JavaScriptCore/ChangeLog

   1 2007-11-03  David D. Kilzer  <ddkilzer@webkit.org>
   2
   3         Sort files(...); sections of Xcode project files.
   4
   5         Rubber-stamped by Darin.
   6
   7         * JavaScriptCore.xcodeproj/project.pbxproj:
   8
   9 2007-11-03  Maciej Stachowiak  <mjs@apple.com>
  10
  11         Reviewed by Oliver.
  12
  13         - remove VarDeclListNode and simplify VarDeclNode evaluation for 0.4% SunSpider speedup
  14
  15         * kjs/grammar.y:
  16         * kjs/nodes.cpp:
  17         (KJS::VarDeclNode::optimizeVariableAccess):
  18         (KJS::VarDeclNode::getDeclarations):
  19         (KJS::VarDeclNode::handleSlowCase):
  20         (KJS::VarDeclNode::evaluateSingle):
  21         (KJS::VarDeclNode::evaluate):
  22         (KJS::VarStatementNode::execute):
  23         * kjs/nodes.h:
  24         (KJS::VarDeclNode::):
  25         (KJS::VarStatementNode::):
  26         * kjs/nodes2string.cpp:
  27         (KJS::VarDeclNode::streamTo):
  28
  29 2007-11-03  Alexey Proskuryakov  <ap@webkit.org>
  30
  31         Reviewed by Darin.
  32
  33         http://bugs.webkit.org/show_bug.cgi?id=15800
  34         REGRESSION (r27303): RegExp leaks
  35
  36         * kjs/regexp_object.h:
  37         (KJS::RegExpImp::setRegExp):
  38         (KJS::RegExpImp::regExp):
  39         (KJS::RegExpImp::classInfo):
  40         * kjs/regexp_object.cpp:
  41         (RegExpImp::RegExpImp):
  42         (RegExpImp::~RegExpImp):
  43         Renamed reg member variable to m_regExp, changed it to use OwnPtr.
  44
  45 2007-11-02  Maciej Stachowiak  <mjs@apple.com>
  46
  47         Reviewed by Oliver.
  48
  49         - add SourceElements as a typedef for Vector<RefPtr<StatementNode> >.
  50
  51         * kjs/grammar.y:
  52         * kjs/nodes.cpp:
  53         (KJS::statementListPushFIFO):
  54         (KJS::statementListGetDeclarations):
  55         (KJS::statementListInitializeDeclarationStacks):
  56         (KJS::statementListInitializeVariableAccessStack):
  57         (KJS::statementListExecute):
  58         (KJS::BlockNode::BlockNode):
  59         (KJS::FunctionBodyNode::FunctionBodyNode):
  60         (KJS::ProgramNode::ProgramNode):
  61         * kjs/nodes.h:
  62         (KJS::CaseClauseNode::):
  63
  64 2007-11-02  Darin Adler  <darin@apple.com>
  65
  66         Reviewed by Maciej.
  67
  68         - http://bugs.webkit.org/show_bug.cgi?id=15791
  69           change property map data structure for less memory use, better speed
  70
  71         The property map now has an array of indices and a separate array of
  72         property map entries. This slightly slows down lookup because of a second
  73         memory acess, but makes property maps smaller and faster to iterate in
  74         functions like mark().
  75
  76         SunSpider says this is 1.2% faster, although it makes the bitwise-end test
  77         more than 10% slower. To fix that we'll need to optimize global variable lookup.
  78
  79         * kjs/property_map.cpp:
  80         (KJS::PropertyMapEntry::PropertyMapEntry):
  81         (KJS::PropertyMapHashTable::entries):
  82         (KJS::PropertyMapHashTable::allocationSize):
  83         (KJS::SavedProperties::SavedProperties):
  84         (KJS::SavedProperties::~SavedProperties):
  85         (KJS::PropertyMap::checkConsistency):
  86         (KJS::PropertyMap::~PropertyMap):
  87         (KJS::PropertyMap::clear):
  88         (KJS::PropertyMap::get):
  89         (KJS::PropertyMap::getLocation):
  90         (KJS::PropertyMap::put):
  91         (KJS::PropertyMap::insert):
  92         (KJS::PropertyMap::createTable):
  93         (KJS::PropertyMap::rehash):
  94         (KJS::PropertyMap::remove):
  95         (KJS::PropertyMap::mark):
  96         (KJS::comparePropertyMapEntryIndices):
  97         (KJS::PropertyMap::containsGettersOrSetters):
  98         (KJS::PropertyMap::getEnumerablePropertyNames):
  99         (KJS::PropertyMap::save):
 100         (KJS::PropertyMap::restore):
 101         * kjs/property_map.h:
 102
 103 2007-11-02  Darin Adler  <darin@apple.com>
 104
 105         Reviewed by Maciej.
 106
 107         - http://bugs.webkit.org/show_bug.cgi?id=15807
 108           HashMap needs a take() function that combines get and remove
 109
 110         * wtf/HashMap.h: Added take function. Simplistic implementation for now,
 111         but still does only one hash table lookup.
