Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1055, Thu Feb 7 20:38:59 2002 UTC	revision 1291, Sat Dec 7 03:41:00 2002 UTC
#	Line 13	Line 13
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/12/06 22:40:00 EST
18		Tag: blume-20021206-cm-fileid
19		Description:
20
21		Added a call of SrcPath.sync at the beginning of Parse.parse (in CM).
22		This fixes the problem of CM getting confused by files that suddenly
23		change their identity (e.g., by getting unlinked and recreated by some
24		text editor such as vi).  There might be a better/cheaper/cleaner way
25		of doing this, but for now this will have to do.
26
27		----------------------------------------------------------------------
28		Name: Matthias Blume
29		Date: 2002/10/28 09:50:00 EST
30		Tag: blume-20021028-typecheck
31		Description:
32
33		Exported structure Typecheck from $smlnj/viscomp/core.cm.
34
35		----------------------------------------------------------------------
36		Name: Matthias Blume
37		Date: 2002/10/17 09:10:00 EDT
38		Tag: Release_110_42
39		Description:
40
41		In good old tradition, there has been a slight hiccup so that we have
42		to patch 110.42 after the fact.   The old release tag has been replaced
43		(see below).
44
45		The change solves a problem with two competing approaches the
46		configuration problem regarding MacOS 10.1 vs. MacOS 10.2 which got in
47		each other's way.
48
49		This change only affects the runtime system code and the installer script.
50		(No new bootfiles.)
51
52		----------------------------------------------------------------------
53		Name: Matthias Blume
54		Date: 2002/10/16 12:00:00 EDT
55		Tag: Release_110_42_removed
56		Description:
57
58		New working release.  New bootfiles.
59
60		----------------------------------------------------------------------
61		Name: Matthias Blume
62		Date: 2002/10/10 13:10:00 EDT
63		Tag: blume-20021010-ppc-divs
64		Description:
65
66		The mltree operator DIVS must be implemented with an overflow check on
67		the PPC because the hardware indicates divide-by-zero using "overflow" as
68		well.
69
70		----------------------------------------------------------------------
71		Name: Matthias Blume
72		Date: 2002/07/23 11:45:00
73		Tag: blume-20020723-smlnj-home
74		Description:
75
76		Sml now senses the SMLNJ_HOME environment variable.  If this is set,
77		then the bin dir is assumed to be in $SMLNJ_HOME/bin and (unless
78		CM_PATHCONFIG is also set), the path configuration file is assumed
79		to be in $SMLNJ_HOME/lib/pathconfig.  This way one can easily move
80		the entire tree to some other place and everything will "just work".
81
82		(Companion commands such as ml-build and ml-makedepend also sense this
83		variable.)
84
85		----------------------------------------------------------------------
86		Name: Matthias Blume
87		Date: 2002/07/12 21:19:00 EDT
88		Tag: blume-20020712-liveness
89		Description:
90
91		Exported two useful "step" functions from liveness module (MLRISC).
92
93		----------------------------------------------------------------------
94		Name: Matthias Blume
95		Date: 2002/07/05 16:00 EDT
96		Tag: Release_110_41
97		Description:
98
99		New working release.  New bootfiles.
100
101		----------------------------------------------------------------------
102		Name: Matthias Blume
103		Date: 2002/07/05 10:25:00 EDT
104		Tag: blume-20020705-btimp
105		Description:
106
107		Exported structure BTImp from $smlnj/viscomp/debugprof.cm so that
108		other clients can set up backtracing support.
109
110		----------------------------------------------------------------------
111		Name: Matthias Blume
112		Date: 2002/06/25 17:23:00 EDT
113		Tag: blume-20020625-fpmax
114		Description:
115
116		Fixed a bug in translation of INLMAX (and INLMIN) for the floating-point
117		case.  (The sense of the isNaN test was reversed -- which made min and
118		max always return their first argument.)
119
120		----------------------------------------------------------------------
121		Name: Matthias Blume
122		Date: 2002/06/11
123		Tag: blume-20020611-unixpath
124		Description:
125
126		Back-ported OS.Path.{from,to}UnixPath from idlbasis-devel branch.
127
128		----------------------------------------------------------------------
129		Name: Matthias Blume
130		Date: 2002/06/10 16:35:00 EDT
131		Tag: blume-20020610-ieeereal
132		Description:
133
134		I back-ported my implementation of IEEEReal.fromString from the
135		idlbasis-devel branch so that we can test it.
136
137		Another small change is that ppDec tries to give more information
138		than just "<sig>" in the case of functors.  However, this code is
139		broken in some mysterious way if the functor's body's signature
140		has not been declared by ascription but gets inferred from the
141		implementation.  This needs fixing...
142
143		----------------------------------------------------------------------
144		Name: Matthias Blume
145		Date: 2002/05/31
146		Tag: blume-20020531-btrace-mode
147		Description:
148
149		Resurrected SMLofNJ.Internals.BTrace.mode.  (It accidentally fell by
150		the wayside when I switched over to using Controls everywhere.)
151
152		----------------------------------------------------------------------
153		Name: Lal George
154		Date: 2002/05/23 12:21:40 EDT
155		Tag: george-20020523-visual-labels
156		Description:
157
158		Labels are now displayed in the graphical output to make
159		the fall-through and target blocks obvious.
160
161		----------------------------------------------------------------------
162		Name: Matthias Blume
163		Date: 2002/05/22 11:03:00 EDT
164		Tag: blume-20020522-shrink
165		Description:
166
167		John tweaked yesterday's fix for 1131 to handle an out-of-memory
168		situation that comes up when allocating huge arrays.
169
170		----------------------------------------------------------------------
171		Name: Matthias Blume
172		Date: 2002/05/21 16:00:00 EDT
173		Tag: Release_110_40
174		Description:
175
176		New working release (110.40).  New bootfiles.
177
178		[Also: John Reppy fixed GC bug 1131.]
179
180		----------------------------------------------------------------------
181		Name: Matthias Blume
182		Date: 2002/05/21 12:35:00 EDT
183		Tag: blume-20020521-cmdoc
184		Description:
185
186		CM documentation update.
187
188		----------------------------------------------------------------------
189		Name: Matthias Blume
190		Date: 2002/05/21 10:55:00 EDT
191		Tag: blume-20020521-misc
192		Description:
193
194		- John tweaked runtime to be silent on heap export (except when
195		GC messages are on).
196		- I added a few more things (cross-compiling versions of CMB) to
197		config/preloads (as suggestions).
198
199		----------------------------------------------------------------------
200		Name: Matthias Blume
201		Date: 2002/05/20 22:25:00 EDT
202		Tag: blume-20020520-controls
203		Description:
204
205		- Added ControlUtil structure to control-lib.cm.
206		- Use it throughout.
207		- Used Controls facility to define MLRISC controls (as opposed to
208		registering MLRISC control ref cells with Controls after the fact)
209		- Fixed messed-up controls priorities.
210
211		* Removed again all the stuff from config/preloads that one wouldn't
212		be able to preload at the time the initial heap image is built.
