- the ugly 'f type parameter is gone (simplifies types tremendously!)
2787		- ml-nlffigen changed accordingly
2788		- tutorial updated
2789
2790		Other changes:
2791
2792		- author's affiliation in CM manual(s) updated
2793		- some more recognized keywords added to Allen's sml.sty
2794
2795		----------------------------------------------------------------------
2796		Name: Matthias Blume
2797		Date: 2001/05/25 15:30:00 EDT
2798		Tag: blume-20010525-iptr
2799		Description:
2800
2801		- put the official 110.33-README (as it appears on the ftp server) under
2802		CVS
2803		- fixed a small bug related to incomplete pointer types in
2804		ml-nlffigen
2805		- small cosmetic change to the ml-nlffi-lib's "arr" type constructor
2806		(it does not need the 'f type parameter)
2807
2808		----------------------------------------------------------------------
2809		Name: Matthias Blume
2810		Date: 2001/05/23 14:30:00 EDT
2811		Tag: Release_110_33
2812		Description:
2813
2814		New version number (110.33).  New bootfiles.
2815
2816		----------------------------------------------------------------------
2817		Name: Matthias Blume
2818		Date: 2001/05/22 18:06:00 EDT
2819		Tag: blume-20010522-targets
2820		Description:
2821
2822		Made install.sh use file config/targets.customized if it exists, falling
2823		back to config/targets if it doesn't.  This way one can have a customized
2824		version of the targets file without touching the "real thing", thus
2825		eliminating the constant fear of accidentally checking something bogus
2826		back into the CVS repository...  (File config/targets.customized must
2827		not be added to the repository!)
2828
2829		----------------------------------------------------------------------
2830		Name: Matthias Blume
2831		Date: 2001/05/22 16:30:00 EDT
2832		Tag: blume-20010522-minitut
2833		Description:
2834
2835		1. Bug fix in ml-nlffigen; now (hopefully) correctly handling
2836		struct returns.
2837		2. Added src/ml-nlffi-lib/Doc/mini-tutorial.txt.  This is some very
2838		incomplete, preliminary documentation for NLFFI.
2839
2840		----------------------------------------------------------------------
2841		Name: Matthias Blume
2842		Date: 2001/05/14 11:30:00 EDT
2843		Tag: blume-20010514-script
2844		Description:
2845
2846		Some bugs in install script fixed.
2847
2848		In addition to that I also made a slight change to the NLFFI API:
2849		Functors generated by ml-nlffigen now take the dynamic library as a
2850		straight functor argument, not as a suspended one.  (The original
2851		functor code used to force the suspension right away anyway, so there
2852		was nothing gained by this complication of the interface.)
2853
2854		----------------------------------------------------------------------
2855		Name: Matthias Blume
2856		Date: 2001/05/11 14:35:00 EDT
2857		Tag: blume-20010511-ml-nlffi
2858		Description:
2859
2860		I finally took the plunge and added my new FFI code to the main
2861		repository.  For x86-linux it is now ready for prime-time.
2862
2863		There are two new subdirectories of "src":
2864
2865		- ml-nlffi-lib:
2866		The utility library for programs using the FFI interface.
2867		Here is the implementation of $/c.cm and its associated low-level
2868		partners $/c-int.cm and $/memory.cm.
2869		- ml-nlffigen:
2870		A stand-alone program for generating ML glue code from C source
2871		code.
2872
2873		Building ml-nlffigen requires $/ckit-lib.cm.
2874
2875		The config/install.sh script has been updates to do the Right Thing
2876		(hopefully).
2877
2878		Notice that the source tree for the C-Kit will not be put under "src"
2879		but directly under the installation root directory.  (This is the
2880		structure that currently exists on the CVS server when you check out
2881		module "sml".)  Fortunately, config/install.sh knows about this oddity.
2882
2883		Bugs: No documentation yet.
2884
2885		----------------------------------------------------------------------
2886		Name: Matthias Blume
2887		Date: 2001/05/09 16:35:00 EDT
2888		Tag: blume-20010509-cpscontract
2889		Description:
2890
2891		Fixed a bug in the accounting code in cpsopt/contract.sml.  (The
2892		wrapper/unwrapper elimination did not decrement usage counts and some
2893		dead variables got overlooked by the dead-up logic.)
2894
2895		----------------------------------------------------------------------
2896		Name: Lal George
2897		Date: 2001/05/08  17:26:09 EDT
2898		Tag: george-20010508-omit-frameptr
2899		Description:
2900
2901		Changes to implement the omit-frame-pointer optimization to support
2902		raw C calls. For now, there is only support on the Intel x86, but
2903		other architectures will follow as more experience is gained with this.
2904
2905
2906		----------------------------------------------------------------------
2907		Name: Matthias Blume
2908		Date: 2001/05/07 14:40:00 EDT
2909		Tag: blume-20010507-proxies
2910		Description:
2911
2912		I made into "proxy libraries" all libraries that qualify for such a
2913		change.  (A qualifying library is a library that has another library or
2914		groups as its sole member and repeats that member's export list
2915		verbatim.  A proxy library avoids this repetition by omitting its export
2916		list, effectively inheriting the list that its (only) member exports.
2917		See the CM manual for more explanation.)
2918		The main effect is that explicit export lists for these libraries
2919		do not have to be kepts in sync, making maintenance a bit easier.
2920
2921		I also added copyright notices to many .cm-files.
2922
2923		Last but not least, I made a new set of bootfiles.
2924
2925		----------------------------------------------------------------------
2926		Name: Matthias Blume
2927		Date: 2001/05/04 17:00:00 EDT
2928		Tag: blume-20010504-cm-lsplit
2929		Description:
2930
2931		0. John merged pending changes to $/smlnj-lib.cm
2932
2933		1. Allen's previous change accidentally backed out of one of Lal's
2934		earlier changes.  I undid this mistake (re-introducing Lal's change).
2935
2936		2. I used the new topOrder' function from graph-scc.sml (from $/smlnj-lib.cm)
2937		within the compiler where applicable.  There is some code simplification
2938		because of that.
2939
2940		3. The "split" phase (in FLINT) is now part of the default list of phases.
2941		Compiler.Control.LambdaSplitting.* can be used to globally control the
2942		lambda-splitting (cross-module-inlining) engine.  In addition to that,
2943		it can now also be controlled on a per-source basis: CM has been taught
2944		a new tool parameter applicable to ML source files.
