Svelte/log

   1 #!/bin/bash
   2 if [ -e ~/.log.lock ]
   3 then
   4   echo ~/.log.lock exists :\(
   5   exit
   6 else
   7   touch ~/.log.lock && nano +99999 ~/log && rm ~/.log.lock && exit
   8 fi
   9
  10 ---------------------------------------------------------------------
  11
  12 CHANGED
  13         IRTests.java
  14         JavaIRTests.java
  15
  16 2007-10-29
  17         * Figured out how to get line # from SSA index, Bytecode index
  18         * Tried "thin slicing" -- regular slicing with "NO_BASE_PTRS".
  19           Didn't really work, put in extra line.
  20         * Talked to Bodik, he suggested understanding more about the IR,
  21           and possibly thinking of prospective tools that could be useful
  22           in understanding the API
  23         * Tried some simple examples (THE GLORIOUS MAIN()) to see what
  24           IR code it made
  25         * Learned about variable numbers & uses and defs, Java method
  26           descriptors, some specific IR instructions from above
  27         * Tried getting all functions to work (one CallGraph with all
  28           functions, without too much luck.) Can choose one function, hackily.
  29         * Thought of idea for "tree-view" tool to show where all these
  30           objects come from in one particular instance -- what method
  31           created them, etc. For example, slice =
  32           Slicer.computerBackwardSlice -- show where those args came
  33           from, where the args to those methods came from, etc.
  34           builder.getPointerAnalysis and builder.makeCallGraph would
  35           point down to the same "builder"...
  36         * In order to understand more, upon Bodik's suggestion looked
  37           for Constant Propagation w/o luck. Found
  38           DeadAssignmentElimination and learned about phi instructions,
  39           dead phi removal, etc.
  40
  41 Misc notes: Ask Bodik if he is aware of Indus, Kaveri
  42
  43 2007-10-30
  44         * Worked on Arbolico (tree-view tool) to learn more about instructions,
  45           analyses. Just looked for get invokespecial instructions (new) and
  46           invokevirtual instructions that define something, and added edges to
  47           generate a dot graph from that. (A better way is probably to use
  48           Eclipse libraries to analyze the source). Kinda worked for simple
  49           example, not so good for real-life WALA stuff. Learned a bit
  50           about instructions, defs, uses, etc. anyway.
  51         * Put slicing test & print instruction in same program
  52           to compare, learn more about purpose of FakeRootClass.
  53           FakeRootClass sets up constants? (strings?), calls main.
  54           arraystore?
  55         * Tried to understand how Slicer works. Got lost in
  56           SDG/PDG stuff. Read some papers on it, didn't understand
  57           much.
  58         * Switched gears and tried to learn up on Eclipse plugins
  59           (lost/tired...)
  60         * Read up a bit on overall Eclipse & Java terms and concepts.
  61
  62 2007-10-31
  63         * Compared results using "ThinSlicer" against regular slicer and
  64           regular size with NO_BASE_PTRS. Same in latter case, but
  65           ThinSlicer doesn't get FakeRootClasses.
  66         * Reread paper. Realized ThinSlicer is the context-insensitive
  67           variant
  68         * Running CISlicer, looked at the CISDG and SDG by getting the
  69           graph and using dot. The graphs have a lot of extra nodes / edges
  70           in them but I pruned it down by grepping for ManusTSlice.
  71           It's hard to see what's going on because there are no more variable
  72           names -- in fact, statements like w=x and z=x have been optimized
  73           out completely.
  74           In addition, invokevirtual calls from the new() calls appear
  75           to be one node because the names are too short for
  76           dot to distinguished (clipped by the dot writer in Wala).
  77           Anyway, the problem with thin slicing is that there is an edge
  78           in the CISDG from x = new A() to w.f = y (which is really x.f = y,
  79           since w, x, and z have been merged to variable '3' in the SSA).
  80           Since the statements w=x and z=x no longer exist, this could mean
  81           we are following base pointer data dependences (TODO: try more
  82           to verify ALL base pointer data dependences are made), even though
  83           we are using the options NO_BASE_NO_HEAP (results are the same, at least
  84           for the final slice, with just NO_HEAP, which was in ThinSlicer.java
  85           originally). For thin slicing with the regular context-sensitive slicer
  86           and with CISlicer, slice results are same -- x = new A() is included,
  87           but w == z is not, suggesting that at least control dependences are not
  88           followed.
  89         * Looked at SDG of context-sensitive slice with NO_BASE_PTRS. There is no edge from
  90           x = new A() to putfield (GOOD), but there is an edge from x = new A() to
  91           getfield (BAD). There is no edge from the conditional branch to v=x.f (GOOD).
  92         * Looked at SDG of traditional slice. Lots of interprocedural stuff. But also,
  93           edges from x = new A() to putfield (w.f = y, effectively x.f = y) and also from
  94           w == z to get v = z.f (effectively x == x and v = x.f). OK.
  95         * Looked where ignoreBasePtrs flag is used: in building the PDG.
  96           Thought maybe the check wasn't working, but indeed it doesn't add an
  97           edge there between the x=new A() and v=x.f (getfield) statements if
  98           ignoreBasePtrs is on.
  99         * Overrode PDG.addEdge() and looked for the offending edge, but it isn't being
 100           added in the PDG evidently. Done after creating the SDG?
 101
 102   @Override
 103   public void addEdge(Statement src, Statement dst) throws IllegalArgumentException {
 104     //if ( dst.getDef() == 7 && dst instanceof SSAGetInstruction ) { System.out.println("oops!"+ dOptions.isIgnoreBasePtrs()); }
 105     if ( dst instanceof NormalStatement ) {
 106       SSAInstruction use = ((NormalStatement) dst).getInstruction();
 107       if ( use.getDef() == 7 && use instanceof SSAGetInstruction ) { System.out.println("oops!"+ dOptions.isIgnoreBasePtrs()); }
 108     }
 109
 110     super.addEdge(src,dst);
 111   }
 112
 113         * Saved SDG graph in CISlicer right after calling SDG().
 114           At this stage, the line connecting x=new A() and v=x.f is absent
 115           if NO_BASE is included (good) and present if NO_BASE is not included
 116           (traditional-like slice, also good). So it seems that in the case of
 117           CISlicer the errant arrow is added AFTER this (mod/ref stuff? getfield
 118           and putfield not linked before this, maybe this does the pointer analysis
 119           stuff, context-insensitively in this case). Maybe this is because
 120           ThinSlicer was meant to be used with NO_HEAP, not NO_BASE_NO_HEAP?
 121           But why doesn't NO_BASE work with context-sensitive slicer?
 122           I really think I should do more testing to see if these are always
 123           followed.
 124
 125         * Tried weird example:
 126 class Main {
 127   public static void main ( String args[] ) {
 128     A x;
 129     x = new A();
 130     x.f = new B();
 131     WeirdContainer c = new WeirdContainer(x);
 132     A a2 = c.getA();
 133     System.out.println(a2.f);                   // seed for now
 134   }
 135 }
 136           Slice is "x = new A();", "x.f = new B();", and the seed. OK except for x = new A()...
 137           Hmm... maybe not so many problems as I thought.
 138         * Read about Eclipse plug-in framework.