213		(Many libraries, e.g., CML, do not exist yet at this time.  The
214		only libraries that can be preloaded via config/preloads are those
215		that come bundled with the bootfiles.)
216
217		----------------------------------------------------------------------
218		Name: Matthias Blume
219		Date: 2002/05/20 10:59:00 EDT
220		Tag: blume-20020520-preloads
221		Description:
222
223		Added a lot of commented-out suggestions for things to be included
224		in config/preloads.
225
226		----------------------------------------------------------------------
227		Name: Allen Leung
228		Date: 2002/05/18 14:20:00 EDT
229		Tag: leunga-20020518-mdl
230		Description:
231
232		o Made the mdl tool stuff compile and run again.
233		o I've disabled all the stuff that depends on RTL specifications; they
234		are all badly broken anyway.
235
236		----------------------------------------------------------------------
237		Name: Matthias Blume
238		Date: 2002/05/17 16:49:00 EDT
239		Tag: blume-20020517-controls
240		Description:
241
242		0. John Reppy made several modifications to the SML/NJ library.
243		In particular, there is a shiny new controls-lib.cm.
244
245		1. Pushed new controls interface through compiler so that everything
246		compiles again.
247
248		2. Added FormatComb and FORMAT_COMB to the CML version of the
249		SML/NJ library (so that CML compiles again).
250
251		3. Modified init scripts because XXX_DEFAULT environment variables
252		are no longer with us.  (Boot-time initialization is now done
253		using the same environment variables that are also used for
254		startup-time initialization of controls.)
255
256		----------------------------------------------------------------------
257		Name: Lal George
258		Date: 2002/05/15 09:20:10 EDT
259		Tag: george-20020515-pseudo-op-decls
260		Description:
261
262		All pseudo-ops emitted before the first segment declaration
263		such as TEXT, DATA, and BSS directives are assumed to be global
264		declarations and are emitted first in the assembly file. This is
265		useful in a number of situations where one has pseudo-ops that are not
266		specific to any segment, and also works around the constraint that one
267		cannot have client pseudo-ops in the TEXT segment.
268
269		Because no segment is associated with these declarations it is
270		an error to allocate any space or objects before the first segment
271		directive and an exception will be raised. However, we cannot make
272		this check for client pseudo-ops.
273
274		These top level declarations are a field in the CFG graph_info.
275		In theory you can continue to add to this field after the CFG has been
276		built -- provided you know what you are doing;-)
277
278		----------------------------------------------------------------------
279		Name: Matthias Blume
280		Date: 2002/05/13 16:40:00 EDT
281		Tag: blume-20020513-pp-etc
282		Description:
283
284		A few minor bugfixes:
285
286		- Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
287		(Bogus printouts for redefined bindings still occur.  Compiler
288		bug should no longer occur now.  We need to redo the prettyprinter
289		from scratch.)
290
291		- CM pathname printer now also adds escape sequences for ( and )
292
293		- commend and docu fixes for ml-nlffi
294
295		----------------------------------------------------------------------
296		Name: Matthias Blume
297		Date: 2002/05/10 16:40:00 EDT
298		Tag: blume-20020510-erg-textio
299		Description:
300
301		Applied the following bugfix provided by Emden Gansner:
302
303		Output is corrupted when outputSubstr is used rather than output.
304
305		The problem occurs when a substring
306
307		ss = (s, dataStart, dataLen)
308
309		where dataStart > 0, fills a stream buffer with avail bytes left.
310		avail bytes of s, starting at index dataStart, are copied into the
311		buffer, the buffer is flushed, and then the remaining dataLen-avail
312		bytes of ss are copied into the beginning of the buffer. Instead of
313		starting this copy at index dataStart+avail in s, the current code
314		starts the copy at index avail.
315
316		Fix:
317		In text-io-fn.sml, change line 695 from
318		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
319		to
320		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
321
322		----------------------------------------------------------------------
323		Name: Matthias Blume
324		Date: 2002/04/12 13:55:00 EDT
325		Tag: blume-20020412-assyntax
326		Description:
327
328		1. Grabbed newer assyntax.h from the XFree86 project.
329		2. Fiddled with how to compile X86.prim.asm without warnings.
330		3. (Very) Minor cleanup in CM.
331
332		----------------------------------------------------------------------
333		Name: Matthias Blume
334		Date: 2002/04/01 (no joke!) 17:07:00 EST
335		Tag: blume-20020401-x86div
336		Description:
337
338		Added full support for div/mod/rem/quot on the x86, using the machine
339		instruction's two results (without clumsily recomputing the remainder)
340		directly where appropriate.
341
342		Some more extensive power-of-two support was added to the x86 instruction
343		selector (avoiding expensive divs, mods, and muls where they can be
344		replaced with cheaper shifts and masks).  However, this sort of thing
345		ought to be done earlier, e.g., within the CPS optimizer so that
346		all architectures benefit from it.
347
348		The compiler compiles to a fixed point, but changes might be somewhat
349		fragile nevertheless.  Please, report any strange things that you might
350		see wrt. div/mod/quot/rem...
351
352		----------------------------------------------------------------------
353		Name: Matthias Blume
354		Date: 2002/03/29 17:22:00
355		Tag: blume-20020329-div
356		Description:
357
358		Fixed my broken div/mod logic.  Unfortunately, this means that the
359		inline code for div/mod now has one more comparison than before.
360		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
361		The problem was with quotient = 0, because that alone does not tell
362		us which way the rounding went.  One then has to look at whether
363		remainder and divisor have the same sign...  :(
364
365		Anyway, I replaced the bootfiles with fresh ones...
366
367		----------------------------------------------------------------------
368		Name: Matthias Blume
369		Date: 2002/03/29 14:10:00 EST
370		Tag: blume-20020329-inlprims
371		Description:
372
373		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
374
375		Primops have changed. This means that the bin/boot-file formats have
376		changed as well.
377
378		To make sure that there is no confusion, I made a new version.
379
380
381		CHANGES:
382
383		* removed REMT from mltree (remainder should never overflow).
384
385		* added primops to deal with divisions of all flavors to the frontend
386
387		* handled these primops all the way through so they map to their respective
388		MLRISC support
389
390		* used these primops in the implementation of Int, Int32, Word, Word32
391
392		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
393
394		* parameterized INLMIN, INLMAX, and INLABS by a numkind
395
396		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
397		floating point
398
399		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
400		and Real (but Real.abs maps to a separate floating-point-only primop)
401
402
403		TODO items:
404
405		* Hacked Alpha32 instruction selection, disabling the selection of REMx
406		instructions because the machine instruction encoder cannot handle
407		them.  (Hppa, PPC, and Sparc instruction selection did not handle
408		REM in the first place, and REM is supported by the x86 machine coder.)
409
410		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
411		selection phase.  (The two can be streamlined because the hardware
412		delivers both quotient and remainder at the same time anyway.)
413
414		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
415		(Currently the behavior is inconsistent both across architectures and
416		wrt. the draft Basis spec.)
417
418		* Word8 should eventually be handled natively, too.
