Home My Page Projects Code Snippets Project Openings SML/NJ
Summary Activity Forums Tracker Lists Tasks Docs Surveys News SCM Files

SCM Repository

[smlnj] Diff of /sml/trunk/HISTORY
ViewVC logotype

Diff of /sml/trunk/HISTORY

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1085, Fri Feb 22 00:15:55 2002 UTC revision 1158, Fri Mar 22 01:21:25 2002 UTC
# Line 14  Line 14 
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Allen Leung  Name: Allen Leung
17    Date: 2002/03/21 20:30:00 EST
18    Tag: leunga-20020321-cfg
19    Description:
20    
21      Fixed a potential problem in cfg edge splitting.
22    
23    ----------------------------------------------------------------------
24    Name: Allen Leung
25    Date: 2002/03/21 17:15:00 EST
26    Tag: leunga-20020321-x86-fp-cfg
27    Description:
28    
29      1. Recoded the buggy parts of x86-fp.
30    
31         a. All the block reordering code has been removed.
32            We now depend on the block placement phases to do this work.
33    
34         b. Critical edge splitting code has been simplified and moved into the
35            CFG modules, as where they belong.
36    
37         Both of these were quite buggy and complex.  The code is now much, much
38         simpler.
39    
40      2. X86 backend.
41    
42         a. Added instructions for 64-bit support.  Instruction selection for
43            64-bit has not been committed, however, since that
44            requires changes to MLTREE which haven't been approved by
45            Lal and John.
46    
47         b. Added support for FUCOMI and FUCOMIP when generating code for
48            PentiumPro and above.  We only generate these instructions in
49            the fast-fp mode.
50    
51         c. Added cases for JP and JNP in X86FreqProps.
52    
53      3. CFG
54    
55         CFG now has a bunch of methods for edge splitting and merging.
56    
57      4. Machine description.
58    
59         John's simplification of MLTREE_BASIS.fcond broke a few machine
60         description things:
61    
62         rtl-build.{sig,sml} and hppa.mdl fixed.
63    
64         NOTE: the machine description stuff in the repository is still broken.
65               Again, I can't put my fixes in because that involves
66               changes to MLTREE.
67    
68    ----------------------------------------------------------------------
69    Name: Matthias Blume
70    Date: 2002/03/20 15:55:00 EST
71    Tag: blume-20020320-kmp
72    Description:
73    
74    Implemented Knuth-Morris-Pratt string matching in PreString and used
75    it for String.isSubstring, Substring.isSubstring, and
76    Substring.position.
77    
78    (Might need some stress-testing.  Simple examples worked fine.)
79    
80    ----------------------------------------------------------------------
81    Name: Matthias Blume
82    Date: 2002/03/19 16:37:00 EST
83    Tag: blume-20020319-witnesses
84    Description:
85    
86    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
87    
88    This implements a generic mechanism for changing constness qualifiers
89    anywhere within big C types without resorting to outright "casts".
90    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
91    modify the constness at the outermost level.)
92    The implementation of "convert" is based on the idea of "witness"
93    values -- values that are not used by the operation but whose types
94    "testify" to their applicability.  On the implementation side, "convert"
95    is simply a projection (returning its second curried argument).  With
96    cross-module inlining, it should not result in any machine code being
97    generated.
98    
99    ----------------------------------------------------------------------
100    Name: Matthias Blume
101    Date: 2002/03/15 16:40:00 EST
102    Tag: blume-20020315-basis
103    Description:
104    
105    Provided (preliminary?) implementations for
106    
107      {String,Substring}.{concatWith,isSuffix,isSubstring}
108    
109    and
110    
111      Substring.full
112    
113    Those are in the Basis spec but they were missing in SML/NJ.
114    
115    ----------------------------------------------------------------------
116    Name: Matthias Blume
117    Date: 2002/03/14 21:30:00 EST
118    Tag: blume-20020314-controls
119    Description:
120    
121    Controls:
122    ---------
123    
124    1. Factored out the recently-added Controls : CONTROLS stuff and put
125       it into its own library $/controls-lib.cm.  The source tree for
126       this is under src/smlnj-lib/Controls.
