Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1063, Wed Feb 13 21:41:53 2002 UTC	revision 1155, Wed Mar 20 20:52:51 2002 UTC
#	Line 14	Line 14
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16	Name: Matthias Blume	Name: Matthias Blume
17		Date: 2002/03/20 15:55:00 EST
18		Tag: blume-20020320-kmp
19		Description:
20
21		Implemented Knuth-Morris-Pratt string matching in PreString and used
22		it for String.isSubstring, Substring.isSubstring, and
23		Substring.position.
24
25		(Might need some stress-testing.  Simple examples worked fine.)
26
27		----------------------------------------------------------------------
28		Name: Matthias Blume
29		Date: 2002/03/19 16:37:00 EST
30		Tag: blume-20020319-witnesses
31		Description:
32
33		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
34
35		This implements a generic mechanism for changing constness qualifiers
36		anywhere within big C types without resorting to outright "casts".
37		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
38		modify the constness at the outermost level.)
39		The implementation of "convert" is based on the idea of "witness"
40		values -- values that are not used by the operation but whose types
41		"testify" to their applicability.  On the implementation side, "convert"
42		is simply a projection (returning its second curried argument).  With
43		cross-module inlining, it should not result in any machine code being
44		generated.
45
46		----------------------------------------------------------------------
47		Name: Matthias Blume
48		Date: 2002/03/15 16:40:00 EST
49		Tag: blume-20020315-basis
50		Description:
51
52		Provided (preliminary?) implementations for
53
54		{String,Substring}.{concatWith,isSuffix,isSubstring}
55
56		and
57
58		Substring.full
59
60		Those are in the Basis spec but they were missing in SML/NJ.
61
62		----------------------------------------------------------------------
63		Name: Matthias Blume
64		Date: 2002/03/14 21:30:00 EST
65		Tag: blume-20020314-controls
66		Description:
67
68		Controls:
69		---------
70
71		1. Factored out the recently-added Controls : CONTROLS stuff and put
72		it into its own library $/controls-lib.cm.  The source tree for
73		this is under src/smlnj-lib/Controls.
74
75		2. Changed the names of types and functions in this interface, so they
76		make a bit more "sense":
77
78		module -> registry
79		'a registry -> 'a group
80
81		3. The interface now deals in ref cells only.  The getter/setter interface
82		is (mostly) gone.
83
84		4. Added a function that lets one register an already-existing ref cell.
85
86		5. Made the corresponding modifications to the rest of the code so that
87		everything compiles again.
88
89		6. Changed the implementation of Controls.MLRISC back to something closer
90		to the original.  In particular, this module (and therefore MLRISC)
91		does not depend on Controls.  There now is some link-time code in
92		int-sys.sml that registers the MLRISC controls with the Controls
93		module.
94
95		CM:
96		---
97
98		* One can now specify the lambda-split aggressiveness in init.cmi.
99
100		----------------------------------------------------------------------
101		Name: Allen Leung
102		Date: 2002/03/13 17:30:00 EST
103		Tag: leunga-20020313-x86-fp-unary
104		Description:
105
106		Bug fix for:
107
108		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
109		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
110		> - fun f(x,(y,z)) = Real.~ y;
111		> [autoloading]
112		> [autoloading done]
113		>       fchsl   (%eax), 184(%esp)
114		> Error: MLRisc bug: X86MCEmitter.emitInstr
115		>
116		> uncaught exception Error
117		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
118
119		The problem was that the code generator did not generate any fp registers
120		in this case, and the ra didn't know that it needed to run the X86FP phase to
121		translate the pseudo fp instruction.   This only happened with unary fp
122		operators in certain situations.
123
124		----------------------------------------------------------------------
125		Name: Matthias Blume
126		Date: 2002/03/13 14:00:00 EST
127		Tag: blume-20020313-overload-etc
128		Description:
129
130		1. Added _overload as a synonym for overload for backward compatibility.
131		(Control.overloadKW must be true for either version to be accepted.)
132
133		2. Fixed bug in install script that caused more things to be installed
134		than what was requested in config/targets.