419
420		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
421		as if there currently is no execution path that could trigger it in
422		SML/NJ.  (The assumptions underlying functions arith and promotable do not
423		hold for things like multiplication and division.)
424
425		----------------------------------------------------------------------
426		Name: Matthias Blume
427		Date: 2002/03/27 16:27:00 EST
428		Tag: blume-20020327-mlrisc-divisions
429		Description:
430
431		Added support for all four division operations (ML's div, mod, quot,
432		and rem) to MLRISC.  In the course of doing so, I also rationalized
433		the naming (no more annoying switch-around of DIV and QUOT), by
434		parameterizing the operation by div_rounding_mode (which can be either
435		DIV_TO_ZERO or DIV_TO_NEGINF).
436
437		The generic MLTreeGen functor takes care of compiling all four
438		operations down to only round-to-zero div.
439
440		Missing pieces:
441
442		* Doing something smarter than relying on MLTreeGen on architectures
443		like, e.g., the x86 where hardware division delivers both quotient and
444		remainder at the same time.  With this, the implementation of the
445		round-to-neginf operations could be further streamlined.
446
447		* Remove inlining support for div/mod/rem from the frontend and replace it
448		with primops that get carried through to the backend.  Do this for all
449		int and word types.
450
451		----------------------------------------------------------------------
452		Name: Matthias Blume
453		Date: 2002/03/25 17:25:00 EST
454		Tag: blume-20020325-divmod
455		Description:
456
457		I improved (hopefully without breaking them) the implementation of Int.div,
458		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
459		advantage of the following observations:
460
461		Let  q = x quot y      r = x rem y
462		d = x div  y      m = x mod y
463
464		where "quot" is the round-to-zero version of integer division that
465		hardware usually provides.  Then we have:
466
467		r = x - q * y        where neither the * nor the - will overflow
468		d = if q >= 0 orelse x = q * y then q else q - 1
469		where neither the * nor the - will overflow
470		m = if q >= 0 orelse r = 0 then r else r + y
471		where the + will not overflow
472
473		This results in substantial simplification of the generated code.
474		The following table shows the number of CFG nodes and edges generated
475		for
476		fun f (x, y) = x OPER y
477		(* with OPER \in div, mod, quot, rem *)
478
479
480		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
481		--------------------------------------------------------
482		div |         24 |         39 |         12 |         16
483		mod |         41 |         71 |         12 |         16
484		quot |          8 |         10 |          8 |         10
485		rem |         10 |         14 |          8 |         10
486
487
488		----------------------------------------------------------------------
489		Name: Matthias Blume
490		Date: 2002/03/25 22:06:00 EST
491		Tag: blume-20020325-cprotobug
492		Description:
493
494		Fixed a bug in cproto (c prototype decoder).
495
496		----------------------------------------------------------------------
497		Name: Matthias Blume
498		Date: 2002/03/25 16:00:00 EST
499		Tag: blume-20020325-raw-primops
500		Description:
501
502		I did some cleanup to Allen's new primop code and
503		replaced yesterday's bootfiles with new ones.
504		(But they are stored in the same place.)
505
506		----------------------------------------------------------------------
507		Name: Matthias Blume
508		Date: 2002/03/24 22:40:00 EST
509		Tag: blume-20020324-bootfiles
510		Description:
511
512		Made the bootfiles that Allen asked for.
513
514		----------------------------------------------------------------------
515		Name: Allen Leung
516		Date: 2002/03/23 15:50:00 EST
517		Tag: leunga-20020323-flint-cps-rcc-primops
518		Description:
519
520		1. Changes to FLINT primops:
521
522		(* make a call to a C-function;
523		* The primop carries C function prototype information and specifies
524		* which of its (ML-) arguments are floating point. C prototype
525		* information is for use by the backend, ML information is for
526		* use by the CPS converter. *)
527		| RAW_CCALL of { c_proto: CTypes.c_proto,
528		ml_args: ccall_type list,
529		ml_res_opt: ccall_type option,
530		reentrant : bool
531		} option
532		(* Allocate uninitialized storage on the heap.
533		* The record is meant to hold short-lived C objects, i.e., they
534		* are not ML pointers.  With the tag, the representation is
535		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
536		*)
537		| RAW_RECORD of {tag:bool,sz:int}
538		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
539
540		2.  These CPS primops are now overloaded:
541
542		rawload of {kind:numkind}
543		rawstore of {kind:numkind}
544
545		The one argument form is:
546
547		rawload {kind} address
548
549		The two argument form is:
550
551		rawload {kind} [ml object, byte-offset]
552
553		3. RAW_CCALL/RCC now takes two extra arguments:
554
555		a. The first is whether the C call is reentrant, i.e., whether
556		ML state should be saved and restored.
557		b. The second argument is a string argument specifying the name of
558		library and the C function.
559
560		These things are currently not handled in the code generator, yet.
561
562		4. In CProto,
563
564		An encoding type of "bool" means "ml object" and is mapped into
565		C prototype of PTR.  Note that "bool" is different than "string",
566		even though "string" is also mapped into PTR, because "bool"
567		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
568
569		5. Pickler/unpicker
570
571		Changed to handle RAW_RECORD and newest RAW_CCALL
572
573		6. MLRiscGen,
574
575		1. Changed to handle the new rawload/rawstore/rawrecord operators.
576		2. Code for handling C Calls has been moved to a new module CPSCCalls,
577		in the file CodeGen/cpscompile/cps-c-calls.sml
578
579		7. Added the conditional move operator
580
581		condmove of branch
582
583		to cps.  Generation of this is still buggy so it is currently
584		disabled.
585
586		----------------------------------------------------------------------
587		Name: Lal George
588		Date: 2002/03/22 14:18:25 EST
589		Tag: george-20020322-cps-branch-prob
590		Description:
591
592		Implemented the Ball-Larus branch prediction-heuristics, and
593		incorporated graphical viewers for control flow graphs.
594
595		Ball-Larus Heuristics:
596		---------------------
597		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
598
599		By design it uses the Dempster-Shafer theory for combining
600		probabilities.  For example, in the function:
601
602		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
603
604		the ball-larus heuristics predicts that the n=0 is unlikely
605		(OH-heuristic), and the 'then' branch is unlikely because of the
606		RH-heuristic -- giving the 'then' branch an even lower combined
607		probability using the Dempster-Shafer theory.
608
609		Finally, John Reppy's loop analysis in MLRISC, further lowers the
610		probability of the 'then' branch because of the loop in the else
611		branch.
612
613
614		Graphical Viewing:
615		------------------
616		I merely plugged in Allen's graphical viewers into the compiler. The
617		additional code is not much. At the top level, saying:
618
619		Control.MLRISC.getFlag "cfg-graphical-view" := true;
620
621		will display the graphical view of the control flow graph just before
622		back-patching.  daVinci must be in your path for this to work. If
623		daVinci is not available, then the default viewer can be changed
624		using:
625
626		Control.MLRISC.getString "viewer"
627
628		which can be set to "dot" or "vcg" for the corresponding viewers. Of
629		course, these viewers must be in your path.