 139
 140 2007-11-01
 141         * Wrote summary email to Prof Bodik.
 142         * Read some more about Eclipse plugins. Created a plugin
 143           with a toolbar action of readme files but I couldn't use the
 144           TextSelection class (error about loading it).
 145         * Switch gears: investigate more Arbolico
 146         * Back to eclipse: figured out TextSelection stuff. Now I want
 147           to make it show up on a JavaEditor. Yay! it works!
 148           Just like readme example but action is org.eclipse.jdt.ui.internal.
 149         * Got back to the ICompilationUnit, thanks to Prospector.
 150         * GOt to slide 40 on Reps' presentation (cis.upenn.edu)
 151
 152 2007-11-02
 153         * Read more of presentation, up to where it seemed irrelevant.
 154         * Used JavaSketch to look for ways to get String data out of
 155           an editor (given offset and length) w/o much success.
 156                 String (ICompilationUnit e) {(int), @call 1 ~[org.eclipse.jdt.ui]}
 157           -> text is messed up, can't see
 158           -> giving int arguments doesn't do anything (?)
 159           -> can I restrict by parameter names/types (I want it to use
 160              an int offset)?
 161           -> I can right click and exclude a method, but it would be
 162              useful to exclude a CLASS (in the case of a static call)
 163           -> Bug: try
 164                 String (org.eclipse.jdt.internal.ui.javaeditor.CompilationUnitEditor cue) {(int), @call 1 ~[org.eclipse.jdt.ui]} @(internal)
 165              then right click -> exclude constructor
 166              DeleteNextSubWordAction everywhere ($ causing problems)
 167         -> Try
 168            String (org.eclipse.ui.texteditor.AbstractTextEditor cue) {(int)}
 169            Gives me all this preference stuff. Can I exclude text from
 170            ANYWHERE in the query?
 171         * These APIs are too compliated. I never have this many problems
 172           doing something this simple in GTK.
 173         * Studying JavaSketch Eclipse/UI-related code to learn more
 174           about writing Eclipse plugins with use Java source files.
 175         * FINALLY found answer to JavaSketch query, just happened upon
 176           compilationunit.getSource() in JavaSketch source:
 177                 String s = cu.getSource().substring(start,end);
 178
 179         * Installed cast.java, cast.js. Couldn't get java JUnit tests to work
 180           due to java.lang.ClassNotFoundException: com.ibm.wala.cast.java.test.SyncDuplicatorTest
 181         * Tried out wala_output/cislictest. The CISlicer did get the
 182           extra call that didn't come from y, but two things I don't
 183           understand:
 184                 * Why did the CS thick slicer get stuff related to x?
 185                 * Why did the CS thick slicer get the calls themselves
 186                   y=cube(y), etc., whereas the other two didn't? Do
 187                   these only have control dependencies?
 188
 189   public static int cube(int a) {
 190     return a*a*a;                       // all (CS thin, CI thin, CS thick)
 191   }
 192
 193   public static void main(String args[]) {
 194     int x = 2;
 195     x = x * x;                          // CI thin slicer, CS thick slicer
 196     x = cube(x);                        // CS thick slicer
 197     int y = 7;
 198     y = y * y;                          // all
 199     y = cube(y);                        // CS thick slicer
 200     System.out.println(y);              // seed
 201     y = y+x;
 202   }
 203
 204 2007-11-05
 205         * Put a couple things on the wiki (from ir to source line
 206           numbers, example slicing code)
 207         * Unable to get CAst or Eclipse to work, so trying to look out
 208           where opts happenning so we can cut 'em off.
 209         * Unable to do that. Posted to wala-wala. Got response:
 210           happens in SSABuilder, but impossible to disable
 211           because SSAInstruction doesn't support assigments. Updated wiki.
 212           If this is the only problem, perhaps we could change SSA?
 213           After all, assignment is a simple instruction which really does
 214           nothing. It just has one use and one def.
 215         * Dave explained to me workings of mods, refs, pointer analysis with
 216           regards to SDG creating (links thru the heap), that this is what is
 217           being done at second stage of creation in CISDG, hotplugging in the
 218           modifications to all abstract objects used to a node to determine
 219           predecessor nodes.
 220         * Looked up a few more papers describing this; will read them fully later.
 221                 UsingHeapFlowPointsTo0.ps
 222                 HeapFlowPointsToInSlicing-HorwitzNew.pdf
 223              somewhat:
 224                 JavaPointsToAnalysis.pdf
 225         * Tried to get CAst & Eclipse things working again in a new workspace,
 226           drowning in dependencies, then noticed it had no errors on
 227           old workspace, magically enough. JUnit test semi-crashed the
 228           system, and now CAst & Eclipse projects magically don't work
 229           again, but fixed on new workspace ~/walacastwork
 230         * Started browsing cast stuff to see how slicing could possibly
 231           be done thru this. (Haven't heard from Hsiu-fan yet since I
 232           just emailed him this morning).
 233         * Made a test case in JavaIRTest.java, but it failed with out
 234           of heap space error.
 235
 236 2007-11-06
 237         * Tried again to get Java CAst IR tests to work w/o success.
 238         * Tried reading "Effect Whole-Program Analysis Paper", didn't
 239           follow too long, unsure if relevant.
 240         * By hijacking JUnit tests, got it to show me the IR thru CAst
 241           from a Java file, but that's about it. Tried to do a slice but
 242           it didn't work for unknown reason (tried catching all
 243           exceptions but that didn't work. Just dies in the middle w/o
 244           saying anything.)
 245         * Figured out how to get an AST node given length, offset from
 246           Eclipse plugin.
 247         * Looked at Kaveri code, from this figured out a bit about
 248           annotations. Got my eclipse plug-in to annotate some
 249           text (e.g. selected text, or parent AST node), by typical
 250           error annotations, or by highlighting it in yellow.
 251
 252 2007-11-07
 253         * Got annotations to come up green on the right bar ("overview
 254           ruler"), and with my own icon of the left bar ("vertical
 255           ruler"). Need to learn about Eclipse Image managing/disposal.
 256           That's good enough for annotations for now (also I should read
 257           FAQ book)
 258         * Talked to Hsiu-Fan about slicing with CAst. Not too much
 259           progress so far, although his CISlicer copies should prove
 260           useful once I figure out how to do CS slices.
 261         * Posted to newsgroup about using CAst to get a slice.
 262         * Was going to try to get CAst to print IR, then find an
 263           instruction and take a slice using the instr. index.
 264           But then everything hanged -- Simple1 AND MiniaturSliceBug!
 265         * Reset computer and tried again, MiniaturSliceBug worked for
 266           2 runs but hanged on third, after adding couple lines to
 267           test source file... Is test source file too big? Accdng to top,
 268           all memory (1GB) used. Probably swapping very slowly?
 269         * Rebooted again, no luck. Reduced MiniaturSliceBug.java to
 270           a very small trivial program (Hello World), but it STILL hangs.