2945
2946		- To turn lambda-splitting off completely:
2947		local open Compiler.Control.LambdaSplitting in
2948		val _ = set Off
2949		end
2950		- To make "no lambda-splitting" the global default (but allow per-source
2951		overriding); this is the initial setting:
2952		local open Compiler.Control.LambdaSplitting in
2953		val _ = set (Default NONE)
2954		end
2955		- To make "lambda-splitting with aggressiveness a" the global default
2956		(and allow per-source overriding):
2957		local open Compiler.Control.LambdaSplitting in
2958		val _ = set (Default (SOME a))
2959		end
2960
2961		- To turn lambda-splitting off for a given ML souce file (say: a.sml)
2962		write (in the respective .cm-file):
2963		a.sml (lambdasplitting:off)
2964		- To turn lambda-splitting for a.sml on with minimal aggressiveness:
2965		a.sml (lambdasplitting:on)
2966		- To turn lambda-splitting for a.sml on with aggressiveness <a> (where
2967		<a> is a decimal non-negative integer):
2968		a.sml (lambdasplitting:<a>)
2969		- To turn lambda-splitting for a.sml on with maximal aggressiveness:
2970		a.sml (lambdasplitting:infinity)
2971		- To use the global default for a.sml:
2972		a.sml (lambdasplitting:default)
2973		or simply
2974		a.sml
2975
2976		----------------------------------------------------------------------
2977		Name: Allen Leung
2978		Date: 2001/05/04 01:57:00 EDT
2979		Tag: leunga-20010504-sync
2980		Description:
2981
2982		MLRISC features.
2983
2984		1. Fix to CMPXCHG instructions.
2985		2. Changed RA interface to allow annotations in callbacks.
2986		3. Added a new method to the stream interface to allow annotations updates.
2987
2988		----------------------------------------------------------------------
2989		Name: Matthias Blume
2990		Date: 2001/05/01 11:45:00 EDT
2991		Tag: blume-20010501-pcedittmp
2992		Description:
2993
2994		Changed install.sh to use the current working directory instead of
2995		/usr/tmp for a temporary file (pcedittmp).  The previous choice
2996		of /usr/tmp caused trouble with MacOS X because of file premission
2997		problems.
2998
2999		----------------------------------------------------------------------
3000		Name: Matthias Blume
3001		Date: 2001/04/20 11:10:00 EDT
3002		Tag: blume-20010420-inMLflag
3003		Description:
3004
3005		- added vp_limitPtrMask to vproc-state.h
3006		(for use by the raw-C-calls mechanism to implement proper interrupt
3007		handling)
3008		- made the ML compiler aware of various data-structure offsets so it
3009		can generate code for accessing the vp_inML flag and vp_limitPtrMask
3010		- tweaked mlriscGen.sml to have it emit interrupt-handling code for
3011		raw C-calls
3012
3013		----------------------------------------------------------------------
3014		Name: Lal George
3015		Date: 2001/04/20 09:15:28 EDT
3016		Tag: george-20010420-macosX
3017		Description:
3018
3019		- Changes to port to Mac OS X; Darwin.
3020
3021		- In the process I found that sqrt was broken on the PPC, because the
3022		fsqrt instruction is not implemented.
3023
3024		----------------------------------------------------------------------
3025		Name: Matthias Blume
3026		Date: 2001/04/18 12:45:00 EDT
3027		Tag: blume-20010418-ccalls
3028		Description:
3029
3030		- fixed two off-by-4 errors in the x86-specific c-calls implementation
3031		(this bug prevented structure arguments containing pointers from being
3032		passed correctly)
3033		- changed the raw-C-call code in mlriscGen.sml in such a way that
3034		structure arguments are represented as a pointer to the beginning
3035		of the structure (instead of having a series of synthesized arguments,
3036		one for each structure member)
3037
3038		- made makeml script's verbosity level configurable via environment
3039		variable (MAKEML_VERBOSITY)
3040
3041		- eliminated placeholder implementations for f32l, w16s, i16s, and f32s
3042		in rawmem-x86.sml; we are now using the real thing
3043
3044		----------------------------------------------------------------------
3045		Name: Matthias Blume
3046		Date: 2001/03/22 16:25:00 EST
3047		Tag: blume-20010322-bootfiles
3048		Description:
3049
3050		Created a new set of bootfiles (for your automatic installation convenience).
3051
3052		----------------------------------------------------------------------
3053		Name: Matthias Blume
3054		Date: 2001/03/22 15:10:00 EST
3055		Tag: blume-20010322-rawmem-parcm
3056		Description:
3057
3058		1. All "raw memory access" primitives for the new FFI are implemented now
3059		(at least on the x86).
3060		2. Some further cleanup of CM's parallel make mechanism.
3061
3062		----------------------------------------------------------------------
3063		Name: Matthias Blume
3064		Date: 2001/03/19 17:53:00 EST
3065		Tag: blume-20010319-parallel
3066		Description:
3067
3068		Parallel make (using compile servers) now works again.
3069
3070		To this end, CM.stabilize and CMB.make have been modified to work in
3071		two passes when compile servers are attached:
3072		1. Compile everything, do not perform stabilization; this pass
3073		uses compile servers
3074		2. Stabilize everything; this pass does not use compile servers
3075		If there are no compile servers, the two passes are combined into one
3076		(as before).  Splitting the passes increases the inherent parallelism
3077		in the dependency graph because the entire graph including all
3078		libraries is available at the same time.  This, in turn, improves
3079		server utilization.  The downside is that the master process will
3080		have to do some extra work after compilation is done (because for
3081		technical reasons it must re-read all the binfiles during stabilization).
3082
3083		----------------------------------------------------------------------
3084		Name: Matthias Blume
3085		Date: 2001/03/16 12:22:00 EST
3086		Tag: blume-20010316-bootfiles
3087		Description:
3088
3089		Created a new set of bootfiles (for your automatic installation convenience).
3090
3091		----------------------------------------------------------------------
3092		Name: Matthias Blume
3093		Date: 2001/03/16 11:00:00 EST
3094		Tag: blume-20010316-MLTREE-fixup
3095		Description:
3096
3097		This is a minor fixup for an (untagged) earlier commit by Allen.
3098		(A file was missing).
3099
3100		----------------------------------------------------------------------
3101		Name: Allen Leung
3102		Date: Mon Mar  5 18:54:57 EST 2001
3103		Tag: leunga-20010305-cut-support
3104
3105		1. New support for alternative control-flow in MLTREE.