135
136		3. Made CM aware of the (_)overload construct so that autoloading
137		works.
138
139		----------------------------------------------------------------------
140		Name: Matthias Blume
141		Date: 2002/03/12 22:03:00 EST
142		Tag: blume-20020312-url
143		Description:
144
145		Forgot to update BOOT and srcarchiveurl.
146
147		----------------------------------------------------------------------
148		Name: Matthias Blume
149		Date: 2002/03/12 17:30:00 EST
150		Tag: blume-20020312-version110392
151		Description:
152
153		Yet another version number bump (because of small changes to the
154		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
155
156		Changes:
157
158		The new pid generation scheme described a few weeks ago was overly
159		complicated.  I implemented a new mechanism that is simpler and
160		provides a bit more "stability":  Once CM has seen a compilation
161		unit, it keeps its identity constant (as long as you do not delete
162		those crucial CM/GUID/* files).  This means that when you change
163		an interface, compile, then go back to the old interface, and
164		compile again, you arrive at the original pid.
165
166		There now also is a mechanism that instructs CM to use the plain
167		environment hash as a module's pid (effectively making its GUID
168		the empty string).  For this, "noguid" must be specified as an
169		option to the .sml file in question within its .cm file.
170		This is most useful for code that is being generated by tools such
171		as ml-nlffigen (because during development programmers tend to
172		erase the tool's entire output directory tree including CM's cached
173		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
174		revert to the old, broken behavior of SML/NJ, but in specific cases
175		where there is no danger of interface confusion, its use is ok
176		(I think).
177
178		ml-nlffigen by default generates "noguid" annotations.  They can be
179		turned off by specifying -guid in its command line.
180
181		----------------------------------------------------------------------
182		Name: Lal George
183		Date: 2002/03/12 12 14:42:36 EST
184		Tag: george-20020312-frequency-computation
185		Description:
186
187		Integrated jump chaining and static block frequency into the
188		compiler. More details and numbers later.
189
190		----------------------------------------------------------------------
191		Name: Lal George
192		Date: 2002/03/11 11 22:38:53 EST
193		Tag: george-20020311-jump-chain-elim
194		Description:
195
196		Tested the jump chain elimination on all architectures (except the
197		hppa).  This is on by default right now and is profitable for the
198		alpha and x86, however, it may not be profitable for the sparc and ppc
199		when compiling the compiler.
200
201		The gc test will typically jump to a label at the end of the cluster,
202		where there is another jump to an external cluster containing the actual
203		code to invoke gc. This is to allow factoring of common gc invocation
204		sequences. That is to say, we generate:
205
206		f:
207		testgc
208		ja   L1      % jump if above to L1
209
210		L1:
211		jmp L2
212
213
214		After jump chain elimination the 'ja L1' instructions is converted to
215		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
216		up being implemented in their long form (if L2 is far away) using:
217
218		jbe     L3      % jump if below or equal to L3
219		jmp     L2
220		L3:
221		...
222
223
224		For large compilation units L2  may be far away.
225
226
227		----------------------------------------------------------------------
228		Name: Matthias Blume
229		Date: 2002/03/11 13:30:00 EST
230		Tag: blume-20020311-mltreeeval
231		Description:
232
233		A functor parameter was missing.
234
235		----------------------------------------------------------------------
236		Name: Allen Leung
237		Date: 2002/03/11 10:30:00 EST
238		Tag: leunga-20020311-runtime-string0
239		Description:
240
241		The representation of the empty string now points to a
242		legal null terminated C string instead of unit.  It is now possible
243		to convert an ML string into C string with InlineT.CharVector.getData.
244		This compiles into one single machine instruction.
245
246		----------------------------------------------------------------------
247		Name: Allen Leung
248		Date: 2002/03/10 23:55:00 EST
249		Tag: leunga-20020310-x86-call
250		Description:
251
252		Added machine generation for CALL instruction (relative displacement mode)
253
254		----------------------------------------------------------------------
255		Name: Matthias Blume
256		Date: 2002/03/08 16:05:00
257		Tag: blume-20020308-entrypoints
258		Description:
259
260		Version number bumped to 110.39.1.  NEW BOOTFILES!
261
262		Entrypoints: non-zero offset into a code object where execution should begin.
263
264		- Added the notion of an entrypoint to CodeObj.