 271           Works for some small programs, but seems to hang with anything
 272           with System.out.println() in it.
 273         * CI-Thin-Sliced a small program by first getting the
 274           instruction index or the seed statement and then going from
 275           there. Even got the line numbers in source!
 276         * Made an independent project MyFunFunWalaCastSlicingTest,
 277           with getStatementFromLineNumber (only works if line
 278           number is in the CGNode passed, e.g. main) working.
 279         * Copied HelloWorld plugin over into Operation Chao. Getting
 280           ready to copy over stuff to make first thin slicing eclipse
 281           plugin!
 282
 283 2007-11-08
 284         * Slicing plug-in works! Kinda.
 285         * Switched to Hsiu-Fan's CISlicer to handle problems with
 286           ModRef visitors (wouldn't work interprocedurally).
 287         * Investigated hanging problem in JavaIRTests. It seems many
 288           system library functions (System.out.println,
 289           Integer.toHexString) causes problems. Took some stack traces,
 290           emailed the newsgroup about it.
 291         * Trying to figure out how to get the method of the selected
 292           statement -- more precisely, the method descriptor.
 293           Getting nowhere on that front.
 294         * Get position, including start and end of columns, not just
 295           line num, and make annotation with that.
 296         * All right, it KINDA works for whatever method you're in now.
 297           Just looks thru all CGNodes in the call graph until it finds
 298           out with the source position covering the line number.
 299           Problems with different files, though...
 300
 301 2007-11-09
 302         * Updated to latest SVN (Java 6.0 exclusions -- don't hang on
 303           CAst analysis).
 304         * Couldn't find the exclusions file when running as a plugin.
 305           Looked for info on ClassLoader resources, where it's looking
 306           for the files, but haven't found anything. Copied exclusions
 307           files into the bin directory of the plugin and it worked.
 308         * Figure out what class & package you're using for the slice,
 309           instead of assuming "Main".
 310         * Fix bug including null IR statements in seed statements,
 311           causing crash.
 312         * Met with Bodik & Hsiu-Fan after lunch.
 313         * Looking into why in this case
 314             cube(a) {
 315                 return a*a*a;
 316             }
 317                 ...
 318                 y = cube(y);
 319                 System.out.println(y)
 320           a*a*a is in the slice but y = cube(y) is not.
 321           Don't know why it doesn't look at that, but per Bodik's
 322           suggestion I am now looking at what CS Slicer does.
 323         * Trying CS slicer with CAst. Have to use AstJavaModRef (?) stuff.
 324         * Action buttons for CI Thin, CS Thin, or CS Thick slicing.
 325         * Discovered that the reason CS Thick slicing gets too much
 326           is probably because of exceptional edges (control may never
 327           even reach that point). Just putting a try/catch around stuff
 328           makes it kosher. Should definitely be a user-option to ignore
 329           exceptional edges (except maybe really "exceptional" cases),
 330           probably even a default.
 331         * Realized that thick slicing doesn't really get y=cube(y) either,
 332           it's just getting it 'cos of the exceptional control edge. This
 333           can be seen by adding a try/catch block around "y=cube(y)".
 334           Shouldn't y=cube(y) add a data dependency node to anything that
 335           uses y? That line definitely affects the value of y...
 336         * Found that by changing createScalarDataDependenceEdges in PDG.java
 337           if we get rid of this
 338                if (!(statement instanceof SSAAbstractInvokeInstruction)) {
 339           we can draw data dependencies directly. But then x = f(y) makes it
 340           like x depends on y ALWAYS. Even if f(y) doesn't depend on y (e.g.
 341           is a constant). Maybe they add another statement to account for this,
 342           like that HEAP_PARAM_CALLEE stuff, and just make that define on the var
 343           "x", and anything the uses x will add an edge but that will only add an edge
 344           back to y thru the function call f so depending on f x may not depend on y.
 345           Actually that's gotta be what they do. So the best thing to do is notice
 346           the CALLEE stuff and find the function call and add a notation by hand.
 347         see pdg-sdg-calls-not-given-in-slice.png on desktop
 348
 349 2007-11-13
 350         * Get used calls:
 351                 CI thin slicing: slice contains PARAM_CALLER and PARAM_CALLEE statements.
 352                         PARAM_CALLER statements have a getCall() method which gives the
 353                         instruction with the call which needs to be in the slice,
 354                         which seems to be sufficient to get needed calls.
 355                         Also have NORMAL_RET_CALLER and NORMAL_RET_CALLEE statements
 356                 CS thin slicing: slice contains HEAP_PARAM_CALLEE, HEAP_PARAM_CALLER,
 357                         HEAP_RET_CALLER, HEAP_RET_CALLEE. HEAP_RET_CALLER is sufficient
 358                         to get call.
 359                 CS thick slicing: has everything, both PARAM_CALLER and
 360                         HEAP_RET_CALLER statements.
 361         * Added a source to get an inter-compilationunit example running.
 362           much work remains to figure out which sources to add.
 363         * Check source filename to select lines only in source file. In the future we'd
 364           like to keep track of these and add annotations when switching editors.
 365         * Tried adding BinaryDirectoryTreeModule for bin directory with class files in. Works.
 366         * Looking at EclipseProjectSourceAnalysisEngine in WALA Eclipse stuff.
 367           Spent a while doing this, figuring out about engines. Didn't get far.
 368         * Went back to using BinaryDirectoryTreeModule only to find it doesn't work
 369           (polyglot error). Took a long time to figure out you NEED the exclude line
 370           in the exclusions file for it to work (it had mysteriously vanished
 371           from the OperationChao/bin directory)
 372         * Tried running it on OperationChao, but it opens all WALA files as Java
 373           source files and opens too many files.
 374         * Fixed this (maybe) by using binary class files for WALA and adding to exclusion
 375           file, but now it won't work because Polyglot only supports Java 1.4 and generics
 376           (Java 1.5) are heavily used in WALA / my slicer plugin (Operation Chao)
 377
 378 2007-11-14
 379         * Polyglot5 won't compile with newest polyglot -- looks for classes that
 380           no longer exist (API completely changed)
 381         * Won't compile if file uses binary nested classes (Bla$Foo.class)
 382         * This can be gotten around by importing only base class (Bla) and
 383           referring to Foo as Bla.Foo
 384         * Now I have the problem "can't find superclass" for the class I'm analyzing
 385         * Included that class as source. Now it can't find main in the class.
 386           This takes some thinking. Slicing depends on an entrypoint.
 387           The entrypoint in real life would probably be something far
 388           away. Like in the slicer plugin code. But this is being ignored
 389           in the callgraph right now due to the exclusions file. But we must have
 390           this exclusions file for binary class files or we get an error (cast from
 391           shrike to polyglot or something). So our options are
 392                 1) do everything from source.
 393                         would require:
 394                         2a) converting <generic> stuff before giving to polyglot
 395                         2b) porting polyglot5 to polyglot 2.3 or porting polyglot 1.3 to WALA
 396                 2) investigate more and try to use binary classes in call graph alongside CAst
 397                 3) forget CAst and use shrike
 398                 4) forget WALA and do our own analyses.