3106		Currently we support
3107
3108		FLOW_TO(CALL ...., [k1,...,kn])
3109
3110		This is needed for 'cuts to' in C-- and try/handle-like constructs
3111		in Moby
3112
3113		New assembler flag "asm-show-cutsto" to turn on control-flow debugging.
3114
3115		2. Register Allocator
3116
3117		Changes in interface [from Fermin, John]
3118
3119		3. Alpha 8-bit SLL support [Fermin]
3120
3121		4. All architectures
3122
3123		A new module (ClusterExpandCopies) for expanding parallel copies.
3124
3125		----------------------------------------------------------------------
3126		Name: Allen Leung
3127		Date: 2001/02/27 23:07:00 EST
3128		Tag: leunga-20010227-minor-stuff
3129
3130		1. Alpha bug fix for CMOVNE
3131		2. Handle mltree COND(..,FCMP ...,...)
3132		3. Bug fix in simplifier
3133
3134		----------------------------------------------------------------------
3135		Name: Matthias Blume
3136		Date: 2001/01/30 17:50:00 EST
3137		Tag: blume-20010130-sync
3138		Description:
3139
3140		This is just a minor update to sync my devel branch with the main brach.
3141		The only visible change is the addition of some README files.
3142
3143		----------------------------------------------------------------------
3144		Name: Matthias Blume
3145		Date: 2001/01/12 23:30:00 JST
3146		Tag: blume-20010112-bootfiles
3147		Description:
3148
3149		Made a new set of bootfiles that goes with the current state of the
3150		repository.
3151
3152		----------------------------------------------------------------------
3153		Name: Matthias Blume
3154		Date: 2001/01/12 21:20:00 JST
3155		Tag: blume-20010112-sync
3156		Description:
3157
3158		I am just flushing out some minor changes that had accumulated in
3159		my private branch in order to sync with the main tree.  (This is
3160		mainly because I had CVS trouble when trying to merge _into_ my
3161		private branch.)
3162
3163		Most people should be completely unaffected by this.
3164
3165		----------------------------------------------------------------------
3166		Name: Allen Leung
3167		Date: Thu Jan 11 21:03:00 EST 2001
3168		Tag: leunga-20010111-labexp=mltree
3169		Description:
3170
3171		1.  Removed the type LabelExp and replace it by MLTree.
3172		2.  Rewritten mltree-simplify with the pattern matcher tool.
3173		3.  There were some bugs in alpha code generator which would break
3174		64-bit code generation.
3175		4.  Redo the tools to generate code with the
3176		5.  The CM files in MLRISC (and in src/system/smlnj/MLRISC)
3177		are now generated by perl scripts.
3178
3179		----------------------------------------------------------------------
3180		Name: Matthias Blume
3181		Date: 2001/01/10 21:55:00 JST
3182		Tag: blume-20010110-rcc
3183		Description:
3184
3185		The RCC stuff now seems to work (but only on the x86).
3186		This required hacking of the c-calls interface (and -implementation) in
3187		MLRISC.
3188
3189		Normal compiler users should be unaffected.
3190
3191		----------------------------------------------------------------------
3192		Name: Matthias Blume
3193		Date: 2001/01/09 01:20:00 JST
3194		Tag: blume-20010109-rcc
3195		Description:
3196
3197		This is a fairly big patch, flushing out a large number of pending
3198		changes that I made to my development copy over the last couple of days.
3199
3200		Of practical relevance at this moment is a workaround for a pickling
3201		bug that Allen ran into the other day.  The cause of the bug itself is
3202		still unknown and it might be hard to fix it properly, but the
3203		workaround has some merits of its own (namely somewhat reducing pickling
3204		overhead for certain libraries).  Therefore, I think this solution should
3205		be satisfactory at this time.
3206
3207		The rest of the changes (i.e., the vast majority) has to do with my
3208		ongoing efforts of providing direct support for C function calls from
3209		ML.  At the moment there is a new primop "RAW_CCALL", typing magic
3210		in types/cproto.sml (invoked from FLINT/trans/translate.sml), a new
3211		case in the FLINT CPS datatype (RCC), changes to cps/convert.sml to
3212		translate uses of RAW_CCALL into RCC, and changes to mlriscGen.sml to
3213		handle RCC.
3214
3215		The last part (the changes to mlriscGen.sml) are still known to be
3216		wrong on the x86 and not implemented on all other architectures.  But
3217		the infrastructure is in place. I had to change a few functor
3218		signatures in the backend to be able to route the CCalls interface
3219		from MLRISC there, and I had to specialize the mltree type (on the
3220		x86) to include the necessary extensions. (The extensions themselves
3221		were already there and redy to go in MLRISC/x86).
3222
3223		Everything should be very happy as soon as someone helps me with
3224		mlriscGen.sml...
3225
3226		In any case, nothing of this should matter to anyone as long as the
3227		new primop is not being used (which is going to be the case unless you
3228		find it where I hid it :). The rest of the compiler is completely
3229		unaffected.
3230
3231		----------------------------------------------------------------------
3232		Name: Matthias Blume
3233		Date: 2001/01/05 00:30:00 JST
3234		Tag: blume-20010105-primops
3235		Description:
3236
3237		Added some experimental support for work that I am doing right now.
3238		These changes mostly concern added primops, but there is also a new
3239		experimental C library in the runtime system (but currently not enabled
3240		anywhere except on Linux/X86).
3241
3242		In the course of adding primops (and playing with them), I discovered that
3243		Zhong's INL_PRIM hack (no type info for certain primops) was, in fact, badly
3244		broken.  (Zhong was very right he labeled this stuff as "major gross hack".)
3245		To recover, I made type information in INL_PRIM mandatory and changed
3246		prim.sml as well as built-in.sml accordingly.  The InLine structure now
3247		has complete, correct type information (i.e., no bottom types).
3248
3249		Since all these changes mean that we need new binfiles, I also bumped the
3250		version number to 110.32.1.
3251
3252		----------------------------------------------------------------------
3253		Name: Matthias Blume
3254		Date: 2000/12/30 22:10:00 JST
3255		Tag: blume-20001230-various
3256		Description:
3257
3258		Added proxy libraries for MLRISC and let MLRISC libraries refer
3259		to each other using path anchors.  (See CM manual for explanation.)
3260
3261		Updated CM documentation.
3262
3263		Fixed some bugs in CM.
3264
3265		Implemented "proxy" libraries (= syntactic sugar for CM).
3266
3267		Added "-quiet" option to makeml and changed runtime system accordingly.
3268
3269		Added cleanup handler for exportML to reset timers and compiler stats.
3270
3271		----------------------------------------------------------------------
3272		Name: Lal George
3273		Date: 2000/12/22 22:22:58 EST 2000
3274		Tag: Release_110_32
3275		Description:
3276
3277		Infinite precision used throughout MLRISC.
3278		see MLRISC/mltree/machine-int.sig
3279
3280		----------------------------------------------------------------------
3281		Name: Matthias Blume
3282		Date: 2000/12/22 23:16:00 JST
3283		Tag: blume-20001222-warn
3284		Description:
3285
3286		Corrected wording and formatting of some CM warning message which I
3287		broke in my previous patch.