265		- Added reading/writing of entrypoint info to Binfile.
266		- Made runtime system bootloader aware of entrypoints.
267		- Use the address of the label of the first function given to mlriscGen
268		as the entrypoint.  This address is currently always 0, but it will
269		not be 0 once we turn on block placement.
270		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
271		with entry points) from mlriscGen.
272
273		----------------------------------------------------------------------
274		Name: Allen Leung
275		Date: 2002/03/07 20:45:00 EST
276		Tag: leunga-20020307-x86-cmov
277		Description:
278
279		Bug fixes for CMOVcc on x86.
280
281		1. Added machine code generation for CMOVcc
282		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
283		3. CMOVcc cannot have an immediate operand as argument.
284
285		----------------------------------------------------------------------
286		Name: Matthias Blume
287		Date: 2002/03/07 16:15:00 EST
288		Tag: blume-20020307-controls
289		Description:
290
291		This is a very large but mostly boring patch which makes (almost)
292		every tuneable compiler knob (i.e., pretty much everything under
293		Control.* plus a few other things) configurable via both the command
294		line and environment variables in the style CM did its configuration
295		until now.
296
297		Try starting sml with '-h' (or, if you are brave, '-H')
298
299		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
300		implements the underlying generic mechanism.
301
302		The interface to some of the existing such facilities has changed somewhat.
303		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
304		(The getFoo interface is still there for backward-compatibility, but its
305		use is deprecated.)
306
307		The ml-build script passes -Cxxx=yyy command-line arguments through so
308		that one can now twiddle the compiler settings when using this "batch"
309		compiler.
310
311		TODO items:
312
313		We should go through and throw out all controls that are no longer
314		connected to anything.  Moreover, we should go through and provide
315		meaningful (and correct!) documentation strings for those controls
316		that still are connected.
317
318		Currently, multiple calls to Controls.new are accepted (only the first
319		has any effect).  Eventually we should make sure that every control
320		is being made (via Controls.new) exactly once.  Future access can then
321		be done using Controls.acc.
322
323		Finally, it would probably be a good idea to use the getter-setter
324		interface to controls rather than ref cells.  For the time being, both
325		styles are provided by the Controls module, but getter-setter pairs are
326		better if thread-safety is of any concern because they can be wrapped.
327
328		*****************************************
329
330		One bug fix: The function blockPlacement in three of the MLRISC
331		backpatch files used to be hard-wired to one of two possibilities at
332		link time (according to the value of the placementFlag).  But (I
333		think) it should rather sense the flag every time.
334
335		*****************************************
336
337		Other assorted changes (by other people who did not supply a HISTORY entry):
338
339		1. the cross-module inliner now works much better (Monnier)
340		2. representation of weights, frequencies, and probabilities in MLRISC
341		changed in preparation of using those for weighted block placement
342		(Reppy, George)
343
344		----------------------------------------------------------------------
345		Name: Lal George
346		Date: 2002/03/07 14:44:24 EST 2002
347		Tag: george-20020307-weighted-block-placement
348
349		Tested the weighted block placement optimization on all architectures
350		(except the hppa) using AMPL to generate the block and edge frequencies.
351		Changes were required in the machine properties to correctly
352		categorize trap instructions. There is an MLRISC flag
353		"weighted-block-placement" that can be used to enable weighted block
354		placement, but this will be ineffective without block/edge
355		frequencies (coming soon).
356
357
358		----------------------------------------------------------------------
359		Name: Lal George
360		Date: 2002/03/05 17:24:48 EST
361		Tag: george-20020305-linkage-cluster
362
363		In order to support the block placement optimization, a new cluster
364		is generated as the very first cluster (called the linkage cluster).
365		It contains a single jump to the 'real' entry point for the compilation
366		unit. Block placement has no effect on the linkage cluster itself, but
367		all the other clusters  have full freedom in the manner in which they
368		reorder blocks or functions.