 399           Time for some new practice data.
 400         * JavaSketch, Kaveri also require Java 5. Let's slice Polyglot 2.3.0.
 401         * Running on Polyglot causes stack overflow. Try something (slightly) smaller: gpsylon
 402         * Gpslyon causes strange error:
 403                 com.ibm.wala.util.debug.UnimplementedError: getCAstTypeFor() passed type that is not primitive, array, or class?
 404         * Enough of this, I'm looking at JavaSketch code to try & understand some things, such as background processing.
 405         * Changed hello world plugin to make a simple job.
 406         * Read Reps 90 quite thoroughly up to page 12.
 407
 408 2007-11-15
 409         * Check filename when looking for method by line number.
 410           Probably got wrong method when testing before. Retested
 411           GPSylon: stack overflow.
 412         * Used binary class files. After adding all the exclusions
 413           I finally get an error:
 414                 com.ibm.wala.util.debug.UnimplementedError: Cannot find super class for <src-class: Lorg/dinopolis/gpstool/Gpsylon$ScaleAction(within Lorg/dinopolis/gpstool/Gpsylon)> in [Ljavax/swing/AbstractAction]
 415           I worked around a similar error before by including the superclass .java file in the sources,
 416           but this isn't going to work for javax.swing.* classes.
 417           This may be worth asking on the mailing list, if we decided to actually use polyglot.
 418         * Read up to pg ~18 or 19 in Reps 90. Lost me with the TDP
 419           creation stuff.
 420         * The three major problems with using Polyglot:
 421                 1) Only for <= Java 1.4
 422                 2) Cannot find super class for <src-class: Lcom/example/delmenow/Main> in [Ljavax/swing/AbstractAction]
 423                   We can't include javax.swing because it's too big.
 424                   -> Maybe we can get around this -- don't really need/want to look at super class. Talk to Ras -- why would want to look in it
 425                         for slice? super() / this() calls?
 426                 3) Stack overflow (for polyglot)
 427                   -> Investigate. Maybe it's a bug with only one file. Or maybe it's because it's too many files.
 428                    or for gpsylon:
 429                         com.ibm.wala.util.debug.UnimplementedError: getCAstTypeFor() passed type that is not primitive, array, or class?
 430                   -> ???
 431         * For 2) found out that we actually CAN include javax.swing and java.awt. Took these out of exclusion file.
 432           Tried to slice gpsylon again, but now I get a very strange "error"
 433
 434 com.ibm.wala.util.debug.UnimplementedError: found [<Primordial,Ljava/beans/PropertyChangeListener>] interfaces for [Ljava/lang/Object,
 435 Ljava/beans/PropertyChangeListener, Lorg/dinopolis/gpstool/GpsylonKeyConstants, Lorg/dinopolis/gpstool/hook/MapNavigationHook,
 436 Lorg/dinopolis/gpstool/hook/StatusHook, Lorg/dinopolis/gpstool/util/Positionable] for <src-class: Lorg/dinopolis/gpstool/Gpsylon>
 437
 438           is there an error here?
 439         * OK, so I investigated this and this is basically a "can't find implemented interface" (like "can't find superclass"). It couldn't
 440           find the interfaces not mentioned in the first array. (Why it found one still, I don't know...) So I have to get out
 441           org/dinopolis/gpstool of the exclusion file.
 442         * Note: ignore 'org\/.*" ignores subpackages
 443         * Of course, we all know what happens when we exclude binary files...
 444                 java.lang.ClassCastException: com.ibm.wala.classLoader.ShrikeClass cannot be cast to com.ibm.wala.cast.java.loader.JavaSourceLoaderImpl$JavaClass
 445         * SOLVED! This means it loaded the class whose file we're analyzing not just from source, but also from binary class file.
 446           Add it to the exclusions file and we're good.
 447         * IT MADE A CALL GRAPH! AND FOUND THE SEED STATEMENTS!! Next error:
 448                 CI thin slice:
 449                 java.lang.IllegalArgumentException: I is null
 450                         at com.ibm.wala.ipa.modref.DelegatingExtendedHeapModel.getPointerKeyForArrayContents(DelegatingExtendedHeapModel.java:71)
 451                         at com.ibm.wala.ipa.modref.ModRef$ModVisitor.visitNew(ModRef.java:276)
 452                         at com.ibm.wala.ssa.SSANewInstruction.visit(SSANewInstruction.java:99)
 453                         at com.ibm.wala.ipa.modref.ModRef.getMod(ModRef.java:340)
 454                         at com.ibm.wala.cast.java.ipa.slicer.CISlicer.scanForMod(CISlicer.java:100)
 455                         at com.ibm.wala.cast.java.ipa.slicer.CISlicer.<init>(CISlicer.java:73)
 456                         at com.ibm.wala.cast.java.ipa.slicer.ThinSlicer.<init>(ThinSlicer.java:39)
 457                         at com.ibm.wala.cast.java.ipa.slicer.ThinSlicer.<init>(ThinSlicer.java:35)
 458                         at operationchao.MyFunFunWalaCastSlicingTest.getPositionsOfThinSliceForLineNumInMain(MyFunFunWalaCastSlicingTest.java:427)
 459                 CS thin slice:
 460                 java.lang.IllegalArgumentException: I is null
 461                         at com.ibm.wala.ipa.modref.DelegatingExtendedHeapModel.getPointerKeyForArrayContents(DelegatingExtendedHeapModel.java:71)
 462                         at com.ibm.wala.ipa.modref.ModRef$ModVisitor.visitNew(ModRef.java:276)
 463                         at com.ibm.wala.ssa.SSANewInstruction.visit(SSANewInstruction.java:99)
 464                         at com.ibm.wala.ipa.modref.ModRef.scanNodeForMod(ModRef.java:164)
 465                         at com.ibm.wala.ipa.modref.ModRef.scanForMod(ModRef.java:131)
 466                         at com.ibm.wala.ipa.modref.ModRef.computeMod(ModRef.java:72)
 467                         at com.ibm.wala.ipa.slicer.SDG.<init>(SDG.java:127)
 468         * This seems to happen while visiting this instruction:
 469                 56 = new <Primordial,[Lsun/util/calendar/Era>@352 dims: 1
 470                 Problem statement is java.util.JapaneseImperialCalendar.<clinit>()V:56 = new <Primordial,[Lsun/util/calendar/Era>@352 dims: 1
 471           I love Java. Sure libraries may be big, but all that useful stuff. What would I do with Japanese Imperial Calendar support?
 472           Was going to try a test case with just using JapaneseImperialCalendar, but it's a private class. grepped whole workspace for
 473           'calendar'... northing. Don't know where this is coming from. Dead end.
 474         * Was going to investigate stack overflow but realized it's a DIFFERENT error for gpsylon using all sources...
 475           This is crazy. I've seen tens of different kinds of errors. I scratch my head, dig around for a really long time and manage
 476           to figure out what they really mean and how to get around (but not really solve) the problem. Then I get another error.