3288
3289		----------------------------------------------------------------------
3290		Name: Matthias Blume
3291		Date: 2000/12/22 21:20:00 JST
3292		Tag: blume-20001222-anchorenv
3293		Description:
3294
3295		Fixed CM's handling of anchor environments in connection with CMB.make.
3296
3297		----------------------------------------------------------------------
3298		Name: Matthias Blume
3299		Date: 2000/12/22 13:15:00 JST
3300		Tag: blume-20001222-cleanup
3301		Description:
3302
3303		Removed src/cm/ffi which does not (and did not) belong here.
3304
3305		----------------------------------------------------------------------
3306		Name: Matthias Blume
3307		Date: 2000/12/21 23:55:00 JST
3308		Tag: blume-20001221-exn
3309		Description:
3310
3311		Probably most important: CM no longer silently swallows all exceptions
3312		in the compiler.
3313		Plus: some other minor CM changes.  For example, CM now reports some
3314		sizes for generated binfiles (code, data, envpickle, lambdapickle).
3315
3316		----------------------------------------------------------------------
3317		Name: Matthias Blume
3318		Date: 2000/12/15 00:01:05 JST
3319		Tag: blume-20001215-dirtool
3320		Description:
3321
3322		- "dir" tool added.
3323		- improvements and cleanup to Tools structure
3324		- documentation updates
3325
3326		----------------------------------------------------------------------
3327		Name: Allen Leung
3328		Date: Thu Dec 14 03:45:24 EST 2000
3329		Description:
3330		Tag:  leunga-20001214-int-inf
3331		Description:
3332
3333		In IntInf, added these standard functions, which are missing from our
3334		implementation:
3335
3336		andb : int * int -> int
3337		xorb : int * int -> int
3338		orb  : int * int -> int
3339		notb : int -> int
3340		<<   : int * word -> int
3341		~>>  : int * word -> int
3342
3343		Not tested, I hope they are correct.
3344
3345		----------------------------------------------------------------------
3346		Name: Allen Leung
3347		Date: Fri Dec  8 19:23:26 EST 2000
3348		Description:
3349		Tag:  leunga-20001208-nowhere
3350		Description:
3351
3352		Slight improvements to the 'nowhere' tool to handle OR-patterns,
3353		to generate better error messages etc.  Plus a brief manual.
3354
3355		----------------------------------------------------------------------
3356		Name: Lal George
3357		Date: 2000/12/08 09:54:02 EST 2000
3358		Tag: Release_110_31
3359		Description:
3360
3361		- Version 110.31
3362		----------------------------------------------------------------------
3363		Name: Allen Leung
3364		Date: Thu Dec  7 22:01:04 EST 2000
3365		Tag:  leunga-20001207-cell-monster-hack
3366		Description:
3367
3368		Major MLRISC internal changes.  Affect all clients.
3369		Summary:
3370
3371		1.  Type CELLS.cell = int is now replaced by a datatype.
3372		As a result, the old regmap is now gone.  Almost all interfaces
3373		in MLRISC change as a consequence.
3374
3375		2.  A new brand version of machine description tool (v3.0) that generates
3376		modules expecting the new interface.  The old version is removed.
3377
3378		3.  The RA interface has been further abstracted into two new functors.
3379		RISC_RA and X86RA.  These functors have much simpler interfaces.
3380		[See also directory MLRISC/demo.]
3381
3382		4.  Some other new source->source code generation tools are available:
3383
3384		a. MLRISC/Tools/RewriteGen -- generate rewriters from rules.
3385		b. MLRISC/Tools/WhereGen -- expands conditional pattern matching rules.
3386		I use this tool to generate the peephole optimizers---with the new
3387		cell type changes, peephole rules are becoming difficult to write
3388		without conditional pattern matching.
3389
3390		5.  More Intmap -> IntHashTable change.  Previous changes by Matthias didn't
3391		cover the entire MLRISC source tree so many things broke.
3392
3393		6.  CM files have been moved to the subdirectory MLRISC/cm.
3394		They are moved because there are a lot of them and they clutter up the
3395		root dir.
3396
3397		7.  More detailed documentation to come...
3398
3399		NOTE: To rebuild from 110.30 (ftp distribution), you'll have to do
3400		a makeml -rebuild first.  This is because of other other
3401		changes that Matthias has made (see below).
3402
3403
3404		----------------------------------------------------------------------
3405		Name: Matthias Blume
3406		Date: 2000/11/30 23:12:00 JST
3407		Tag: blume-20001130-filereorg
3408		Description:
3409
3410		Some manual updates and some file reorganizations in CM.
3411
3412		----------------------------------------------------------------------
3413		Name: Matthias Blume
3414		Date: 2000/11/24 17:45:00 JST
3415		Tag: blume-20001124-link
3416		Description:
3417
3418		Drastically improved link traversal code for the case that the dynamic
3419		value was already loaded at bootstrap time.  As a result, CM and CMB
3420		now both load blazingly fast -- even on a very slow machine.  Also,
3421		memory consumption has been further reduced by this.
3422
3423		Warning: The format of the PIDMAP file has changed.  THerefore, to
3424		bootstrap you have to do this:
3425
3426		1. Run CMB.make
3427		2. Make a symbolic link for the boot directory:
3428		ln -s sml.boot.ARCH-OS xxx
3429		3. "Rebuild" the boot directory:
3430		./makeml -boot xxx -rebuild sml ; rm xxx
3431		4. Boot normally:
3432		./makeml
3433
3434		----------------------------------------------------------------------
3435		Name: Matthias Blume
3436		Date: 2000/11/21 21:20:00 JST
3437		Tag: blume-20001121-tools
3438		Description:
3439
3440		Continued hacking on autoloading problem -- with success this time.
3441		Also changed tool-plugin mechanism.  See new CM manual.
3442
3443		----------------------------------------------------------------------
3444		Name: Matthias Blume
3445		Date: 2000/11/19 14:30:00 JST
3446		Tag:  blume-20001119-autoload
3447		Description:
3448
3449		Some hacking to make autoloading faster.  Success for CMB, no success
3450		so far for CM.  There is a reduced structure CM' that autoloads faster.
3451		(This is a temporary, non-documented hack to be eliminated again when
3452		the general problem is solved.)