369
370		On the x86 the typical linkage code that is generated is:
371		----------------------
372		.align 2
373		L0:
374		addl    $L1-L0, 72(%esp)
375		jmp     L1
376
377
378		.align  2
379		L1:
380		----------------------
381
382		72(%esp) is the memory location for the stdlink register. This
383		must contain the address of the CPS function being called. In the
384		above example, it contains the address of  L0; before
385		calling L1 (the real entry point for the compilation unit), it
386		must contain the address for L1, and hence
387
388		addl $L1-L0, 72(%esp)
389
390		I have tested this on all architectures except the hppa.The increase
391		in code size is of course negligible
392
393		----------------------------------------------------------------------
394		Name: Allen Leung
395		Date: 2002/03/03 13:20:00 EST
396		Tag: leunga-20020303-mlrisc-tools
397
398		Added #[ ... ] expressions to mlrisc tools
399
400		----------------------------------------------------------------------
401		Name: Matthias Blume
402		Date: 2002/02/27 12:29:00 EST
403		Tag: blume-20020227-cdebug
404		Description:
405
406		- made types in structure C and C_Debug to be equal
407		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
408		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
409
410		----------------------------------------------------------------------
411		Name: Matthias Blume
412		Date: 2002/02/26 12:00:00 EST
413		Tag: blume-20020226-ffi
414		Description:
415
416		1. Fixed a minor bug in CM's "noweb" tool:
417		If numbering is turned off, then truly don't number (i.e., do not
418		supply the -L option to noweb).  The previous behavior was to supply
419		-L'' -- which caused noweb to use the "default" line numbering scheme.
420		Thanks to Chris Richards for pointing this out (and supplying the fix).
421
422		2. Once again, I reworked some aspects of the FFI:
423
424		A. The incomplete/complete type business:
425
426		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
427		gone!
428		- ML types representing an incomplete type are now *equal* to
429		ML types representing their corresponding complete types (just like
430		in C).  This is still safe because ml-nlffigen will not generate
431		RTTI for incomplete types, nor will it generate functions that
432		require access to such RTTI.   But when ML code generated from both
433		incomplete and complete versions of the C type meet, the ML types
434		are trivially interoperable.
435
436		NOTE:  These changes restore the full generality of the translation
437		(which was previously lost when I eliminated functorization)!
438
439		B. Enum types:
440
441		- Structure C now has a type constructor "enum" that is similar to
442		how the "su" constructor works.  However, "enum" is not a phantom
443		type because each "T enum" has values (and is isomorphic to
444		MLRep.Signed.int).
445		- There are generic access operations for enum objects (using
446		MLRep.Signed.int).
447		- ml-nlffigen will generate a structure E_foo for each "enum foo".
448		* The structure contains the definition of type "mlrep" (the ML-side
449		representation type of the enum).  Normally, mlrep is the same
450		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
451		then mlrep will be defined as a datatype -- thus facilitating
452		pattern matching on mlrep values.
453		("-ec" will be suppressed if there are duplicate values in an
454		enumeration.)
455		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
456		will be generated for each C enum constant xxx.
457		* Conversion functions m2i and i2m convert between mlrep and
458		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
459		* Coversion functions c and ml convert between mlrep and "tag enum".
460		* Access functions (get/set) fetch and store mlrep values.
461		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
462		enumerations are merged into one single enumeration represented by
463		structure E_'.
464
465		----------------------------------------------------------------------
466		Name: Allen Leung
467		Date: 2002/02/25 04:45:00 EST
468		Tag: leunga-20020225-cps-spill
469
470		This is a new implementation of the CPS spill phase.
471		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
472		In case of problems, replace it with the old file spill.sml
473
474		The current compiler runs into some serious performance problems when
475		constructing a large record.  This can happen when we try to compile a
476		structure with many items.  Even a very simple structure like the following
477		makes the compiler slow down.
478
479		structure Foo = struct
480		val x_1 = 0w1 : Word32.int
481		val x_2 = 0w2 : Word32.int
482		val x_3 = 0w3 : Word32.int
483		...