127    
128    2. Changed the names of types and functions in this interface, so they
129       make a bit more "sense":
130    
131          module -> registry
132          'a registry -> 'a group
133    
134    3. The interface now deals in ref cells only.  The getter/setter interface
135       is (mostly) gone.
136    
137    4. Added a function that lets one register an already-existing ref cell.
138    
139    5. Made the corresponding modifications to the rest of the code so that
140       everything compiles again.
141    
142    6. Changed the implementation of Controls.MLRISC back to something closer
143       to the original.  In particular, this module (and therefore MLRISC)
144       does not depend on Controls.  There now is some link-time code in
145       int-sys.sml that registers the MLRISC controls with the Controls
146       module.
147    
148    CM:
149    ---
150    
151      * One can now specify the lambda-split aggressiveness in init.cmi.
152    
153    ----------------------------------------------------------------------
154    Name: Allen Leung
155    Date: 2002/03/13 17:30:00 EST
156    Tag: leunga-20020313-x86-fp-unary
157    Description:
158    
159    Bug fix for:
160    
161    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
162    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
163    > - fun f(x,(y,z)) = Real.~ y;
164    > [autoloading]
165    > [autoloading done]
166    >       fchsl   (%eax), 184(%esp)
167    > Error: MLRisc bug: X86MCEmitter.emitInstr
168    >
169    > uncaught exception Error
170    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
171    
172    The problem was that the code generator did not generate any fp registers
173    in this case, and the ra didn't know that it needed to run the X86FP phase to
174    translate the pseudo fp instruction.   This only happened with unary fp
175    operators in certain situations.
176    
177    ----------------------------------------------------------------------
178    Name: Matthias Blume
179    Date: 2002/03/13 14:00:00 EST
180    Tag: blume-20020313-overload-etc
181    Description:
182    
183    1. Added _overload as a synonym for overload for backward compatibility.
184       (Control.overloadKW must be true for either version to be accepted.)
185    
186    2. Fixed bug in install script that caused more things to be installed
187       than what was requested in config/targets.
188    
189    3. Made CM aware of the (_)overload construct so that autoloading
190       works.
191    
192    ----------------------------------------------------------------------
193    Name: Matthias Blume
194    Date: 2002/03/12 22:03:00 EST
195    Tag: blume-20020312-url
196    Description:
197    
198    Forgot to update BOOT and srcarchiveurl.
199    
200    ----------------------------------------------------------------------
201    Name: Matthias Blume
202    Date: 2002/03/12 17:30:00 EST
203    Tag: blume-20020312-version110392
204    Description:
205    
206    Yet another version number bump (because of small changes to the
207    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
208    
209    Changes:
210    
211      The new pid generation scheme described a few weeks ago was overly
212      complicated.  I implemented a new mechanism that is simpler and
213      provides a bit more "stability":  Once CM has seen a compilation
214      unit, it keeps its identity constant (as long as you do not delete
215      those crucial CM/GUID/* files).  This means that when you change
216      an interface, compile, then go back to the old interface, and
217      compile again, you arrive at the original pid.
218    
219      There now also is a mechanism that instructs CM to use the plain
220      environment hash as a module's pid (effectively making its GUID
221      the empty string).  For this, "noguid" must be specified as an
222      option to the .sml file in question within its .cm file.
223      This is most useful for code that is being generated by tools such
224      as ml-nlffigen (because during development programmers tend to
225      erase the tool's entire output directory tree including CM's cached
226      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
227      revert to the old, broken behavior of SML/NJ, but in specific cases
228      where there is no danger of interface confusion, its use is ok
229      (I think).
230    
231      ml-nlffigen by default generates "noguid" annotations.  They can be
232      turned off by specifying -guid in its command line.
233    
234    ----------------------------------------------------------------------
235    Name: Lal George
236    Date: 2002/03/12 12 14:42:36 EST
237    Tag: george-20020312-frequency-computation
238    Description:
239    
240    Integrated jump chaining and static block frequency into the
241    compiler. More details and numbers later.