3453
3454		----------------------------------------------------------------------
3455		Name: Matthias Blume
3456		Date: 2000/11/17 14:10:00 JST
3457		Tag: blume-20001117-pickle-lib
3458		Description:
3459
3460		1. Eliminated comp-lib.cm
3461		2. Made pickle-lib.cm
3462		3. Eliminated all uses of intset.sml (from comp-lib.cm)
3463		4. Replaced all uses of intmap.{sig,sml} (from comp-lib.cm) with
3464		equivalent constructs from smlnj-lib.cm (INtHashTable).
3465		5. Point 4. also goes for those uses of intmap.* in MLRISC.
3466		Duplicated intmap modules thrown out.
3467		6. Hunted down all duplicated SCC code and replaced it with
3468		equivalent stuff (GraphSCCFn from smlnj-lib.cm).
3469		7. Rewrote Feedback module.
3470		8. Moved sortedlist.sml into viscomp-lib.cm.  Eventually it
3471		should be thrown out and equivalent modules from smlnj-lib.cm
3472		should be used (IntRedBlackSet, IntListSet, ...).
3473
3474		Confirmed that compiler compiles to fixpoint.
3475
3476		----------------------------------------------------------------------
3477		Name: Allen Leung
3478		Date: 2000/11/10 18:00:00
3479		Tag: leunga-20001110-new-x86-fp
3480
3481		A new x86 floating point code generator has been added.
3482		By default this is turned off.  To turn this on, do:
3483
3484		CM.autoload "$smlnj/compiler.cm";
3485		Compiler.Control.MLRISC.getFlag "x86-fast-fp" := true;
3486
3487		Changes:
3488
3489		1.  Changed FTAN to FPTAN so that the assembly output is correct.
3490		2.  Changed the extension callback for FTANGENT to generate:
3491
3492		fptan
3493		fstp  %st(0)
3494		instead of
3495		fptan
3496		fstpl ftempmem
3497
3498		3.  Numerous assembly fixes for x86.
3499
3500		5.  Cleaned up the machine code output module x86/x86MC.sml and added
3501		support for a whole bunch of instructions and addressing modes:
3502
3503		fadd/fsub/fsubr/fmul/fdiv/fdivr  %st, %st(n)
3504		faddp/fsubp/fsubrp/fmulp/fdivp/fdivrp  %st, %st(n)
3505		fadd/fsub/fsubr/fmul/fdiv/fdivr  %st(n), %st
3506		fiadd/fisub/fisubr/fimul/fidiv/fidivr mem
3507		fxch %st(n)
3508		fld %st(n)
3509		fst %st(n)
3510		fst mem
3511		fstp %st(n)
3512		fucom %st(n)
3513		fucomp %st(n)
3514
3515		All these are now generated when the fast fp mode is turned on.
3516
3517		6.  Removed the dedicated registers %st(0), ..., %st(7) from X86CpsRegs
3518
3519		----------------------------------------------------------------------
3520		Name: Matthias Blume
3521		Date: 2000/11/09 11:20:00 JST
3522		Tag: blume-20001109-scc
3523		Description:
3524
3525		Eliminated some code duplication:
3526
3527		1. Added "where" clause to GraphSCCFn in SML/NJ Library.
3528		(Otherwise the functor is useless.)
3529		2. Used GraphSCCFn where SCCUtilFun was used previously.
3530		3. Got rid of SCCUtilFun (in comp-lib.cm).
3531
3532		----------------------------------------------------------------------
3533		Name: Lal George
3534		Date: 2000/11/06 09:02:21 EST 2000
3535		Tag: Release_110_30
3536		Description:
3537
3538		- Version 110.30
3539		----------------------------------------------------------------------
3540		Name: Matthias Blume
3541		Date: 2000/11/04 14:45:00
3542		Tag: blume-20001104-mlbuild
3543		Description:
3544
3545		- Made ml-build faster on startup.
3546		- Documentation fixes.
3547
3548		----------------------------------------------------------------------
3549		Name: Matthias Blume
3550		Date: 2000/11/02 17:00:00 JST
3551		Tag: blume-20001102-condcomp
3552		Description:
3553
3554		- Small tweaks to pickler -- new BOOTFILES!
3555		- Version bumped to 110.29.2.
3556		- Added conditional compilation facility to init.cmi (see comment there).
3557		----------------------------------------------------------------------
3558		Name: Allen Leung
3559		Date: 2000/10/23 19:31:00
3560		Tag: leunga-20001023-demo-ra
3561
3562		1. Minor RA changes that improves spilling on x86 (affects Moby and C-- only)
3563		2. Test programs for the graph library updated
3564		3. Some new MLRISC demo programs added
3565
3566		----------------------------------------------------------------------
3567		Name: Matthias Blume
3568		Date: 2000/08/31 22:15:00 JST
3569		Tag: blume-20001017-errmsg
3570		Description:
3571
3572		More error message grief: Where there used to be no messages, there
3573		now were some that had bogus error regions.  Fixed.
3574
3575		----------------------------------------------------------------------
3576		Name: Matthias Blume
3577		Date: 2000/08/31 17:30:00 JST
3578		Tag: blume-20001017-v110p29p1
3579		Description:
3580
3581		I made a version 110.29.1 with new bootfiles.
3582
3583		Changes:  Modified pickler/unpickler for faster and leaner unpickling.
3584		CM documentation changes and a small bugfix in CM's error reporting.
3585
3586		----------------------------------------------------------------------
3587		Name: Lal George
3588		Date: 2000/09/27 14:42:35 EDT
3589		Tag: george-20000927-nodestatus
3590		Description:
3591
3592		Changed the type of the nodestatus, so that:
3593
3594		SPILLED(~1)             is now SPILLED
3595		SPILLED(m) where m>=0   is now MEMREG(m)
3596		SPILLED(s) where s<~1   is now SPILL_LOC(~s)
3597
3598		----------------------------------------------------------------------
3599		Name: Matthias Blume
3600		Date: 2000/09/07 14:45:00 JST
3601		Tag: blume-20000907-cmerrmsg
3602		Description:
3603
3604		Small tweak to CM to avoid getting ML syntax error messages twice.
3605
3606		----------------------------------------------------------------------
3607		Name: Matthias Blume
3608		Date: 2000/08/31 18:00:00 JST
3609		Tag: blume-20000831-cvsbootfiles
3610		Description:
3611
3612		New URL for boot files (because the 110.29 files on the BL server do
3613		now work correctly with my updated install scripts for yacc and lex).