630
631		The above will display the compilation unit at the level of clusters,
632		many of which are small, boring, and un-interesting. Also setting:
633
634		Control.MLRISC.getInt "cfg-graphical-view_size"
635
636		will display clusters that are larger than the value set by the above.
637
638
639		----------------------------------------------------------------------
640		Name: Matthias Blume
641		Date: 2002/03/21 22:20:00 EST
642		Tag: blume-20020321-kmp-bugfix
643		Description:
644
645		Changed the interface to the KMP routine in PreString and fixed
646		a minor bug in one place where it was used.
647
648		----------------------------------------------------------------------
649		Name: Allen Leung
650		Date: 2002/03/21 20:30:00 EST
651		Tag: leunga-20020321-cfg
652		Description:
653
654		Fixed a potential problem in cfg edge splitting.
655
656		----------------------------------------------------------------------
657		Name: Allen Leung
658		Date: 2002/03/21 17:15:00 EST
659		Tag: leunga-20020321-x86-fp-cfg
660		Description:
661
662		1. Recoded the buggy parts of x86-fp.
663
664		a. All the block reordering code has been removed.
665		We now depend on the block placement phases to do this work.
666
667		b. Critical edge splitting code has been simplified and moved into the
668		CFG modules, as where they belong.
669
670		Both of these were quite buggy and complex.  The code is now much, much
671		simpler.
672
673		2. X86 backend.
674
675		a. Added instructions for 64-bit support.  Instruction selection for
676		64-bit has not been committed, however, since that
677		requires changes to MLTREE which haven't been approved by
678		Lal and John.
679
680		b. Added support for FUCOMI and FUCOMIP when generating code for
681		PentiumPro and above.  We only generate these instructions in
682		the fast-fp mode.
683
684		c. Added cases for JP and JNP in X86FreqProps.
685
686		3. CFG
687
688		CFG now has a bunch of methods for edge splitting and merging.
689
690		4. Machine description.
691
692		John's simplification of MLTREE_BASIS.fcond broke a few machine
693		description things:
694
695		rtl-build.{sig,sml} and hppa.mdl fixed.
696
697		NOTE: the machine description stuff in the repository is still broken.
698		Again, I can't put my fixes in because that involves
699		changes to MLTREE.
700
701		----------------------------------------------------------------------
702		Name: Matthias Blume
703		Date: 2002/03/20 15:55:00 EST
704		Tag: blume-20020320-kmp
705		Description:
706
707		Implemented Knuth-Morris-Pratt string matching in PreString and used
708		it for String.isSubstring, Substring.isSubstring, and
709		Substring.position.
710
711		(Might need some stress-testing.  Simple examples worked fine.)
712
713		----------------------------------------------------------------------
714		Name: Matthias Blume
715		Date: 2002/03/19 16:37:00 EST
716		Tag: blume-20020319-witnesses
717		Description:
718
719		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
720
721		This implements a generic mechanism for changing constness qualifiers
722		anywhere within big C types without resorting to outright "casts".
723		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
724		modify the constness at the outermost level.)
725		The implementation of "convert" is based on the idea of "witness"
726		values -- values that are not used by the operation but whose types
727		"testify" to their applicability.  On the implementation side, "convert"
728		is simply a projection (returning its second curried argument).  With
729		cross-module inlining, it should not result in any machine code being
730		generated.
731
732		----------------------------------------------------------------------
733		Name: Matthias Blume
734		Date: 2002/03/15 16:40:00 EST
735		Tag: blume-20020315-basis
736		Description:
737
738		Provided (preliminary?) implementations for
739
740		{String,Substring}.{concatWith,isSuffix,isSubstring}
741
742		and
743
744		Substring.full
745
746		Those are in the Basis spec but they were missing in SML/NJ.
747
748		----------------------------------------------------------------------
749		Name: Matthias Blume
750		Date: 2002/03/14 21:30:00 EST
751		Tag: blume-20020314-controls
752		Description:
753
754		Controls:
755		---------
756
757		1. Factored out the recently-added Controls : CONTROLS stuff and put
758		it into its own library $/controls-lib.cm.  The source tree for
759		this is under src/smlnj-lib/Controls.
760
761		2. Changed the names of types and functions in this interface, so they
762		make a bit more "sense":
763
764		module -> registry
765		'a registry -> 'a group
766
767		3. The interface now deals in ref cells only.  The getter/setter interface
768		is (mostly) gone.
769
770		4. Added a function that lets one register an already-existing ref cell.
771
772		5. Made the corresponding modifications to the rest of the code so that
773		everything compiles again.
774
775		6. Changed the implementation of Controls.MLRISC back to something closer
776		to the original.  In particular, this module (and therefore MLRISC)
777		does not depend on Controls.  There now is some link-time code in
778		int-sys.sml that registers the MLRISC controls with the Controls
779		module.
780
781		CM:
782		---
783
784		* One can now specify the lambda-split aggressiveness in init.cmi.
785
786		----------------------------------------------------------------------
787		Name: Allen Leung
788		Date: 2002/03/13 17:30:00 EST
789		Tag: leunga-20020313-x86-fp-unary
790		Description:
791
792		Bug fix for:
793
794		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
795		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
796		> - fun f(x,(y,z)) = Real.~ y;
797		> [autoloading]
798		> [autoloading done]
799		>       fchsl   (%eax), 184(%esp)
800		> Error: MLRisc bug: X86MCEmitter.emitInstr
801		>
802		> uncaught exception Error
803		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
804
805		The problem was that the code generator did not generate any fp registers
806		in this case, and the ra didn't know that it needed to run the X86FP phase to
807		translate the pseudo fp instruction.   This only happened with unary fp
808		operators in certain situations.
809
810		----------------------------------------------------------------------
811		Name: Matthias Blume
812		Date: 2002/03/13 14:00:00 EST
813		Tag: blume-20020313-overload-etc
814		Description:
815
816		1. Added _overload as a synonym for overload for backward compatibility.
817		(Control.overloadKW must be true for either version to be accepted.)
818
819		2. Fixed bug in install script that caused more things to be installed
820		than what was requested in config/targets.
821
822		3. Made CM aware of the (_)overload construct so that autoloading
823		works.
824
825		----------------------------------------------------------------------
826		Name: Matthias Blume
827		Date: 2002/03/12 22:03:00 EST
828		Tag: blume-20020312-url
829		Description:
830
831		Forgot to update BOOT and srcarchiveurl.
832
833		----------------------------------------------------------------------
834		Name: Matthias Blume
835		Date: 2002/03/12 17:30:00 EST
836		Tag: blume-20020312-version110392
837		Description:
838
839		Yet another version number bump (because of small changes to the
840		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
841
842		Changes:
843
844		The new pid generation scheme described a few weeks ago was overly
845		complicated.  I implemented a new mechanism that is simpler and
846		provides a bit more "stability":  Once CM has seen a compilation
847		unit, it keeps its identity constant (as long as you do not delete
848		those crucial CM/GUID/* files).  This means that when you change
849		an interface, compile, then go back to the old interface, and
850		compile again, you arrive at the original pid.