242    
243    ----------------------------------------------------------------------
244    Name: Lal George
245    Date: 2002/03/11 11 22:38:53 EST
246    Tag: george-20020311-jump-chain-elim
247    Description:
248    
249    Tested the jump chain elimination on all architectures (except the
250    hppa).  This is on by default right now and is profitable for the
251    alpha and x86, however, it may not be profitable for the sparc and ppc
252    when compiling the compiler.
253    
254    The gc test will typically jump to a label at the end of the cluster,
255    where there is another jump to an external cluster containing the actual
256    code to invoke gc. This is to allow factoring of common gc invocation
257    sequences. That is to say, we generate:
258    
259            f:
260               testgc
261               ja   L1      % jump if above to L1
262    
263            L1:
264               jmp L2
265    
266    
267    After jump chain elimination the 'ja L1' instructions is converted to
268    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
269    up being implemented in their long form (if L2 is far away) using:
270    
271            jbe     L3      % jump if below or equal to L3
272            jmp     L2
273         L3:
274            ...
275    
276    
277    For large compilation units L2  may be far away.
278    
279    
280    ----------------------------------------------------------------------
281    Name: Matthias Blume
282    Date: 2002/03/11 13:30:00 EST
283    Tag: blume-20020311-mltreeeval
284    Description:
285    
286    A functor parameter was missing.
287    
288    ----------------------------------------------------------------------
289    Name: Allen Leung
290    Date: 2002/03/11 10:30:00 EST
291    Tag: leunga-20020311-runtime-string0
292    Description:
293    
294       The representation of the empty string now points to a
295    legal null terminated C string instead of unit.  It is now possible
296    to convert an ML string into C string with InlineT.CharVector.getData.
297    This compiles into one single machine instruction.
298    
299    ----------------------------------------------------------------------
300    Name: Allen Leung
301    Date: 2002/03/10 23:55:00 EST
302    Tag: leunga-20020310-x86-call
303    Description:
304    
305       Added machine generation for CALL instruction (relative displacement mode)
306    
307    ----------------------------------------------------------------------
308    Name: Matthias Blume
309    Date: 2002/03/08 16:05:00
310    Tag: blume-20020308-entrypoints
311    Description:
312    
313    Version number bumped to 110.39.1.  NEW BOOTFILES!
314    
315    Entrypoints: non-zero offset into a code object where execution should begin.
316    
317    - Added the notion of an entrypoint to CodeObj.
318    - Added reading/writing of entrypoint info to Binfile.
319    - Made runtime system bootloader aware of entrypoints.
320    - Use the address of the label of the first function given to mlriscGen
321      as the entrypoint.  This address is currently always 0, but it will
322      not be 0 once we turn on block placement.
323    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
324      with entry points) from mlriscGen.
325    
326    ----------------------------------------------------------------------
327    Name: Allen Leung
328    Date: 2002/03/07 20:45:00 EST
329    Tag: leunga-20020307-x86-cmov
330    Description:
331    
332       Bug fixes for CMOVcc on x86.
333    
334       1. Added machine code generation for CMOVcc
335       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
336       3. CMOVcc cannot have an immediate operand as argument.
337    
338    ----------------------------------------------------------------------
339    Name: Matthias Blume
340    Date: 2002/03/07 16:15:00 EST
341    Tag: blume-20020307-controls
342    Description:
343    
344    This is a very large but mostly boring patch which makes (almost)
345    every tuneable compiler knob (i.e., pretty much everything under
346    Control.* plus a few other things) configurable via both the command
347    line and environment variables in the style CM did its configuration
348    until now.
349    
350    Try starting sml with '-h' (or, if you are brave, '-H')
351    
352    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
353    implements the underlying generic mechanism.
354    
355    The interface to some of the existing such facilities has changed somewhat.
356    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
357    (The getFoo interface is still there for backward-compatibility, but its
358    use is deprecated.)
359    
360    The ml-build script passes -Cxxx=yyy command-line arguments through so
361    that one can now twiddle the compiler settings when using this "batch"
362    compiler.