3614
3615		----------------------------------------------------------------------
3616		Name: Matthias Blume
3617		Date: 2000/08/08 12:33:00 JST
3618		Tag: blume-20000808-manual
3619		Description:
3620
3621		Tiny update to CM manual.
3622
3623		----------------------------------------------------------------------
3624		Name: Allen Leung
3625		Date: 2000/08/7 19:31:00
3626		Tag: leunga-20000807-a-whole-bunch-of-stuff
3627
3628		Moby, C--, SSA, x86, machine descriptions etc.  Should only affect C--
3629		and Mobdy.
3630
3631		1.  x86
3632
3633		a.  Fixes to peephole module by John and Dan.
3634		b.  Assembly fix to SETcc by Allen.
3635		c.  Fix to c-call by John.
3636		d.  Fix to spilling by John.  (This one deals with the missing FSTPT case)
3637		e.  Instruction selection optimization to SETcc as suggested by John.
3638
3639		For example,
3640
3641		MV(32, x, COND(32, CMP(32, LT, a, b), LI 1, LI 0))
3642
3643		should generate:
3644
3645		MOVL a, x
3646		SUBL b, x
3647		SHRL 31, x
3648
3649		2.  IR stuff
3650
3651		A bunch of new DJ-graph related algorithms added.  These
3652		speed up SSA construction.
3653
3654		3.  SSA + Scheduling
3655
3656		Added code for SSA and scheduling to the repository
3657
3658		----------------------------------------------------------------------
3659		Name: Lal George
3660		Date: 2000/07/27 11:53:14 EDT
3661
3662		Tag: lal-20000727-linux-ppc
3663		Description:
3664
3665		Made changes to support Linux PPC.
3666		p.s. I have confirmation that the 110.29 boot files work fine.
3667
3668		----------------------------------------------------------------------
3669		Name: Matthias Blume
3670		Date: 2000/07/27 17:40:00 JST
3671		Tag: blume-20000727-scripts
3672		Description:
3673
3674		!!!! WARNING !!!!
3675		You must recompile the runtime system!
3676		!!!! WARNING !!!!
3677
3678		This is basically another round of script-enhancements:
3679
3680		1. sml, ml-build, and ml-makedepend accept options -D and -U to define
3681		and undefine CM preprocessor symbols.
3682
3683		2. ml-build avoids generating a new heap image if it finds that the
3684		existing one is still ok.  (The condition is that no ML file had to
3685		be recompiled and all ML files are found to be older that the heap
3686		file.)
3687
3688		To make this work smoothly, I also hacked the runtime system as
3689		well as SMLofNJ.SysInfo to get access to the heap image suffix
3690		(.sparc-solaris, ...) that is currently being used.
3691
3692		Moreover, the signature of CM.mk_standalone has changed.  See the
3693		CM manual.
3694
3695		3. ml-makedepend accepts additional options -n, -a, and -o.  (See the
3696		CM manual for details.)
3697
3698		4. More CM manual updates:
3699		- all of the above has been documented.
3700		- there is now a section describing the (CM-related) command line
3701		arguments that are accepted by the "sml" command
3702
3703		----------------------------------------------------------------------
3704		Name: Matthias Blume
3705		Date: 2000/07/25 16:20:00 JST
3706		Tag: blume-20000725-makedepend
3707		Description:
3708
3709		Added a script called ml-makedepend.  This can be used in makefiles
3710		for Unix' make in a way very similar to the "makedepend" command for
3711		C.
3712
3713		The script internally uses function CM.sources.
3714
3715		Synopsis:
3716
3717		ml-makedepend [-f makefile] cmfile targetname
3718
3719		The default for the makefile is "makefile" (or "Makefile" should
3720		"makefile" not exist).
3721
3722		ml-makedepend adds a cmfile/targetname-specific section to this
3723		makefile (after removing the previous version of this section).  The
3724		section contains a single dependency specification with targetname on
3725		the LHS (targetname is an arbitrary name), and a list of files derived
3726		from the cmfile on the RHS.  Some of the files on the RHS are
3727		ARCH/OPSYS-specific.  Therefore, ml-makedepend inserts references to
3728		"make" variables $(ARCH) and $(OPSYS) in place of the corresponding
3729		path names.  The makefile writer is responsible for making sure that
3730		these variables have correct at the time "make" is invoked.
3731
3732		----------------------------------------------------------------------
3733		Name: Matthias Blume
3734		Date: 2000/07/22 23:30:00 JST
3735		Tag: blume-20000722-urlupdate
3736		Description:
3737
3738		Changed BOOT and config/srcarchiveurl to point to BL server:
3739
3740		ftp://ftp.research.bell-labs.com/dist/smlnj/working/110.29/
3741
3742		----------------------------------------------------------------------
3743		Name: Matthias Blume
3744		Date: 2000/07/18 18:00:00 JST
3745		Tag: blume-20000718-Version_110_29
3746		Description:
3747
3748		1. Updated src/compiler/TopLevel/main/version.sml to version 110.29
3749
3750		2. Updated config/version to 110.29
3751
3752		3. Updated config/srcarchiveurl
3753
3754		3. New boot files!
3755		ftp://ftp.cs.princeton.edu/pub/people/blume/sml/110.29-autofetch
3756
3757		----------------------------------------------------------------------
3758		Name: Matthias Blume
3759		Date: 2000/07/11 13:58:00 JST
3760		Tag: blume-20000711-doctypo
3761		Description:
3762
3763		Fixed a few typos in CM manual.
3764
3765		----------------------------------------------------------------------
3766		Name: Allen Leung
3767		Date: 2000/06/15 00:38:00
3768		Tag: leunga-20000704-sparc-x86
3769
3770		1. x86 peephole improvement sp += k; sp -= k => nop  [from John]
3771		2. fix to x86 RET bug [found by Dan Grossman]
3772		3. sparc assembly bug fix for ticc instructions [found by Fermin]
3773
3774		Affects c-- and moby only
3775
3776		----------------------------------------------------------------------
3777		Name: Matthias Blume
3778		Date: 2000/07/04 15:26:00
3779		Tag: blume-20000704-trigger
3780		Description:
3781
3782		1. Improvements to CM manual.
3783		2. SMLofNJ.Internals.BTrace.trigger reinstated as an alternative way
3784		of getting a back-trace.  The function, when called, raises an
3785		internal exception which explicitly carries the full back-trace history,
3786		so it is unaffected by any intervening handle-raise pairs ("trivial"
3787		or not).  The interactive loop will print that history once it arrives
3788		at top level.