851
852		There now also is a mechanism that instructs CM to use the plain
853		environment hash as a module's pid (effectively making its GUID
854		the empty string).  For this, "noguid" must be specified as an
855		option to the .sml file in question within its .cm file.
856		This is most useful for code that is being generated by tools such
857		as ml-nlffigen (because during development programmers tend to
858		erase the tool's entire output directory tree including CM's cached
859		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
860		revert to the old, broken behavior of SML/NJ, but in specific cases
861		where there is no danger of interface confusion, its use is ok
862		(I think).
863
864		ml-nlffigen by default generates "noguid" annotations.  They can be
865		turned off by specifying -guid in its command line.
866
867		----------------------------------------------------------------------
868		Name: Lal George
869		Date: 2002/03/12 12 14:42:36 EST
870		Tag: george-20020312-frequency-computation
871		Description:
872
873		Integrated jump chaining and static block frequency into the
874		compiler. More details and numbers later.
875
876		----------------------------------------------------------------------
877		Name: Lal George
878		Date: 2002/03/11 11 22:38:53 EST
879		Tag: george-20020311-jump-chain-elim
880		Description:
881
882		Tested the jump chain elimination on all architectures (except the
883		hppa).  This is on by default right now and is profitable for the
884		alpha and x86, however, it may not be profitable for the sparc and ppc
885		when compiling the compiler.
886
887		The gc test will typically jump to a label at the end of the cluster,
888		where there is another jump to an external cluster containing the actual
889		code to invoke gc. This is to allow factoring of common gc invocation
890		sequences. That is to say, we generate:
891
892		f:
893		testgc
894		ja   L1      % jump if above to L1
895
896		L1:
897		jmp L2
898
899
900		After jump chain elimination the 'ja L1' instructions is converted to
901		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
902		up being implemented in their long form (if L2 is far away) using:
903
904		jbe     L3      % jump if below or equal to L3
905		jmp     L2
906		L3:
907		...
908
909
910		For large compilation units L2  may be far away.
911
912
913		----------------------------------------------------------------------
914		Name: Matthias Blume
915		Date: 2002/03/11 13:30:00 EST
916		Tag: blume-20020311-mltreeeval
917		Description:
918
919		A functor parameter was missing.
920
921		----------------------------------------------------------------------
922		Name: Allen Leung
923		Date: 2002/03/11 10:30:00 EST
924		Tag: leunga-20020311-runtime-string0
925		Description:
926
927		The representation of the empty string now points to a
928		legal null terminated C string instead of unit.  It is now possible
929		to convert an ML string into C string with InlineT.CharVector.getData.
930		This compiles into one single machine instruction.
931
932		----------------------------------------------------------------------
933		Name: Allen Leung
934		Date: 2002/03/10 23:55:00 EST
935		Tag: leunga-20020310-x86-call
936		Description:
937
938		Added machine generation for CALL instruction (relative displacement mode)
939
940		----------------------------------------------------------------------
941		Name: Matthias Blume
942		Date: 2002/03/08 16:05:00
943		Tag: blume-20020308-entrypoints
944		Description:
945
946		Version number bumped to 110.39.1.  NEW BOOTFILES!
947
948		Entrypoints: non-zero offset into a code object where execution should begin.
949
950		- Added the notion of an entrypoint to CodeObj.
951		- Added reading/writing of entrypoint info to Binfile.
952		- Made runtime system bootloader aware of entrypoints.
953		- Use the address of the label of the first function given to mlriscGen
954		as the entrypoint.  This address is currently always 0, but it will
955		not be 0 once we turn on block placement.
956		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
957		with entry points) from mlriscGen.
958
959		----------------------------------------------------------------------
960		Name: Allen Leung
961		Date: 2002/03/07 20:45:00 EST
962		Tag: leunga-20020307-x86-cmov
963		Description:
964
965		Bug fixes for CMOVcc on x86.
966
967		1. Added machine code generation for CMOVcc
968		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
969		3. CMOVcc cannot have an immediate operand as argument.
970
971		----------------------------------------------------------------------
972		Name: Matthias Blume
973		Date: 2002/03/07 16:15:00 EST
974		Tag: blume-20020307-controls
975		Description:
976
977		This is a very large but mostly boring patch which makes (almost)
978		every tuneable compiler knob (i.e., pretty much everything under
979		Control.* plus a few other things) configurable via both the command
980		line and environment variables in the style CM did its configuration
981		until now.
982
983		Try starting sml with '-h' (or, if you are brave, '-H')
984
985		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
986		implements the underlying generic mechanism.
987
988		The interface to some of the existing such facilities has changed somewhat.
989		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
990		(The getFoo interface is still there for backward-compatibility, but its
991		use is deprecated.)
992
993		The ml-build script passes -Cxxx=yyy command-line arguments through so
994		that one can now twiddle the compiler settings when using this "batch"
995		compiler.
996
997		TODO items:
998
999		We should go through and throw out all controls that are no longer
1000		connected to anything.  Moreover, we should go through and provide
1001		meaningful (and correct!) documentation strings for those controls
1002		that still are connected.
1003
1004		Currently, multiple calls to Controls.new are accepted (only the first
1005		has any effect).  Eventually we should make sure that every control
1006		is being made (via Controls.new) exactly once.  Future access can then
1007		be done using Controls.acc.
1008
1009		Finally, it would probably be a good idea to use the getter-setter
1010		interface to controls rather than ref cells.  For the time being, both
1011		styles are provided by the Controls module, but getter-setter pairs are
1012		better if thread-safety is of any concern because they can be wrapped.
1013
1014		*****************************************
1015
1016		One bug fix: The function blockPlacement in three of the MLRISC
1017		backpatch files used to be hard-wired to one of two possibilities at
1018		link time (according to the value of the placementFlag).  But (I
1019		think) it should rather sense the flag every time.
1020
1021		*****************************************
1022
1023		Other assorted changes (by other people who did not supply a HISTORY entry):
1024
1025		1. the cross-module inliner now works much better (Monnier)
1026		2. representation of weights, frequencies, and probabilities in MLRISC
1027		changed in preparation of using those for weighted block placement
1028		(Reppy, George)
1029
1030		----------------------------------------------------------------------
1031		Name: Lal George
1032		Date: 2002/03/07 14:44:24 EST 2002
1033		Tag: george-20020307-weighted-block-placement
1034
1035		Tested the weighted block placement optimization on all architectures
1036		(except the hppa) using AMPL to generate the block and edge frequencies.
1037		Changes were required in the machine properties to correctly
1038		categorize trap instructions. There is an MLRISC flag
1039		"weighted-block-placement" that can be used to enable weighted block
1040		placement, but this will be ineffective without block/edge
1041		frequencies (coming soon).
1042
1043
1044		----------------------------------------------------------------------
1045		Name: Lal George
1046		Date: 2002/03/05 17:24:48 EST
1047		Tag: george-20020305-linkage-cluster
1048
1049		In order to support the block placement optimization, a new cluster
1050		is generated as the very first cluster (called the linkage cluster).