484		val x_N = 0wN : Word32.int
485		end
486
487		The following table shows the compile time, from N=1000 to N=4000,
488		with the old compiler:
489
490		N
491		1000   CPS 100 spill                           0.04u  0.00s  0.00g
492		MLRISC ra                               0.06u  0.00s  0.05g
493		(spills = 0 reloads = 0)
494		TOTAL                                   0.63u  0.07s  0.21g
495
496		1100   CPS 100 spill                           8.25u  0.32s  0.64g
497		MLRISC ra                               5.68u  0.59s  3.93g
498		(spills = 0 reloads = 0)
499		TOTAL                                   14.71u  0.99s  4.81g
500
501		1500   CPS 100 spill                           58.55u  2.34s  1.74g
502		MLRISC ra                               5.54u  0.65s  3.91g
503		(spills = 543 reloads = 1082)
504		TOTAL                                   65.40u  3.13s  6.00g
505
506		2000   CPS 100 spill                           126.69u  4.84s  3.08g
507		MLRISC ra                               0.80u  0.10s  0.55g
508		(spills = 42 reloads = 84)
509		TOTAL                                   129.42u  5.10s  4.13g
510
511		3000   CPS 100 spill                           675.59u  19.03s  11.64g
512		MLRISC ra                               2.69u  0.27s  1.38g
513		(spills = 62 reloads = 124)
514		TOTAL                                   682.48u  19.61s  13.99g
515
516		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
517		MLRISC ra                               4.96u  0.27s  2.72g
518		(spills = 85 reloads = 170)
519		TOTAL                                   2375.26u  57.21s  48.00g
520
521		As you can see the old cps spill module suffers from some serious
522		performance problem.  But since I cannot decipher the old code fully,
523		instead of patching the problems up, I'm reimplementing it
524		with a different algorithm.  The new code is more modular,
525		smaller when compiled, and substantially faster
526		(O(n log n) time and O(n) space).  Timing of the new spill module:
527
528		4000  CPS 100 spill                           0.02u  0.00s  0.00g
529		MLRISC ra                               0.25u  0.02s  0.15g
530		(spills=1 reloads=3)
531		TOTAL                                   7.74u  0.34s  1.62g
532
533		Implementation details:
534
535		As far as I can tell, the purpose of the CPS spill module is to make sure the
536		number of live variables at any program point (the bandwidth)
537		does not exceed a certain limit, which is determined by the
538		size of the spill area.
539
540		When the bandwidth is too large, we decrease the register pressure by
541		packing live variables into spill records.  How we achieve this is
542		completely different than what we did in the old code.
543
544		First, there is something about the MLRiscGen code generator
545		that we should be aware of:
546
547		o MLRiscGen performs code motion!
548
549		In particular, it will move floating point computations and
550		address computations involving only the heap pointer to
551		their use sites (if there is only a single use).
552		What this means is that if we have a CPS record construction
553		statement
554
555		RECORD(k,vl,w,e)
556
557		we should never count the new record address w as live if w
558		has only one use (which is often the case).
559
560		We should do something similar to floating point, but the transformation
561		there is much more complex, so I won't deal with that.
562
563		Secondly, there are now two new cps primops at our disposal:
564
565		1. rawrecord of record_kind option
566		This pure operator allocates some uninitialized storage from the heap.
567		There are two forms:
568
569		rawrecord NONE [INT n]  allocates a tagless record of length n
570		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
571		and initializes the tag.
572
573		2. rawupdate of cty
574		rawupdate cty (v,i,x)
575		Assigns to x to the ith component of record v.
576		The storelist is not updated.
577
578		We use these new primops for both spilling and increment record construction.
579
580		1. Spilling.
581
582		This is implemented with a linear scan algorithm (but generalized
583		to trees).  The algorithm will create a single spill record at the
584		beginning of the cps function and use rawupdate to spill to it,
585		and SELECT or SELp to reload from it.  So both spills and reloads
586		are fine-grain operations.  In contrast, in the old algorithm
587		"spills" have to be bundled together in records.