 477           The CAst-based slicer works on a stupid tiny test program but it's never going to work on real code. The problem of
 478           disappearing "x=y" type things in Shrike is insignificant compared to trying to actually get CAst work on something useful
 479           -- much less the fact that it only works for Java <= 1.4.
 480
 481 2007-11-16
 482     GPSYLON BINARY:
 483         * Gives the JapaneseImperialCalendar error wherever I seed the slice from ANYWHERE the Gpsylon class.
 484         * Searched source for "calendar", there is some code in other .class files (not the .java file being
 485           analyzed) that uses a calendar. Added code to use calendars how they are used there (create them,
 486           get the date, set fields) and added to delmenow/Main.java (analyzed) and delmenow/Olvidao.java
 487           (in a function used by Main.java / in the slice, included as binary). But slicing this works...
 488     POLYGLOT BINARY:
 489         * Since I've figured out things since I last tried it (ShrikeClass cast thing), I tried polyglot again
 490           using Main.java source but other polyglot files as binaries. This time I get an "out of memory" error.
 491         * I somehow got it to make a slice in the main function with added x=x+8 type code!
 492         * Trying on some real code isn't working ... can't find seed statement (although it finds the method).
 493           Seems to be nondeterministic. If the array it makes to hold the statements is too small (I don't see
 494           how that could be because it counts the statements it will take and then takes them), it gets fine to
 495           that point. If I make the array bigger, it won't even get to that point, but will be unable to find
 496           the seed statements! (I think it's just random.) Finally I made it not add elements to the array if
 497           it's not big enough. Then I got a slice (!!!), but it seems to be practically the whole file...
 498         * Talked with Ras
 499         * Talked with Hsiu-fan about problem with plugin with Windows,
 500           JDT AST instead of Polyglot?
 501         * Use ArrayList instead of calculating number of slices used then getting them (2 loops)
 502         * Talked more with Hsiu-fan. Setting up git repository & installing git.
 503         * Talked in again with Ras. Will look up why extra things are getting in the slice.
 504
 505 2007-11-19
 506         * Analyzing SDGs to figure out what is wrong with the CI slicer.
 507           Maybe I don't fully understand the algorithm.
 508           In scanForMod(), what does it mean to modify a pointer key?
 509         * Learned a bit about git, set up git repository, uploaded stuff.
 510         * Dave set up my username on cobol.cs.berkeley.edu CVS server,
 511           so I downloaded the packages Manu mentioned. Looking at
 512           SDGThinSlicerDriver.java, BugTestInfo.java
 513         * (Distacted by Dave's Mozilla compiling issues)
 514         * Back to investigating CI Slicer, why it gets the extra
 515           "x=new Foo()" line. What does it mean to mod a pointer key?
 516         * Realized there is a InstanceFieldKey representing not an
 517           abstract object but an abstract object's field.
 518
 519 2007-11-20
 520         * OK, so I think the issue is that a = new Foo() sets up all the
 521           instance field keys and by default modifies all of them.
 522           Because after all, it initializes them by creating them (even
 523           if it gives them an undefined value). However if they really
 524           are initialized in some other way ("int d = 7" in the class
 525           definition, or in a constructor, etc.) this is also
 526           highlighted already. So Making the a=new Foo() statement
 527           modify it is really redundant.
 528         * Oops, pointer keys are actually the abstraction of "variables",
 529           whereas INSTANCE keys are the abstraction of objects. Makes a
 530           bit more sense now. A pointer key is a variable or field of an
 531           abstract object.
 532         * Thought I found the cause in ModRef.java:286 -- for "new"
 533           instructions, it adds all instance field keys as pointer keys
 534           that the instructions "mods" or "writes". This can be disabled
 535           directly or by setting PDG.IGNORE_ALLOC_HEAP_DEFS to true (may
 536           have other side effects in PDG.java) This fixes it for the CS
 537           slicer but something more sinister seems to be happening in the
 538           CI slicer, in fact, slicing from x.y.z, x = new Foo() is in the
 539           slice!
 540         * CHANGED in PDG.java:77
 541         * It looks like the problem with the CI slicer only happens with
 542           the CAst representation... Oops, forgot to change from NO_HEAP
 543           to NO_BASE_NO_HEAP in the CAst CISlicer.
 544         * Timed the CISlicer. Making the call graph (and I assume also
 545           pointer analysis took 5.455 seconds, while *everything* else
 546           took 0.237 seconds. Another: CG 5.894, Other 0.203, Total 6.097
 547                 test code (foo has one bar called f, bar has integers b,c,d):
 548                         Foo p = new Foo();
 549                         Foo p2 = p;
 550                         Bar r = new Bar();
 551                         p.f = r;
 552                         Bar z = p2.f;
 553                         p2.f.d = 3;
 554
 555                         p2.f.c = 4;
 556                         z.b = 6;
 557
 558                         z.d=6;
 559                         z.d=7;
 560                         z.d=8;
 561                         z.d=9;
 562
 563                         System.out.println(z.d);
 564         * CS Thin Slicer: 6.697, 6.993
 565           CS Full Slicer: 7.605, 8.004
 566            some random variation but call graph always takes up 90%+ of time
 567         * Tracing what the call graph builder does.
 568                 CallGraphBuilder is a
 569                 com.ibm.wala.cast.java.ipa.callgraph.AstJavaZeroXCFABuilder
 570         * Class hierarchy takes 4.730s, buildCallGraph() 1.797
 571           Class hierarchy takes 3.600s, buildCallGraph() 1.629
 572         * Spent some time trying to get TPTP Eclipse profiler to work...
 573           No luck...
 574         * Class hierarchy stuff done in
 575                 JavaSourceAnalysisEngine.buildClassHierarchy()
 576           probably loading classes and reading files, other stuff
 577           building analyses?
 578           Question -- how much of that time is using Polyglot to
 579           parse that file?
 580         * Ran hprof but that didn't give me much useful...
 581           I want something which gives times of every function.
 582           I.e. it should show main as 100%.
 583         * Ran JIP and got some interesting results (see
 584           ~/home/evan/walacastwork/OperationChao/profile.txt,
 585           try viewing w/ grep -v '0..          |'  profile.txt |less
 586                 Summary: issues in JavaSourceAnalysisEngine:buildClassHierarchy:
 587                         * Most of this in ClassLoaderImpl:init (thru PolyglotClassLoaderFactory:makeNewClassLoader)
 588                         * At least half of this from ClassLoaderImpl:loadAllClasses / ShrikeClass<init> -- loading system .jars etc. I assume.
 589                 Issues in AbstractAnalysisEngine:buildCallGraph:
 590                         * All from StandardSolver:solve -> SSAPropagationCallGraphBuilder:unconditionallyAddConstraintsFromNode
 591           The profiler is giving me different relative numbers than my own tests gave me
 592           (slicing itself is taking longer than the rest!), but I think it is fair so say that both loading
 593           the classes and doing the pointer analysis are quite significant bottlenecks.