1051		It contains a single jump to the 'real' entry point for the compilation
1052		unit. Block placement has no effect on the linkage cluster itself, but
1053		all the other clusters  have full freedom in the manner in which they
1054		reorder blocks or functions.
1055
1056		On the x86 the typical linkage code that is generated is:
1057		----------------------
1058		.align 2
1059		L0:
1060		addl    $L1-L0, 72(%esp)
1061		jmp     L1
1062
1063
1064		.align  2
1065		L1:
1066		----------------------
1067
1068		72(%esp) is the memory location for the stdlink register. This
1069		must contain the address of the CPS function being called. In the
1070		above example, it contains the address of  L0; before
1071		calling L1 (the real entry point for the compilation unit), it
1072		must contain the address for L1, and hence
1073
1074		addl $L1-L0, 72(%esp)
1075
1076		I have tested this on all architectures except the hppa.The increase
1077		in code size is of course negligible
1078
1079		----------------------------------------------------------------------
1080		Name: Allen Leung
1081		Date: 2002/03/03 13:20:00 EST
1082		Tag: leunga-20020303-mlrisc-tools
1083
1084		Added #[ ... ] expressions to mlrisc tools
1085
1086		----------------------------------------------------------------------
1087		Name: Matthias Blume
1088		Date: 2002/02/27 12:29:00 EST
1089		Tag: blume-20020227-cdebug
1090		Description:
1091
1092		- made types in structure C and C_Debug to be equal
1093		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
1094		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
1095
1096		----------------------------------------------------------------------
1097		Name: Matthias Blume
1098		Date: 2002/02/26 12:00:00 EST
1099		Tag: blume-20020226-ffi
1100		Description:
1101
1102		1. Fixed a minor bug in CM's "noweb" tool:
1103		If numbering is turned off, then truly don't number (i.e., do not
1104		supply the -L option to noweb).  The previous behavior was to supply
1105		-L'' -- which caused noweb to use the "default" line numbering scheme.
1106		Thanks to Chris Richards for pointing this out (and supplying the fix).
1107
1108		2. Once again, I reworked some aspects of the FFI:
1109
1110		A. The incomplete/complete type business:
1111
1112		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
1113		gone!
1114		- ML types representing an incomplete type are now *equal* to
1115		ML types representing their corresponding complete types (just like
1116		in C).  This is still safe because ml-nlffigen will not generate
1117		RTTI for incomplete types, nor will it generate functions that
1118		require access to such RTTI.   But when ML code generated from both
1119		incomplete and complete versions of the C type meet, the ML types
1120		are trivially interoperable.
1121
1122		NOTE:  These changes restore the full generality of the translation
1123		(which was previously lost when I eliminated functorization)!
1124
1125		B. Enum types:
1126
1127		- Structure C now has a type constructor "enum" that is similar to
1128		how the "su" constructor works.  However, "enum" is not a phantom
1129		type because each "T enum" has values (and is isomorphic to
1130		MLRep.Signed.int).
1131		- There are generic access operations for enum objects (using
1132		MLRep.Signed.int).
1133		- ml-nlffigen will generate a structure E_foo for each "enum foo".
1134		* The structure contains the definition of type "mlrep" (the ML-side
1135		representation type of the enum).  Normally, mlrep is the same
1136		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
1137		then mlrep will be defined as a datatype -- thus facilitating
1138		pattern matching on mlrep values.
1139		("-ec" will be suppressed if there are duplicate values in an
1140		enumeration.)
1141		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
1142		will be generated for each C enum constant xxx.
1143		* Conversion functions m2i and i2m convert between mlrep and
1144		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
1145		* Coversion functions c and ml convert between mlrep and "tag enum".
1146		* Access functions (get/set) fetch and store mlrep values.
1147		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
1148		enumerations are merged into one single enumeration represented by
1149		structure E_'.
1150
1151		----------------------------------------------------------------------
1152		Name: Allen Leung
1153		Date: 2002/02/25 04:45:00 EST
1154		Tag: leunga-20020225-cps-spill
1155
1156		This is a new implementation of the CPS spill phase.
1157		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1158		In case of problems, replace it with the old file spill.sml
1159
1160		The current compiler runs into some serious performance problems when
1161		constructing a large record.  This can happen when we try to compile a
1162		structure with many items.  Even a very simple structure like the following
1163		makes the compiler slow down.
1164
1165		structure Foo = struct
1166		val x_1 = 0w1 : Word32.int
1167		val x_2 = 0w2 : Word32.int
1168		val x_3 = 0w3 : Word32.int
1169		...
1170		val x_N = 0wN : Word32.int
1171		end
1172
1173		The following table shows the compile time, from N=1000 to N=4000,
1174		with the old compiler:
1175
1176		N
1177		1000   CPS 100 spill                           0.04u  0.00s  0.00g
1178		MLRISC ra                               0.06u  0.00s  0.05g
1179		(spills = 0 reloads = 0)
1180		TOTAL                                   0.63u  0.07s  0.21g
1181
1182		1100   CPS 100 spill                           8.25u  0.32s  0.64g
1183		MLRISC ra                               5.68u  0.59s  3.93g
1184		(spills = 0 reloads = 0)
1185		TOTAL                                   14.71u  0.99s  4.81g
1186
1187		1500   CPS 100 spill                           58.55u  2.34s  1.74g
1188		MLRISC ra                               5.54u  0.65s  3.91g
1189		(spills = 543 reloads = 1082)
1190		TOTAL                                   65.40u  3.13s  6.00g
1191
1192		2000   CPS 100 spill                           126.69u  4.84s  3.08g
1193		MLRISC ra                               0.80u  0.10s  0.55g
1194		(spills = 42 reloads = 84)
1195		TOTAL                                   129.42u  5.10s  4.13g
1196
1197		3000   CPS 100 spill                           675.59u  19.03s  11.64g
1198		MLRISC ra                               2.69u  0.27s  1.38g
1199		(spills = 62 reloads = 124)
1200		TOTAL                                   682.48u  19.61s  13.99g
1201
1202		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1203		MLRISC ra                               4.96u  0.27s  2.72g
1204		(spills = 85 reloads = 170)
1205		TOTAL                                   2375.26u  57.21s  48.00g
1206
1207		As you can see the old cps spill module suffers from some serious
1208		performance problem.  But since I cannot decipher the old code fully,
1209		instead of patching the problems up, I'm reimplementing it
1210		with a different algorithm.  The new code is more modular,
1211		smaller when compiled, and substantially faster
1212		(O(n log n) time and O(n) space).  Timing of the new spill module:
1213
1214		4000  CPS 100 spill                           0.02u  0.00s  0.00g
1215		MLRISC ra                               0.25u  0.02s  0.15g
1216		(spills=1 reloads=3)
1217		TOTAL                                   7.74u  0.34s  1.62g
1218
1219		Implementation details:
1220
1221		As far as I can tell, the purpose of the CPS spill module is to make sure the
1222		number of live variables at any program point (the bandwidth)
1223		does not exceed a certain limit, which is determined by the
1224		size of the spill area.