3789		Short of having all exceptions implicitly carry the full history, the
3790		recommended way of using this facility is:
3791		- compile your program with instrumentation "on"
3792		- run it, when it raises an exception, look at the history
3793		- if the history is "cut off" because of some handler, go and modify
3794		your program so that it explicitly calls BTrace.trigger
3795		- recompile (still instrumented), and rerun; look at the full history
3796
3797		----------------------------------------------------------------------
3798		Name: Matthias Blume
3799		Date: 2000/07/03 15:36:00 JST
3800		Tag: blume-20000702-manual
3801		Description:
3802
3803		Small corrections and updates to CM manual.
3804
3805		----------------------------------------------------------------------
3806		Name: Matthias Blume
3807		Date: 2000/06/29 16:04:00 JST
3808		Tag: blume-20000629-yacctool
3809		Description:
3810
3811		Changes:
3812
3813		1. Class "mlyacc" now takes separate arguments to pass options to
3814		generated .sml- and .sig-files independently.
3815		2. Corresponding CM manual updates.
3816		3. BTrace module now also reports call sites.  (However, for loop clusters
3817		it only shows from where the cluster was entered.)  There are associated
3818		modifications to core.sml, internals.{sig,sml}, btrace.sml, and btimp.sml.
3819
3820		----------------------------------------------------------------------
3821		Name: Matthias Blume
3822		Date: 2000/06/27 16:51:00 JST
3823		Tag: blume-20000627-noweb
3824		Description:
3825
3826		Changes:
3827
3828		1. Implemented "subdir" and "witness" options for noweb tool.
3829		This caused some slight internal changes in CM's tool implementation.
3830		2. Fixed bug in "tool plugin" mechanism.  This is essentially cleaning
3831		some remaining issues from earlier path anchor changes.
3832		3. Updated CM manual accordingly.
3833
3834		4. Changed implementation of back-tracing so that I now consider it
3835		ready for prime-time.
3836
3837		In particular, you don't have to explicitly trigger the back-trace
3838		anymore.  Instead, if you are running BTrace-instrumented code and
3839		there is an uncaught exception (regardless of whether or not it was
3840		raised in instrumented code), the top-level evalloop will print
3841		the back-trace.
3842
3843		Features:
3844
3845		- Instrumented and uninstrumented code work together seemlessly.
3846		(Of course, uninstrumented code is never mentioned in actual
3847		back-traces.)
3848
3849		- Asymptotic time- and space-complexity of instrumented code is
3850		equal to that of uninstrumented code.  (This means that
3851		tail-recursion is preserved by the instrumentation phase.)
3852
3853		- Modules whose code has been instrumented in different sessions
3854		work together without problem.
3855
3856		- There is no penalty whatsoever on uninstrumented code.
3857
3858		- There is no penalty on "raise" expressions, even in
3859		instrumented code.
3860
3861		A potential bug (or perhaps it is a feature, too):
3862
3863		A back-trace reaches no further than the outermost instrumented
3864		non-trivial "raise".  Here, a "trivial" raise is one that is the
3865		sole RHS of a "handle" rule.  Thus, back-traces reach trough
3866
3867		<exp> handle e => raise e
3868
3869		and even
3870
3871		<exp> handle Foo => raise Bar
3872
3873		and, of course, through
3874
3875		<exp> handle Foo => ...
3876
3877		if the exception was not Foo.
3878
3879		Back-traces always reach right through any un-instrumented code
3880		including any of its "handle" expressions, trivial or not.
3881
3882		To try this out, do the following:
3883
3884		- Erase all existing binfiles for your program.
3885		(You may keep binfiles for those modules where you think you
3886		definitely don't need back-tracing.)
3887		- Turn on back-trace instrumentation:
3888		SMLofNJ.Internals.BTrace.mode (SOME true);
3889		- Recompile your program.  (I.e., run "CM.make" or "use".)
3890		- You may now turn instrumentation off again (if you want):
3891		SMLofNJ.Internals.BTrace.mode (SOME false);
3892		- Run your program as usual.  If it raises an exception that
3893		reaches the interactive toplevel, then a back-trace will
3894		automatically be printed.  After that, the toplevel loop
3895		will print the exception history as usual.
3896
3897		----------------------------------------------------------------------
3898		Name: Matthias Blume
3899		Date: 2000/06/26 09:56:46 JST
3900		Tag: blume-20000626-setup
3901		Description:
3902
3903		CM: - setup-parameter to "sml" added; this can be used to run arbitrary
3904		ML code before and after compiling a file (e.g., to set compiler
3905		flags)
3906
3907		Compiler: - improved btrace API (in core.sml, internals.{sig,sml})
3908		- associated changes to btrace.sml (BTrace instrumentation pass)
3909		- cleaner implementation of btimp.sml (BTrace tracing and report
3910		module)
3911
3912		CM manual: * new path encoding documented
3913		* description of setup-parameter to "sml" added
3914
3915		The biggest user-visible change to back-tracing is that it is no
3916		longer necessary to compile all traced modules within the same
3917		session.  (This was a real limitation.)
3918
3919		----------------------------------------------------------------------
3920		Name: Matthias Blume
3921		Date: 2000/06/24 12:40:00 JST
3922		Tag: blume-20000624-startup
3923		Description:
3924
3925		Fixes startup slowdown problem.  (I was calling SrcPath.sync a _tad_
3926		bit too often -- to put it mildly. :)
3927
3928		----------------------------------------------------------------------
3929		Name: Matthias Blume
3930		Date: 2000/06/23 18:20:00 JST
3931		Tag: blume-20000623-btrace
3932		Description:
3933
3934		This updates adds a backtrace facility to aid programmers in debugging
3935		their programs.  This involves the following changes:
3936
3937		1. Module system/smlnj/init/core.sml (structure _Core) now has hooks for
3938		keeping track of the current call stack.  When programs are compiled
3939		in a special mode, the compiler will insert calls to these hooks
3940		into the user program.
3941		"Hook" means that it is possible for different implementations of
3942		back-tracing to register themselves (at different times).
3943
3944		2. compiler/MiscUtil/profile/btrace.sml implements the annotation phase
3945		as an Absyn.dec->Absyn.dec rewrite.  Normally this phase is turned off.
3946		It can be turned on using this call:
3947		SMLofNJ.Internals.BTrace.mode (SOME true);
3948		Turning it off again:
3949		SMLofNJ.Internals.BTrace.mode (SOME false);
3950		Querying the current status:
3951		SMLofNJ.Internals.BTrace.mode NONE;
3952		Annotated programs are about twice as big as normal ones, and they
3953		run a factor of 2 to 4 slower with a dummy back-trace plugin (one
3954		where all hooks do nothing).  The slowdown with a plugin that is
3955		actually useful (such as the one supplied by default) is even greater,
3956		but in the case of the default plugin it is still only an constant
3957		factor (amortized).