 594         * Working on cleaning up / reorganizing useful parts of slicer plugin
 595           to make it more readable/usable/maintainable.
 596
 597 2007-11-21
 598         * Cleaned up a bit more, reorganized into edu.berkeley.cs.bodik.svelte packages.
 599         * Looked back at the Polyglot from source issue. Found out about using 'jlc' in
 600           the polyglot distribution to test files. Tested the Grm.java file with this --
 601           still gave StackOverFlow error. Narrowed down the problem to big list of
 602           concatenated strings ("bla" + "bla" + "bla" + "bla" ...). Passed Java VM arg
 603           -Xss1024k to change stack size and jlc worked. 512k, maybe even smaller, ok
 604         * Now I have the same problem as I did with analyzing the gpsylon source (full stack trace)
 605                 !ENTRY org.eclipse.ui 4 0 2007-11-21 13:54:25.099
 606                 !MESSAGE Unhandled event loop exception
 607                 !STACK 0
 608                 com.ibm.wala.util.debug.UnimplementedError: getCAstTypeFor() passed type that is not primitive, array, or class?but type(null)
 609                         at com.ibm.wala.util.debug.Assertions.UNREACHABLE(Assertions.java:87)
 610                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotTypeDictionary.getCAstTypeFor(PolyglotTypeDictionary.java:97)
 611                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator$JavaTranslatingVisitorImpl.visit(PolyglotJava2CAstTranslator.java:665)
 612                         at com.ibm.wala.cast.java.translator.polyglot.ASTTraverser.visit(ASTTraverser.java:116)
 613                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkNodes(PolyglotJava2CAstTranslator.java:2621)
 614                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator$JavaTranslatingVisitorImpl.visit(PolyglotJava2CAstTranslator.java:614)
 615                         at com.ibm.wala.cast.java.translator.polyglot.ASTTraverser.visit(ASTTraverser.java:112)
 616                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkNodes(PolyglotJava2CAstTranslator.java:2621)
 617                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator$JavaTranslatingVisitorImpl.visit(PolyglotJava2CAstTranslator.java:1220)
 618                         at com.ibm.wala.cast.java.translator.polyglot.ASTTraverser.visit(ASTTraverser.java:190)
 619                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkNodes(PolyglotJava2CAstTranslator.java:2621)
 620                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator$JavaTranslatingVisitorImpl.visit(PolyglotJava2CAstTranslator.java:956)
 621                         at com.ibm.wala.cast.java.translator.polyglot.ASTTraverser.visit(ASTTraverser.java:162)
 622                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkNodes(PolyglotJava2CAstTranslator.java:2621)
 623                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator$JavaTranslatingVisitorImpl.visit(PolyglotJava2CAstTranslator.java:280)
 624                         at com.ibm.wala.cast.java.translator.polyglot.ASTTraverser.visit(ASTTraverser.java:88)
 625                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkNodes(PolyglotJava2CAstTranslator.java:2621)
 626                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkEntity(PolyglotJava2CAstTranslator.java:2549)
 627                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.processClassMembers(PolyglotJava2CAstTranslator.java:2440)
 628                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkEntity(PolyglotJava2CAstTranslator.java:2524)
 629                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.walkEntity(PolyglotJava2CAstTranslator.java:2516)
 630                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotJava2CAstTranslator.translate(PolyglotJava2CAstTranslator.java:2359)
 631                         at com.ibm.wala.cast.java.translator.Java2IRTranslator.translate(Java2IRTranslator.java:57)
 632                         at com.ibm.wala.cast.java.translator.polyglot.JavaIRPass.run(JavaIRPass.java:37)
 633                         at polyglot.frontend.Scheduler.runPass(Scheduler.java:596)
 634                         at polyglot.frontend.Scheduler.runGoal(Scheduler.java:499)
 635                         at polyglot.frontend.Scheduler.attemptGoal(Scheduler.java:440)
 636                         at polyglot.frontend.Scheduler.attemptGoal(Scheduler.java:412)
 637                         at polyglot.frontend.Scheduler.attemptGoal(Scheduler.java:364)
 638                         at polyglot.frontend.Scheduler.runToCompletion(Scheduler.java:297)
 639                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotFrontEnd.compile(PolyglotFrontEnd.java:72)
 640                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotSourceModuleTranslator.loadAllSources(PolyglotSourceModuleTranslator.java:114)
 641                         at com.ibm.wala.cast.java.loader.JavaSourceLoaderImpl.loadAllSources(JavaSourceLoaderImpl.java:400)
 642                         at com.ibm.wala.classLoader.ClassLoaderImpl.init(ClassLoaderImpl.java:367)
 643                         at com.ibm.wala.cast.java.translator.polyglot.PolyglotClassLoaderFactory.makeNewClassLoader(PolyglotClassLoaderFactory.java:40)
 644                         at com.ibm.wala.classLoader.ClassLoaderFactoryImpl.getLoader(ClassLoaderFactoryImpl.java:66)
 645                         at com.ibm.wala.classLoader.ClassLoaderFactoryImpl.getLoader(ClassLoaderFactoryImpl.java:64)
 646                         at com.ibm.wala.ipa.cha.ClassHierarchy.<init>(ClassHierarchy.java:185)
 647                         at com.ibm.wala.ipa.cha.ClassHierarchy.<init>(ClassHierarchy.java:145)
 648                         at com.ibm.wala.ipa.cha.ClassHierarchy.make(ClassHierarchy.java:1118)
 649                         at com.ibm.wala.cast.java.client.JavaSourceAnalysisEngine.buildClassHierarchy(JavaSourceAnalysisEngine.java:151)
 650                         at com.ibm.wala.client.impl.AbstractAnalysisEngine.defaultCallGraphBuilder(AbstractAnalysisEngine.java:331)
 651                         at com.ibm.wala.client.impl.AbstractAnalysisEngine.buildDefaultCallGraph(AbstractAnalysisEngine.java:343)
 652                         at edu.berkeley.cs.bodik.svelte.TempSlicingDriver.getPositionsOfThinSliceForLineNumInMain(TempSlicingDriver.java:257)
 653           In this function getCAstTypeFor, polyglotType seems to be null (or .toString() is null?)
 654           The last entity walked is "final public polyglot.ast.Branch Break(...)", a function in AbstractNodeFactory_c.java
 655           This seems to be the same.
 656         * This happens even if I replace the source directory tree module with many
 657           Source file modules.
 658         * I can't include files as .class files (some error mentioning polyglot, not a .jar or .java file), but
 659           I can add some class files to a jar, load the jar and not load the .java file. Doing this for AbstractNodeFactory_c.java
 660           Seems to make problems with this files go away. We should still get slice results from this file, but they will not have
 661           column information, only line number information.
 662         * Now it takes a LONG time to run... I'm killing it... It could be working.
 663         * Simplified a test case in which polyglot example fails.
 664
 665 2007-11-26 (note: wired ethernet doesn't work, using laptop for online stuff)
 666         * Debugged latest Polyglot source test... never stops running.