588
589		Ideally, we should sink the spill record construction to where
590		it is needed.  We can even split the spill record into multiple ones
591		at the places where they are needed.  But CPS is not a good
592		representation for global code motion, so I'll keep it simple and
593		am not attempting this.
594
595		2. Incremental record construction (aka record splitting).
596
597		Long records with many component values which are simulatenously live
598		(recall that single use record addresses are not considered to
599		be live) are constructed with rawrecord and rawupdate.
600		We allocate space on the heap with rawrecord first, then gradually
601		fill it in with rawupdate.  This is the technique suggested to me
602		by Matthias.
603
604		Some restrictions on when this is applicable:
605		1. It is not a VECTOR record.  The code generator currently does not handle
606		this case. VECTOR record uses double indirection like arrays.
607		2. All the record component values are defined in the same "basic block"
608		as the record constructor.  This is to prevent speculative
609		record construction.
610
611		----------------------------------------------------------------------
612		Name: Allen Leung
613		Date: 2002/02/22 01:02:00 EST
614		Tag: leunga-20020222-mlrisc-tools
615
616		Minor bug fixes in the parser and rewriter
617
618		----------------------------------------------------------------------
619		Name: Allen Leung
620		Date: 2002/02/21 20:20:00 EST
621		Tag: leunga-20020221-peephole
622
623		Regenerated the peephole files.  Some contained typos in the specification
624		and some didn't compile because of pretty printing bugs in the old version
625		of 'nowhere'.
626
627		----------------------------------------------------------------------
628		Name: Allen Leung
629		Date: 2002/02/19 20:20:00 EST
630		Tag: leunga-20020219-mlrisc-tools
631		Description:
632
633		Minor bug fixes to the mlrisc-tools library:
634
635		1.  Fixed up parsing colon suffixed keywords
636		2.  Added the ability to shut the error messages up
637		3.  Reimplemented the pretty printer and fixed up/improved
638		the pretty printing of handle and -> types.
639		4.  Fixed up generation of literal symbols in the nowhere tool.
640		5.  Added some SML keywords to to sml.sty
641
642		----------------------------------------------------------------------
643		Name: Matthias Blume
644		Date: 2002/02/19 16:20:00 EST
645		Tag: blume-20020219-cmffi
646		Description:
647
648		A wild mix of changes, some minor, some major:
649
650		* All C FFI-related libraries are now anchored under $c:
651		$/c.cm      --> $c/c.cm
652		$/c-int.cm  --> $c/internals/c-int.cm
653		$/memory.cm --> $c/memory/memory.cm
654
655		* "make" tool (in CM) now treats its argument pathname slightly
656		differently:
657		1. If the native expansion is an absolute name, then before invoking
658		the "make" command on it, CM will apply OS.Path.mkRelative
659		(with relativeTo = OS.FileSys.getDir()) to it.
660		2. The argument will be passed through to subsequent phases of CM
661		processing without "going native".  In particular, if the argument
662		was an anchored path, then "make" will not lose track of that anchor.
663
664		* Compiler backends now "know" their respective C calling conventions
665		instead of having to be told about it by ml-nlffigen.  This relieves
666		ml-nlffigen from one of its burdens.
667
668		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
669
670		* Export C_DEBUG and C_Debug from $c/c.cm.
671
672		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
673		subtyping relationship between incomplete pointers and their complete
674		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
675		with the convention of instantiating 'o with ('t, 'c) obj whenever
676		the pointer target type is complete.  In the incomplete case, 'o
677		will be instantiated with some "'c iobj" -- a type obtained by
678		using one of the functors PointerToIncompleteType or PointerToCompleteType.
679
680		Operations that work on both incomplete and complete pointer types are
681		typed as taking an 'o ptr while operations that require the target to
682		be known are typed as taking some ('t, 'c) obj ptr.
683
684		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
685		where void is an eqtype without any values.  This makes it possible
686		to work on voidptr values using functions meant to operate on light
687		incomplete pointers.
688
689		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
690		been vastly simplified.