363    
364    TODO items:
365    
366    We should go through and throw out all controls that are no longer
367    connected to anything.  Moreover, we should go through and provide
368    meaningful (and correct!) documentation strings for those controls
369    that still are connected.
370    
371    Currently, multiple calls to Controls.new are accepted (only the first
372    has any effect).  Eventually we should make sure that every control
373    is being made (via Controls.new) exactly once.  Future access can then
374    be done using Controls.acc.
375    
376    Finally, it would probably be a good idea to use the getter-setter
377    interface to controls rather than ref cells.  For the time being, both
378    styles are provided by the Controls module, but getter-setter pairs are
379    better if thread-safety is of any concern because they can be wrapped.
380    
381    *****************************************
382    
383    One bug fix: The function blockPlacement in three of the MLRISC
384    backpatch files used to be hard-wired to one of two possibilities at
385    link time (according to the value of the placementFlag).  But (I
386    think) it should rather sense the flag every time.
387    
388    *****************************************
389    
390    Other assorted changes (by other people who did not supply a HISTORY entry):
391    
392    1. the cross-module inliner now works much better (Monnier)
393    2. representation of weights, frequencies, and probabilities in MLRISC
394       changed in preparation of using those for weighted block placement
395       (Reppy, George)
396    
397    ----------------------------------------------------------------------
398    Name: Lal George
399    Date: 2002/03/07 14:44:24 EST 2002
400    Tag: george-20020307-weighted-block-placement
401    
402    Tested the weighted block placement optimization on all architectures
403    (except the hppa) using AMPL to generate the block and edge frequencies.
404    Changes were required in the machine properties to correctly
405    categorize trap instructions. There is an MLRISC flag
406    "weighted-block-placement" that can be used to enable weighted block
407    placement, but this will be ineffective without block/edge
408    frequencies (coming soon).
409    
410    
411    ----------------------------------------------------------------------
412    Name: Lal George
413    Date: 2002/03/05 17:24:48 EST
414    Tag: george-20020305-linkage-cluster
415    
416    In order to support the block placement optimization, a new cluster
417    is generated as the very first cluster (called the linkage cluster).
418    It contains a single jump to the 'real' entry point for the compilation
419    unit. Block placement has no effect on the linkage cluster itself, but
420    all the other clusters  have full freedom in the manner in which they
421    reorder blocks or functions.
422    
423    On the x86 the typical linkage code that is generated is:
424       ----------------------
425            .align 2
426       L0:
427            addl    $L1-L0, 72(%esp)
428            jmp     L1
429    
430    
431            .align  2
432       L1:
433       ----------------------
434    
435    72(%esp) is the memory location for the stdlink register. This
436    must contain the address of the CPS function being called. In the
437    above example, it contains the address of  L0; before
438    calling L1 (the real entry point for the compilation unit), it
439    must contain the address for L1, and hence
440    
441            addl $L1-L0, 72(%esp)
442    
443    I have tested this on all architectures except the hppa.The increase
444    in code size is of course negligible
445    
446    ----------------------------------------------------------------------
447    Name: Allen Leung
448    Date: 2002/03/03 13:20:00 EST
449    Tag: leunga-20020303-mlrisc-tools
450    
451      Added #[ ... ] expressions to mlrisc tools
452    
453    ----------------------------------------------------------------------
454    Name: Matthias Blume
455    Date: 2002/02/27 12:29:00 EST
456    Tag: blume-20020227-cdebug
457    Description:
458    
459    - made types in structure C and C_Debug to be equal
460    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
461    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
462    
463    ----------------------------------------------------------------------
464    Name: Matthias Blume
465    Date: 2002/02/26 12:00:00 EST
466    Tag: blume-20020226-ffi
467    Description:
468    
469    1. Fixed a minor bug in CM's "noweb" tool:
470       If numbering is turned off, then truly don't number (i.e., do not
471       supply the -L option to noweb).  The previous behavior was to supply
472       -L'' -- which caused noweb to use the "default" line numbering scheme.