1225
1226		When the bandwidth is too large, we decrease the register pressure by
1227		packing live variables into spill records.  How we achieve this is
1228		completely different than what we did in the old code.
1229
1230		First, there is something about the MLRiscGen code generator
1231		that we should be aware of:
1232
1233		o MLRiscGen performs code motion!
1234
1235		In particular, it will move floating point computations and
1236		address computations involving only the heap pointer to
1237		their use sites (if there is only a single use).
1238		What this means is that if we have a CPS record construction
1239		statement
1240
1241		RECORD(k,vl,w,e)
1242
1243		we should never count the new record address w as live if w
1244		has only one use (which is often the case).
1245
1246		We should do something similar to floating point, but the transformation
1247		there is much more complex, so I won't deal with that.
1248
1249		Secondly, there are now two new cps primops at our disposal:
1250
1251		1. rawrecord of record_kind option
1252		This pure operator allocates some uninitialized storage from the heap.
1253		There are two forms:
1254
1255		rawrecord NONE [INT n]  allocates a tagless record of length n
1256		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1257		and initializes the tag.
1258
1259		2. rawupdate of cty
1260		rawupdate cty (v,i,x)
1261		Assigns to x to the ith component of record v.
1262		The storelist is not updated.
1263
1264		We use these new primops for both spilling and increment record construction.
1265
1266		1. Spilling.
1267
1268		This is implemented with a linear scan algorithm (but generalized
1269		to trees).  The algorithm will create a single spill record at the
1270		beginning of the cps function and use rawupdate to spill to it,
1271		and SELECT or SELp to reload from it.  So both spills and reloads
1272		are fine-grain operations.  In contrast, in the old algorithm
1273		"spills" have to be bundled together in records.
1274
1275		Ideally, we should sink the spill record construction to where
1276		it is needed.  We can even split the spill record into multiple ones
1277		at the places where they are needed.  But CPS is not a good
1278		representation for global code motion, so I'll keep it simple and
1279		am not attempting this.
1280
1281		2. Incremental record construction (aka record splitting).
1282
1283		Long records with many component values which are simulatenously live
1284		(recall that single use record addresses are not considered to
1285		be live) are constructed with rawrecord and rawupdate.
1286		We allocate space on the heap with rawrecord first, then gradually
1287		fill it in with rawupdate.  This is the technique suggested to me
1288		by Matthias.
1289
1290		Some restrictions on when this is applicable:
1291		1. It is not a VECTOR record.  The code generator currently does not handle
1292		this case. VECTOR record uses double indirection like arrays.
1293		2. All the record component values are defined in the same "basic block"
1294		as the record constructor.  This is to prevent speculative
1295		record construction.
1296
1297		----------------------------------------------------------------------
1298		Name: Allen Leung
1299		Date: 2002/02/22 01:02:00 EST
1300		Tag: leunga-20020222-mlrisc-tools
1301
1302		Minor bug fixes in the parser and rewriter
1303
1304		----------------------------------------------------------------------
1305		Name: Allen Leung
1306		Date: 2002/02/21 20:20:00 EST
1307		Tag: leunga-20020221-peephole
1308
1309		Regenerated the peephole files.  Some contained typos in the specification
1310		and some didn't compile because of pretty printing bugs in the old version
1311		of 'nowhere'.
1312
1313		----------------------------------------------------------------------
1314		Name: Allen Leung
1315		Date: 2002/02/19 20:20:00 EST
1316		Tag: leunga-20020219-mlrisc-tools
1317		Description:
1318
1319		Minor bug fixes to the mlrisc-tools library:
1320
1321		1.  Fixed up parsing colon suffixed keywords
1322		2.  Added the ability to shut the error messages up
1323		3.  Reimplemented the pretty printer and fixed up/improved
1324		the pretty printing of handle and -> types.
1325		4.  Fixed up generation of literal symbols in the nowhere tool.
1326		5.  Added some SML keywords to to sml.sty
1327
1328		----------------------------------------------------------------------
1329		Name: Matthias Blume
1330		Date: 2002/02/19 16:20:00 EST
1331		Tag: blume-20020219-cmffi
1332		Description:
1333
1334		A wild mix of changes, some minor, some major:
1335
1336		* All C FFI-related libraries are now anchored under $c:
1337		$/c.cm      --> $c/c.cm
1338		$/c-int.cm  --> $c/internals/c-int.cm
1339		$/memory.cm --> $c/memory/memory.cm
1340
1341		* "make" tool (in CM) now treats its argument pathname slightly
1342		differently:
1343		1. If the native expansion is an absolute name, then before invoking
1344		the "make" command on it, CM will apply OS.Path.mkRelative
1345		(with relativeTo = OS.FileSys.getDir()) to it.
1346		2. The argument will be passed through to subsequent phases of CM
1347		processing without "going native".  In particular, if the argument
1348		was an anchored path, then "make" will not lose track of that anchor.
1349
1350		* Compiler backends now "know" their respective C calling conventions
1351		instead of having to be told about it by ml-nlffigen.  This relieves
1352		ml-nlffigen from one of its burdens.
1353
1354		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1355
1356		* Export C_DEBUG and C_Debug from $c/c.cm.
1357
1358		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
1359		subtyping relationship between incomplete pointers and their complete
1360		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1361		with the convention of instantiating 'o with ('t, 'c) obj whenever
1362		the pointer target type is complete.  In the incomplete case, 'o
1363		will be instantiated with some "'c iobj" -- a type obtained by
1364		using one of the functors PointerToIncompleteType or PointerToCompleteType.
1365
1366		Operations that work on both incomplete and complete pointer types are
1367		typed as taking an 'o ptr while operations that require the target to
1368		be known are typed as taking some ('t, 'c) obj ptr.
1369
1370		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1371		where void is an eqtype without any values.  This makes it possible
1372		to work on voidptr values using functions meant to operate on light
1373		incomplete pointers.
1374
1375		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1376		been vastly simplified.
1377
1378		----------------------------------------------------------------------
1379		Name: Matthias Blume
1380		Date: 2002/02/19 10:48:00 EST
1381		Tag: blume-20020219-pqfix
1382		Description:
1383
1384		Applied Chris Okasaki's bug fix for priority queues.
1385
1386		----------------------------------------------------------------------
1387		Name: Matthias Blume
1388		Date: 2002/02/15 17:05:00
1389		Tag: Release_110_39
1390		Description:
1391
1392		Last-minute retagging is becoming a tradition... :-(
1393
1394		This is the working release 110.39.
1395
1396		----------------------------------------------------------------------
1397		Name: Matthias Blume
1398		Date: 2002/02/15 16:00:00 EST
1399		Tag: Release_110_39-orig
1400		Description:
1401
1402		Working release 110.39.  New bootfiles.
1403
1404		(Update: There was a small bug in the installer so it wouldn't work
1405		with all shells.  So I retagged. -Matthias)
1406
1407		----------------------------------------------------------------------
1408		Name: Matthias Blume
1409		Date: 2002/02/15 14:17:00 EST
1410		Tag: blume-20020215-showbindings
1411		Description:
1412
1413		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1414		the latter can be useful for exploring what bindings are available at
1415		the interactive prompt.  (The first function returns only the list
1416		of symbols that are really bound, the second prints those but also the
1417		ones that CM's autoloading mechanism knows about.)