3958
3959		3. system/Basis/Implementation/NJ/internals.{sig,sml} have been augmented
3960		with a sub-structure BTrace for controlling back-tracing.  In particular,
3961		the above-mentioned function "mode" controls whether the annotation
3962		phase is invoked by the compiler.  Another important function is
3963		"trigger": when called it aborts the current execution and causes
3964		the top-level loop to print a full back-trace.
3965
3966		4. compiler/MiscUtil/profile/btimp.sml is the current default plugin
3967		for back-tracing.  It keeps track of the dynamic call stack and in
3968		addition to that it keeps a partial history at each "level" of that
3969		stack.  For example, if a tail-calls b, b tail-calls c, and c tail-calls
3970		d and b (at separate times, dynamically), then the report will show:
3971
3972		GOTO   d
3973		/c
3974		GOTO  \b
3975		CALL   a
3976
3977		This shows that there was an initial non-tail call of a, then a
3978		tail-call to b or c, looping behavior in a cluster of functions that
3979		consist of b and c, and then a goto from that cluster (i.e., either from
3980		b or from c) to d.
3981
3982		Note that (depending on the user program) the amount of information
3983		that the back-trace module has to keep track of at each level is bounded
3984		by a constant.  Thus, the whole implementation has the same asymptotical
3985		complexity as the original program (both in space and in time).
3986
3987		5. compiler/TopLevel/interact/evalloop.sml has been modified to
3988		handle the special exception SMLofNJ.Internals.BTrace.BTrace
3989		which is raised by the "trigger" function mentioned above.
3990
3991		Notes on usage:
3992
3993		- Annotated code works well together with unannotated code:
3994		Unannotated calls simply do not show up at all in the backtrace.
3995
3996		- It is not a good idea to let modules that were annotated during
3997		different sessions run at the same time.  This is because the compiler
3998		chooses small integers to identify individual functions, and there
3999		will be clashes if different modules were compiled in separate sessions.
4000		(Nothing will crash, and you will even be told about the clashes, but
4001		back-trace information will in general not be useful.)
4002
4003		- Back-tracing can be confused by callcc and capture.
4004
4005		- The only way of getting a back-trace right now is to explicitly
4006		invoke the "trigger" function from your user program.  Eventually, we
4007		should make every exception carry back-trace information (if
4008		available).  But since this creates more overhead at "raise"-time
4009		(similar to the current exnHistory overhead), I have not yet
4010		implemented this.  (The implementation will be rather easy.)  With
4011		exceptions carrying back-trace information, this facility will be even
4012		more useful because users don't need to modify their programs...
4013
4014		- While it is possible to compile the compiler with back-trace
4015		annotations turned on (I did it to get some confidence in
4016		correctness), you must make absolutely sure that core.sml and
4017		btimp.sml are compiled WITHOUT annotation!  (core.sml cannot actually
4018		be compiled with annotation because there is no core access yet, but
4019		if you compile btimp.sml with annotation, then the system will go into
4020		an infinite recursion and crash.)
4021		Since CM currently does not know about BTrace, the only way to turn
4022		annotations on and off for different modules of the compiler is to
4023		interrupt CMB.make, change the settings, and re-invoke it.  Of course,
4024		this is awkward and clumsy.
4025
4026		Sample sessions:
4027
4028		Standard ML of New Jersey v110.28.1 [FLINT v1.5], June 5, 2000
4029		- SMLofNJ.Internals.BTrace.mode (SOME true);
4030		[autoloading]
4031		[autoloading done]
4032		val it = false : bool
4033		- structure X = struct
4034		-     fun main n = let
4035		-         fun a (x, 0) = d x
4036		-           | a (x, n) = b (x, n - 1)
4037		-         and b (x, n) = c (x, n)
4038		-         and c (x, n) = a (x, n)
4039		-         and d x = e (x, 3)
4040		-         and e (x, 0) = f x
4041		-           | e (x, n) = e (x, n - 1)
4042		-         and f 0 = SMLofNJ.Internals.BTrace.trigger ()
4043		-           | f n = n * g (n - 1)
4044		-         and g n = a (n, 3)
4045		-     in
4046		-         f n
4047		-     end
4048		- end;
4049		structure X : sig val main : int -> int end
4050		- X.main 3;
4051		*** BACK-TRACE ***
4052		GOTO   stdIn:4.2-13.20: X.main[2].f
4053		GOTO-( stdIn:4.2-13.20: X.main[2].e
4054		GOTO   stdIn:4.2-13.20: X.main[2].d
4055		/ stdIn:4.2-13.20: X.main[2].a
4056		| stdIn:4.2-13.20: X.main[2].b
4057		GOTO-\ stdIn:4.2-13.20: X.main[2].c
4058		CALL   stdIn:4.2-13.20: X.main[2].g
4059		GOTO   stdIn:4.2-13.20: X.main[2].f
4060		GOTO-( stdIn:4.2-13.20: X.main[2].e
4061		GOTO   stdIn:4.2-13.20: X.main[2].d
4062		/ stdIn:4.2-13.20: X.main[2].a
4063		| stdIn:4.2-13.20: X.main[2].b
4064		GOTO-\ stdIn:4.2-13.20: X.main[2].c
4065		CALL   stdIn:4.2-13.20: X.main[2].g
4066		GOTO   stdIn:4.2-13.20: X.main[2].f
4067		GOTO-( stdIn:4.2-13.20: X.main[2].e
4068		GOTO   stdIn:4.2-13.20: X.main[2].d
4069		/ stdIn:4.2-13.20: X.main[2].a
4070		| stdIn:4.2-13.20: X.main[2].b
4071		GOTO-\ stdIn:4.2-13.20: X.main[2].c
4072		CALL   stdIn:4.2-13.20: X.main[2].g
4073		GOTO   stdIn:4.2-13.20: X.main[2].f
4074		CALL   stdIn:2.15-17.4: X.main[2]
4075		-
4076
4077		(Note that because of a FLINt bug the above code currently does not
4078		compile without BTrace turned on.)
4079
4080		Here is another example, using my modified Tiger compiler:
4081
4082		Standard ML of New Jersey v110.28.1 [FLINT v1.5], June 5, 2000
4083		- SMLofNJ.Internals.BTrace.mode (SOME true);
4084		[autoloading]
4085		[autoloading done]
4086		val it = false : bool
4087		- CM.make "sources.cm";
4088		[autoloading]
4089