473       Thanks to Chris Richards for pointing this out (and supplying the fix).
474    
475    2. Once again, I reworked some aspects of the FFI:
476    
477       A. The incomplete/complete type business:
478    
479       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
480         gone!
481       - ML types representing an incomplete type are now *equal* to
482         ML types representing their corresponding complete types (just like
483         in C).  This is still safe because ml-nlffigen will not generate
484         RTTI for incomplete types, nor will it generate functions that
485         require access to such RTTI.   But when ML code generated from both
486         incomplete and complete versions of the C type meet, the ML types
487         are trivially interoperable.
488    
489         NOTE:  These changes restore the full generality of the translation
490         (which was previously lost when I eliminated functorization)!
491    
492       B. Enum types:
493    
494       - Structure C now has a type constructor "enum" that is similar to
495         how the "su" constructor works.  However, "enum" is not a phantom
496         type because each "T enum" has values (and is isomorphic to
497         MLRep.Signed.int).
498       - There are generic access operations for enum objects (using
499         MLRep.Signed.int).
500       - ml-nlffigen will generate a structure E_foo for each "enum foo".
501         * The structure contains the definition of type "mlrep" (the ML-side
502         representation type of the enum).  Normally, mlrep is the same
503         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
504         then mlrep will be defined as a datatype -- thus facilitating
505         pattern matching on mlrep values.
506         ("-ec" will be suppressed if there are duplicate values in an
507          enumeration.)
508         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
509         will be generated for each C enum constant xxx.
510         * Conversion functions m2i and i2m convert between mlrep and
511         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
512         * Coversion functions c and ml convert between mlrep and "tag enum".
513         * Access functions (get/set) fetch and store mlrep values.
514       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
515         enumerations are merged into one single enumeration represented by
516         structure E_'.
517    
518    ----------------------------------------------------------------------
519    Name: Allen Leung
520    Date: 2002/02/25 04:45:00 EST
521    Tag: leunga-20020225-cps-spill
522    
523    This is a new implementation of the CPS spill phase.
524    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
525    In case of problems, replace it with the old file spill.sml
526    
527    The current compiler runs into some serious performance problems when
528    constructing a large record.  This can happen when we try to compile a
529    structure with many items.  Even a very simple structure like the following
530    makes the compiler slow down.
531    
532        structure Foo = struct
533           val x_1 = 0w1 : Word32.int
534           val x_2 = 0w2 : Word32.int
535           val x_3 = 0w3 : Word32.int
536           ...
537           val x_N = 0wN : Word32.int
538        end
539    
540    The following table shows the compile time, from N=1000 to N=4000,
541    with the old compiler:
542    
543    N
544    1000   CPS 100 spill                           0.04u  0.00s  0.00g
545           MLRISC ra                               0.06u  0.00s  0.05g
546              (spills = 0 reloads = 0)
547           TOTAL                                   0.63u  0.07s  0.21g
548    
549    1100   CPS 100 spill                           8.25u  0.32s  0.64g
550           MLRISC ra                               5.68u  0.59s  3.93g
551              (spills = 0 reloads = 0)
552           TOTAL                                   14.71u  0.99s  4.81g
553    
554    1500   CPS 100 spill                           58.55u  2.34s  1.74g
555           MLRISC ra                               5.54u  0.65s  3.91g
556              (spills = 543 reloads = 1082)
557           TOTAL                                   65.40u  3.13s  6.00g
558    
559    2000   CPS 100 spill                           126.69u  4.84s  3.08g
560           MLRISC ra                               0.80u  0.10s  0.55g
561              (spills = 42 reloads = 84)
562           TOTAL                                   129.42u  5.10s  4.13g
563    
564    3000   CPS 100 spill                           675.59u  19.03s  11.64g
565           MLRISC ra                               2.69u  0.27s  1.38g
566              (spills = 62 reloads = 124)
567           TOTAL                                   682.48u  19.61s  13.99g
568    
569    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
570           MLRISC ra                               4.96u  0.27s  2.72g
571              (spills = 85 reloads = 170)
572           TOTAL                                   2375.26u  57.21s  48.00g
573    
574    As you can see the old cps spill module suffers from some serious
575    performance problem.  But since I cannot decipher the old code fully,
576    instead of patching the problems up, I'm reimplementing it
577    with a different algorithm.  The new code is more modular,
578    smaller when compiled, and substantially faster
579    (O(n log n) time and O(n) space).  Timing of the new spill module:
580    
581    4000  CPS 100 spill                           0.02u  0.00s  0.00g
582          MLRISC ra                               0.25u  0.02s  0.15g
583             (spills=1 reloads=3)
584          TOTAL                                   7.74u  0.34s  1.62g
585    
586    Implementation details:
587    
588    As far as I can tell, the purpose of the CPS spill module is to make sure the
589    number of live variables at any program point (the bandwidth)
590    does not exceed a certain limit, which is determined by the
591    size of the spill area.