1418
1419		----------------------------------------------------------------------
1420		Name: Matthias Blume
1421		Date: 2002/02/15 12:08:00 EST
1422		Tag: blume-20020215-iptrs
1423		Description:
1424
1425		Two improvements to ml-nlffigen:
1426
1427		1. Write files only if they do not exist or if their current contents
1428		do not coincide with what's being written.  (That is, avoid messing
1429		with the time stamps unless absolutely necessary.)
1430
1431		2. Implement a "repository" mechanism for generated files related
1432		to "incomplete pointer types".   See the README file for details.
1433
1434		----------------------------------------------------------------------
1435		Name: Matthias Blume
1436		Date: 2002/02/14 11:50:00 EST
1437		Tag: blume-20020214-quote
1438		Description:
1439
1440		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1441		because of the new and improved tag generation scheme.  (Thanks to Allen
1442		Leung for pointing it out.)
1443
1444		----------------------------------------------------------------------
1445		Name: Lal George
1446		Date: 2002/02/14 09:55:27 EST 2002
1447		Tag: george-20020214-isabelle-bug
1448		Description:
1449
1450		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1451		of Isabelle on the x86.
1452
1453		From Allen:
1454		-----------
1455		I've found the problem:
1456
1457		in ra-core.sml, I use the counter "blocked" to keep track of the
1458		true number of elements in the freeze queue.  When the counter goes
1459		to zero, I skip examining the queue.  But I've messed up the
1460		bookkeeping in combine():
1461
1462		else ();
1463		case !ucol of
1464		PSEUDO => (if !cntv > 0 then
1465		(if !cntu > 0 then blocked := !blocked - 1 else ();
1466		^^^^^^^^^^^^^^^^^^^^^^^
1467		moveu := mergeMoveList(!movev, !moveu)
1468		)
1469		else ();
1470
1471		combine() is called to coalesce two nodes u and v.
1472		I think I was thinking that if the move counts of u and v are both
1473		greater than zero then after they are coalesced then one node is
1474		removed from the freeze queue.  Apparently I was thinking that
1475		both u and v are of low degree, but that's clearly not necessarily true.
1476
1477
1478		02/12/2002:
1479		Here's the patch.  HOL now compiles.
1480
1481		I don't know how this impact on performance (compile
1482		time or runtime).  This bug caused the RA (especially on the x86)
1483		to go thru the potential spill phase when there are still nodes on the
1484		freeze queue.
1485
1486
1487
1488
1489		----------------------------------------------------------------------
1490		Name: Matthias Blume
1491		Date: 2002/02/13 22:40:00 EST
1492		Tag: blume-20020213-fptr-rtti
1493		Description:
1494
1495		Fixed a bug in ml-nlffigen that was introduced with one of the previous
1496		updates.
1497
1498		----------------------------------------------------------------------
1499		Name: Matthias Blume
1500		Date: 2002/02/13 16:41:00 EST
1501		Tag: blume-20020213-cmlpq
1502		Description:
1503
1504		Added new priority queue export symbols (which have just been added to
1505		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1506		not compile and the installer would choke.)
1507
1508		----------------------------------------------------------------------
1509		Name: Matthias Blume
1510		Date: 2002/02/13 16:15:00 EST
1511		Tag: blume-20020213-various
1512		Description:
1513
1514		1. More tweaks to ml-nlffigen:
1515
1516		- better internal datastructures (resulting in slight speedup)
1517		- "-match" option requires exact match
1518		- "localized" gensym counters (untagged structs/unions nested within
1519		other structs/unions or within typedefs get a fresh counter; their
1520		tag will be prefixed by a concatenation of their parents' tags)
1521		- bug fixes (related to calculation of transitive closure of types
1522		to be included in the output)
1523
1524		2. Minor Basis updates:
1525
1526		- added implementations for List.collate and Option.app
1527
1528		----------------------------------------------------------------------
1529		Name: Matthias Blume
1530		Date: 2002/02/11 15:55:00 EST
1531		Tag: blume-20020211-gensym
1532		Description:
1533
1534		Added a "-gensym" option to command line of ml-nlffigen.  This can be
1535		used to specify a "stem" -- a string that is inserted in all "gensym'd"
1536		names (ML structure names that correspond to unnamed C structs, unions,
1537		and enums), so that separate runs of ml-nlffigen do not clash.
1538
1539		----------------------------------------------------------------------
1540		Name: Matthias Blume
1541		Date: 2002/02/11 12:05:00 EST
1542		Tag: blume-20020211-gensml
1543		Description:
1544
1545		A quick fix for a problem with GenSML (in the pgraph-util library):
1546		Make generation of toplevel "local" optional.  (Strictly speaking,
1547		signature definitions within "local" are not legal SML.)
1548
1549		Other than that: updates to INSTALL and cm/TODO.
1550
1551		----------------------------------------------------------------------
1552		Name: Matthias Blume
1553		Date: 2002/02/08 15:00:00 EST
1554		Tag: blume-20020208-uniquepid
1555		Description:
1556
1557		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1558
1559		1. The installer (config/install.sh) has gotten smarter:
1560
1561		- Configuration options are a bit easier to specify now
1562		(in config/targets).
1563		- Bug in recognizing .tar.bz2 files fixed.
1564		- Installer automatically resolves dependencies between
1565		configuration options (e.g., if you ask for eXene, you will
1566		also get cml -- regardless whether you asked for it or not).
1567		- Installer can run in "quieter mode" by setting the environment
1568		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1569		"completely silent", though.
1570		- Build HashCons library as part of smlnj-lib.
1571
1572		2. A new scheme for assigning persistent identifiers to compilation
1573		units (and, by extension, to types etc.) has been put into place.
1574		This fixes a long-standing bug where types and even dynamic values
1575		can get internally confused, thereby compromising type safety
1576		(abstraction) and dynamic correctness.  See
1577
1578		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1579
1580		for an example of how things could go wrong until now.
1581
1582		The downside of the new scheme is that pids are not quite as
1583		persistent as they used to be: CM will generate a fresh pid
1584		for every compilation unit that it thinks it sees for the first
1585		time.  That means that if you compile starting from a clean, fresh
1586		source tree at two different times, you end up with different
1587		binaries.
1588
1589		Cutoff recompilation, however, has not been compromised because
1590		CM keeps pid information in special caches between runs.
1591
1592		----------------------------------------------------------------------
1593	Name: Lal George	Name: Lal George
1594	Date: 2002/02/07 15:34:13 EST 2002	Date: 2002/02/07 15:34:13 EST 2002
1595	Tag: <none>	Tag: <none>

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1055   changed lines   Added in v.1291

---

root@smlnj-gforge.cs.uchicago.edu	ViewVC Help
Powered by ViewVC 1.0.0