 667           Stopped and found that it was printing a cast node for an assert
 668           debugging statement... calculating a string that never gets
 669           printed. Many iterations of CAstPrinter.printTo() but it's not
 670           a stack overflow / infinite recursion... just takes a long time.
 671           Modified assert statement to not call CAstPrinter.print()
 672           (...wala.cast.ir.translator.AstTranslator.java:2623
 673         * Now I get a stack overflow... increasing to 2048k, now 8192k.
 674         * Now I get the same problem as before ... "getCAstTypeFor() passed type that is not primitive, array, or class?"
 675         * Adding class file to 'problem classes' jar file, and continuinug...
 676         * NullPointerException on MyFunFunWalaCastSlicingTest.java:965 -- can't find the main method
 677         * Oops, was using different location of the polyglot source library
 678         * GOT A SLICE!!! Not half bad, either.
 679         * Reorganizing code. After thinking a little bit, making class SvelteAnalysisEngine
 680         * Changing everything to make plugin use SvelteAnalysisEngine. Now can't rind the regression file...
 681
 682 2007-11-27
 683         * After figuring out bug with loading system jars, got slice of Polyglot with new svelte package.
 684         * Get Hsiu-fan's auto-call-graph building from Git.
 685         * After merging source bases, it seems the call graph is being build four times! That takes 4 minutes each for Polyglot...
 686         * This seems to happen with Hsiu-fan's original version too. Static variables to stop this don't work.
 687         * Got new computer today. Plugged in and ran ... need to set up w/ network first, can't have two computers on same network
 688           port... I'll keep using this one until we decide which OSs to put on.
 689         * Testing on old Main.java test to investigate rebuilding of call graph issue...
 690         * Make Auto-cg button a check button
 691         * Upload stuff to git
 692
 693 2007-11-28
 694         * Fixed "one save, multiple resourceChanged callbacks" problem -- kind of -- still runs twice,
 695           but does a System.currentTimeMillis() check to prevent building call graph twice.
 696         * Tried out cached call graph slicing on Polyglot. Looks good. But now I have to solve the issue
 697           of multiple files. Revisiting the issues of markers and annotation.
 698         * Use marker with the annotation. No multiple files yet, and we have to delete this.
 699         * The changing of markers due to the slicing operation is causing the CG to be rebuilt!
 700         * Fixed this, and the earlier "one save, multiple resourceChanged callbacks" -- changed markers
 701           were the culprit back then too (when a file is compiled, its markers change, setting off
 702           a resource change event). Check that the CONTENT of a .java file has changed.
 703         * Markers aren't as persistent as I thought... when I add an error marker it stays,
 704           but when I subclass the error marker and make that it doesn't... how useless..
 705         * Fixed -- markers now persistent. Using problem markers, but level "info". Shows up
 706           in Problems view, under Infos. OK for now.
 707         * Figured out how to get a path from a filename string and now multiple files works! Yay!
 708         * Going to look at Manu's tests from the paper from SIR. Specifically, the jtopas FAULT_5 test
 709           in v1 or jtopas (see jtopas/versions.alt/seeded/v1/src/de/susebox/java/util)
 710         * Found the bug and all, but there is no main() ... have to find/look at the JUnit test for the entrypoints.
 711         * Make a main which run testLinkParsing(). Change testLinkParsing() to read a test (test5.html for now).
 712
 713 2007-11-29
 714         * Class hierarchy builds, but cannot find entrypoint AbstractTokenizer.main... oops, had it in wrong class.
 715         * Slicing seems to work...
 716         * Just noticed that this bug, JTOPAS_1 (jtopas v1 FAULT_5), seems to have the same "seed" and "bug" statements.
 717           Trying JTOPAS_2 (jtopas v1 FAULT_6)
 718         * Looks like it works as expected... This is kind of a "bug of omission" (setLength instead of setType in test4Normal,
 719           then later we check if type==KEYWORD)... I'm not sure what Manu's target "bug" was, but this gets the setType() lexically close.
 720         * Looked at EclipseProjectSourceAnalysisEngine, and IClasspasthEntry documentation
 721         * Still not exactly sure the best way but wrote code to load analysis scope from Eclipse project class paths.
 722         * Looked at nanoxml example... call graph builds nicely, but I don't feel like figuring out how the JUnit test works to make a main()
 723         * ant example won't work, can't find import org.apache.bcel.classfile.* in org/apache/tools/ant/filters/util/JavaClassHelper.java (v7) ...
 724           Manu's test .jar doesn't even have this file...
 725         * Working on making my code general enough to support Shrike OR CAst. Slight problem: can't get source filename from a Shrike Method
 726           (thus a statement in the slice). Will have to record class name and use something to get the source filename from that.
 727         * Get filename from classname by looking thru sources for the presumed filename (Lcom/example/delmenow/Olvidao$Foo ->
 728           com/example/delmenow/Olvidao.java) Cheap, but works. Need hashtable in future.
 729         * Yay! Shrike OR CAst slicing from the plug-in!
 730
 731 2007-11-30
 732         * Cache on a project basis. Now we can start slice from different file than main()
 733         * Shrike or CAst button to switch off. Fix bug in CAst slicing that way.
 734         * Seems like more statements in slices now with CAst...?
 735         * Used git to go back and check.... Yea, less slices before. Investigating...
 736         * Fixed bug -- conditionals and gotos can cover more than one line. So it was getting way more
 737           statements in the seed (it looks for seed statements by comparing line numbers)
 738         * Fixed bug in which cache not being rebuilt on re-save.
 739         * Small test that slice from different file works and gets things in other files.
 740         * Started "Call Graph in its own job" thing
 741
 742 2007-12-03
 743         * Fix "Call Graph in its own job"
 744         * Read up about launch configurations. We can use these to determine the entrypoint for building a call graph.
 745         * If the Java file which induces the call graph building isn't mentioned as a main class in a launch configuration,
 746           use a main class that is. I'd really like to do exactly what clicking on the run and debug buttons do, and get a
 747           list of launch configurations equivalent to it's drop-down menu from the toolbar.
 748         * Trying to slice with old version slicing plugin with Shrike. Didn't seem to give any errors
 749           when not loading WALA libraries! Seems like it should, since they are used in the project. Gives slice,
 750           but of course doesn't include any WALA code.
 751         * Added code for Eclipse classpath project dependencies so bin/ directories of WALA would be incuded. Takes only 36 sec
 752           to build class hierarchy, but then gives out of memory error later on in building call graph.
 753         * Added Eclipse libs to exclusion file. Takes 12 sec to build class hierarchy. Call graph builds. Slice still doesn't
 754           seem to take any WALA code into account. hmmm...
 755         * Maybe I was chooisng a bad seed statement... Choosing a different got me a slice with over 65,000 statements (too
 756           many to see with present marker code...)
 757         * Changed slightly to get an intersting seed statement that works:
 758                 CallGraph cg = engine.buildDefaultCallGraph();
 759                 System.out.println(cg); // seed here
 760           This gets a slice with 11 statements (many duplicates), in this file and 3 WALA files.
 761           It correctly figures out that the CallGraphBulider is a PropagationCallGraphBuilder, traces
 762           the flow of variable callgraph thru from the return statement of makeCallGraph() to its
 763           assignment in the constructor, etc.