592    
593    When the bandwidth is too large, we decrease the register pressure by
594    packing live variables into spill records.  How we achieve this is
595    completely different than what we did in the old code.
596    
597    First, there is something about the MLRiscGen code generator
598    that we should be aware of:
599    
600    o MLRiscGen performs code motion!
601    
602       In particular, it will move floating point computations and
603       address computations involving only the heap pointer to
604       their use sites (if there is only a single use).
605       What this means is that if we have a CPS record construction
606       statement
607    
608           RECORD(k,vl,w,e)
609    
610       we should never count the new record address w as live if w
611       has only one use (which is often the case).
612    
613       We should do something similar to floating point, but the transformation
614       there is much more complex, so I won't deal with that.
615    
616    Secondly, there are now two new cps primops at our disposal:
617    
618     1. rawrecord of record_kind option
619        This pure operator allocates some uninitialized storage from the heap.
620        There are two forms:
621    
622         rawrecord NONE [INT n]  allocates a tagless record of length n
623         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
624                                     and initializes the tag.
625    
626     2. rawupdate of cty
627          rawupdate cty (v,i,x)
628          Assigns to x to the ith component of record v.
629          The storelist is not updated.
630    
631    We use these new primops for both spilling and increment record construction.
632    
633     1. Spilling.
634    
635        This is implemented with a linear scan algorithm (but generalized
636        to trees).  The algorithm will create a single spill record at the
637        beginning of the cps function and use rawupdate to spill to it,
638        and SELECT or SELp to reload from it.  So both spills and reloads
639        are fine-grain operations.  In contrast, in the old algorithm
640        "spills" have to be bundled together in records.
641    
642        Ideally, we should sink the spill record construction to where
643        it is needed.  We can even split the spill record into multiple ones
644        at the places where they are needed.  But CPS is not a good
645        representation for global code motion, so I'll keep it simple and
646        am not attempting this.
647    
648     2. Incremental record construction (aka record splitting).
649    
650        Long records with many component values which are simulatenously live
651        (recall that single use record addresses are not considered to
652         be live) are constructed with rawrecord and rawupdate.
653        We allocate space on the heap with rawrecord first, then gradually
654        fill it in with rawupdate.  This is the technique suggested to me
655        by Matthias.
656    
657        Some restrictions on when this is applicable:
658        1. It is not a VECTOR record.  The code generator currently does not handle
659           this case. VECTOR record uses double indirection like arrays.
660        2. All the record component values are defined in the same "basic block"
661           as the record constructor.  This is to prevent speculative
662           record construction.
663    
664    ----------------------------------------------------------------------
665    Name: Allen Leung
666    Date: 2002/02/22 01:02:00 EST
667    Tag: leunga-20020222-mlrisc-tools
668    
669    Minor bug fixes in the parser and rewriter
670    
671    ----------------------------------------------------------------------
672    Name: Allen Leung
673  Date: 2002/02/21 20:20:00 EST  Date: 2002/02/21 20:20:00 EST
674  Tag: leunga-20020221-peephole  Tag: leunga-20020221-peephole
675    

Legend:
Removed from v.1085  
changed lines
  Added in v.1158

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