691
692		----------------------------------------------------------------------
693		Name: Matthias Blume
694		Date: 2002/02/19 10:48:00 EST
695		Tag: blume-20020219-pqfix
696		Description:
697
698		Applied Chris Okasaki's bug fix for priority queues.
699
700		----------------------------------------------------------------------
701		Name: Matthias Blume
702		Date: 2002/02/15 17:05:00
703		Tag: Release_110_39
704		Description:
705
706		Last-minute retagging is becoming a tradition... :-(
707
708		This is the working release 110.39.
709
710		----------------------------------------------------------------------
711		Name: Matthias Blume
712		Date: 2002/02/15 16:00:00 EST
713		Tag: Release_110_39-orig
714		Description:
715
716		Working release 110.39.  New bootfiles.
717
718		(Update: There was a small bug in the installer so it wouldn't work
719		with all shells.  So I retagged. -Matthias)
720
721		----------------------------------------------------------------------
722		Name: Matthias Blume
723		Date: 2002/02/15 14:17:00 EST
724		Tag: blume-20020215-showbindings
725		Description:
726
727		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
728		the latter can be useful for exploring what bindings are available at
729		the interactive prompt.  (The first function returns only the list
730		of symbols that are really bound, the second prints those but also the
731		ones that CM's autoloading mechanism knows about.)
732
733		----------------------------------------------------------------------
734		Name: Matthias Blume
735		Date: 2002/02/15 12:08:00 EST
736		Tag: blume-20020215-iptrs
737		Description:
738
739		Two improvements to ml-nlffigen:
740
741		1. Write files only if they do not exist or if their current contents
742		do not coincide with what's being written.  (That is, avoid messing
743		with the time stamps unless absolutely necessary.)
744
745		2. Implement a "repository" mechanism for generated files related
746		to "incomplete pointer types".   See the README file for details.
747
748		----------------------------------------------------------------------
749		Name: Matthias Blume
750		Date: 2002/02/14 11:50:00 EST
751		Tag: blume-20020214-quote
752		Description:
753
754		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
755		because of the new and improved tag generation scheme.  (Thanks to Allen
756		Leung for pointing it out.)
757
758		----------------------------------------------------------------------
759		Name: Lal George
760		Date: 2002/02/14 09:55:27 EST 2002
761		Tag: george-20020214-isabelle-bug
762		Description:
763
764		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
765		of Isabelle on the x86.
766
767		From Allen:
768		-----------
769		I've found the problem:
770
771		in ra-core.sml, I use the counter "blocked" to keep track of the
772		true number of elements in the freeze queue.  When the counter goes
773		to zero, I skip examining the queue.  But I've messed up the
774		bookkeeping in combine():
775
776		else ();
777		case !ucol of
778		PSEUDO => (if !cntv > 0 then
779		(if !cntu > 0 then blocked := !blocked - 1 else ();
780		^^^^^^^^^^^^^^^^^^^^^^^
781		moveu := mergeMoveList(!movev, !moveu)
782		)
783		else ();
784
785		combine() is called to coalesce two nodes u and v.
786		I think I was thinking that if the move counts of u and v are both
787		greater than zero then after they are coalesced then one node is
788		removed from the freeze queue.  Apparently I was thinking that
789		both u and v are of low degree, but that's clearly not necessarily true.
790
791
792		02/12/2002:
793		Here's the patch.  HOL now compiles.
794
795		I don't know how this impact on performance (compile
796		time or runtime).  This bug caused the RA (especially on the x86)
797		to go thru the potential spill phase when there are still nodes on the
798		freeze queue.
799
800
801
802
803		----------------------------------------------------------------------
804		Name: Matthias Blume
805		Date: 2002/02/13 22:40:00 EST
806		Tag: blume-20020213-fptr-rtti
807		Description:
808
809		Fixed a bug in ml-nlffigen that was introduced with one of the previous
810		updates.
811
812		----------------------------------------------------------------------
813		Name: Matthias Blume
814	Date: 2002/02/13 16:41:00 EST	Date: 2002/02/13 16:41:00 EST
815	Tag: blume-20020213-cmlpq	Tag: blume-20020213-cmlpq
816	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1063   changed lines   Added in v.1155

---

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