 764         * Got line numbers for seed & bug statements of Manu's benchmarks by modifying his sanity check program,
 765           but they are near useless because I don't have the source they came from... Tried just running cpp to
 766           add the fault for jtopas, but the line number is off by 6 lines... Also, ant fault DF_HD_1 doesn't work:
 767                 com.ibm.wala.util.debug.UnimplementedError: couldn't find node for method Node: < Application, Lorg/apache/tools/ant/taskdefs/Definer, setName(Ljava/lang/String;)V > Context: Everywhere
 768         * Decided to look a JTopas test a little closer using the Sanity Check program. Strangely, Manu's bug statement is
 769           recorded as the conditional branch:
 770             if (prop != null) {
 771           //    token.setType(Token.KEYWORD);
 772                 // FAULT_6
 773               token.setLength(Token.KEYWORD);
 774             } else {
 775               token.setType(Token.NORMAL);
 776             }
 777           I wonder if he taught it to follow a control dependency back?
 778         * Prevent multiple workspasce builds / resourceChanged calls by using a IWorkspaceRunnable to add all markers in 1 batch
 779         * Looking more into converting line + col numbers into offset numbers. I think I want/need a ILineTracker.
 780           It looks like when I add a marker with an offset it doesn't give me a line number in the list of "problems", just says
 781           Location: Unknown. You can add a line number and i will show up but it doesn't have to match up with the offset
 782           numbers. I wonder if compile errors use both line and offset numbers to show both line number and exact position?
 783         * I think our basic options are:
 784                 1) show only line numbers  (disappointing)
 785                 2) open every source file we want column/offset numbers for and use a line tracker (slow)
 786                 3) show only line numbers for the time being, but make a hook for when opening
 787                    a file, use the linetracker (aassuming it is created) and modify all the markers.
 788                    this is kind of a hack, but has the upside that in the future we may want to do
 789                    other things here, like link back to an eclipse AST node.
 790                 4) modify WALA/Polyglot? to give offsets instead of/as well as columns (might not be too bad)
 791                         -> hmm, can't find this in the code... all 2 of the anon classes "new AbstractSourcePosition() {}"
 792                            are unused (?!) and the other has identical first and last line/col.
 793
 794 2007-12-04
 795         * Slicing itself in a job
 796         * Looking up about cancelling. CancelException / MonitorUtil not used much throughout WALA yet.
 797         * Updating WALA from SVN to see if CancelException used more now. No, not really. Doesn't look like it.
 798           Maybe I should wait.
 799         * Looking into locking of resources and issues related to concurrency.
 800         * Slicing and call graph building now lock main project. Works, but more thinking may need to be done.
 801           At least now I can't make a slice using a call graph for the project is being built.
 802         * Don't show duplicate line numbers -- on slicing, make a hash table with (lineno + " " + filename)
 803         * Looked into issue of getting back copies, like "y=x". Tried undoing copy propagation with with
 804           SSAConversion.undoCopyPropagation(). This makes new variable numbers and adds assignment instructions
 805           which use them into the IR... but slicing after this doesn't work (DefUse's uses[] variable isn't updated,
 806           tries to access out of bounds index for newly created variable...) I added a hack which makes new DefUses
 807           for the changed CGNodes... this doesn't stop the slicer from working right if the node's IR is unmodified,
 808           but it still doesn't work  (just null slice) when taking a slice after doing this after undoing copy prop.
 809         * Changed saving to build once for each project in which files have changed... But it seems not to matter,
 810           since "Save All" saves files one by one, individually, anyway.
 811         * Trying to slice the int  = pos.getFirstCol(); doesn't enter function getFirstCol -- why?
 812                 private static void addStatement(Statement s, int rv[][], int i)
 813                 {
 814                         if (s != null && s.getKind() == Statement.Kind.NORMAL && s.getNode().getMethod() instanceof ConcreteJavaMethod) {
 815                         // ignore special kinds of statements
 816                                 int instructionIndex = ((NormalStatement) s).getInstructionIndex();
 817                                 Position pos = ((ConcreteJavaMethod) s.getNode().getMethod()).getSourcePosition(instructionIndex);
 818                                 rv[i][0] = pos.getFirstLine();
 819                                 rv[i][1] = pos.getFirstCol();
 820                                 rv[i][2] = pos.getLastLine();
 821                                 rv[i][3] = pos.getLastCol();
 822                                 int x = pos.getFirstCol();
 823                                 System.out.println(x);
 824                         }
 825                 }
 826         Was hoping it would shed light on where Position is coming from. Instead pretty much useless. gets a bunch of random stuff.
 827         Doesn't include the function getFirstCol() in that actual class (PolyglotSourcePosition)
 828         Going outside of function and slicing from "rv" brings up more junk. Taking slice from x = pos.getFirstCol() gives almost nothing (looking at just pos)
 829
 830 2007-12-05
 831         * Changed Position and PolyglotSourcePosition to give an offset, but Polyglot isn't giving an offset!
 832                 (what did: just changed Position interface to have getFirstOffset() and getLastOffset() functions,
 833                  then fixed errors -- all classes that didn't implement it -- mostly just return -1, except PolyglotSourcePosition
 834                  should return p.offset() and p.endOffset())
 835         * Changed Polyglot to give an offset. Had to change JFlex directive. In java.flex
 836                 --- java.flex.old       2007-12-05 10:03:25.000000000 -0800
 837                 +++ java.flex   2007-12-05 09:46:13.000000000 -0800
 838                 @@ -32,6 +32,7 @@
 839
 840                  %line
 841                  %column
 842                 +%char
 843
 844                  %state STRING, CHARACTER, TRADITIONAL_COMMENT, END_OF_LINE_COMMENT
 845
 846                 @@ -113,7 +114,7 @@
 847
 848                      private Position pos() {
 849                          return new Position(path, file, yyline+1, yycolumn, yyline+1,
 850                 -                            yycolumn + yytext().length());
 851                 +                            yycolumn + yytext().length(), yychar, yychar + yytext().length());
 852                      }
 853
 854                      private Position pos(int len) {
 855         * Move SvelteExclusions.xml to a dat/ directory. Hopefully it'll stay put here.
 856         * Email Polyglot mailing list about patch.
 857         * Change presentationLayer attribute to a higiher layer so occurence markers (grey highlight) won't cover up slice markers (in body & vertical ruler)
 858         * Change default annotation to a box; don't remove duplicate per line for CAst (offset information)
 859         * After a bit of thinking & investigating: for each invoke instruction used as a seed, don't use that instruction but use the return instruction corresponding to
 860           that call site. Before, seeding a slice with something like "x = foo()" wouldn't bring up anything, because no variables are USED in the instruction.. Now it
 861           the variable used is the output of the function (the same as x)
 862         * Don't use source positions until the end... show more information in message for each slice statement -- whether seed or not, also full statement code