 112
 113         * kjs/array_instance.cpp: (KJS::ArrayInstance::put): Use take rather than
 114         a find followed by a remove.
 115
 116 2007-11-02  David Carson  <dacarson@gmail.com>
 117
 118         Reviewed by Darin.
 119
 120         Fix compiler warning "warning: suggest parentheses around && within ||"
 121         http://bugs.webkit.org/show_bug.cgi?id=15764
 122
 123         * kjs/value.h: (KJS::JSValue::isNumber): Add parentheses.
 124
 125 2007-11-01  Geoffrey Garen  <ggaren@apple.com>
 126
 127         Reviewed by Maciej Stachowiak.
 128
 129         In preparation for making List a simple stack-allocated Vector:
 130
 131         Removed all instances of List copying and/or assignment, and made List
 132         inherit from Noncopyable.
 133
 134         Functions that used to return a List by copy now take List& out
 135         parameters.
 136
 137         Layout tests and JS tests pass.
 138
 139         * kjs/list.cpp:
 140         (KJS::List::slice): Replaced copyTail with a more generic slice
 141         alternative. (JavaScriptCore only calls slice(1), but WebCore calls
 142         slice(2)).
 143
 144 2007-11-01  Geoffrey Garen  <ggaren@apple.com>
 145
 146         Reviewed by Maciej Stachowiak.
 147
 148         Fixed http://bugs.webkit.org/show_bug.cgi?id=15785
 149         REGRESSION(r27344): Crash on load at finance.yahoo.com
 150
 151         Reverted a small portion of my last check-in. (The speedup and the List
 152         removal are still there, though.)
 153
 154         ActivationImp needs to hold a pointer to its function, and mark that
 155         pointer (rather than accessing its function through its ExecState, and
 156         counting on the active scope to mark its function) because a closure
 157         can cause an ActivationImp to outlive its ExecState along with any
 158         active scope.
 159
 160         * kjs/ExecState.cpp:
 161         (KJS::ExecState::ExecState):
 162         * kjs/function.cpp:
 163         (KJS::FunctionImp::~FunctionImp):
 164         (KJS::ActivationImp::ActivationImp):
 165         * kjs/function.h:
 166         (KJS::ActivationImp::ActivationImpPrivate::ActivationImpPrivate):
 167
 168         Also made HashTable a little more crash-happy in debug builds, so
 169         problems like this will show up earlier:
 170
 171         * wtf/HashTable.h:
 172         (WTF::HashTable::~HashTable):
 173
 174 2007-11-01  Geoffrey Garen  <ggaren@apple.com>
 175
 176         Reviewed by Adam Roben.
 177
 178         Addressed some of Darin's review comments.
 179
 180         Used perl -p, which is the shorthand while(<>) {}.
 181
 182         Made sure not to suppress bison's output.
 183
 184         Added line to removed bison_out.txt, since this script removes other
 185         intermediate files, too.
 186
 187         * DerivedSources.make:
 188
 189 2007-11-01  Geoffrey Garen  <ggaren@apple.com>
 190
 191         Reviewed by Oliver Hunt.
 192
 193         Removed List from ActivationImp, in preparation for making all lists
 194         stack-allocated.
 195
 196         Tests pass.
 197
 198         1.0% speedup on SunSpider, presumably due to reduced List refcount thrash.
 199
 200         * kjs/ExecState.cpp:
 201         (KJS::ExecState::ExecState):
 202         (KJS::ExecState::~ExecState):
 203         * kjs/function.cpp:
 204         (KJS::ActivationImp::ActivationImp):
 205         (KJS::ActivationImp::createArgumentsObject):
 206         * kjs/function.h:
 207         (KJS::ActivationImp::ActivationImpPrivate::ActivationImpPrivate):
 208
 209 2007-11-01  Adam Roben  <aroben@apple.com>
 210
 211         Use jsNumberCell instead of jsNumber when converting double constants to JSValues
 212
 213         This fixes fast/js/math.html, ecma/Date/15.9.5.10-1.js, and
 214         ecma/Date/15.9.5.12-1.js, which were suffering from a bug in MSVC.
 215
 216         It also gets rid of an MSVC warning that we previously had to silence.
 217
 218         Reviewed by Geoff.
 219
 220         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj: Turn
 221         back on the "overflow in constant arithmetic" warning.
 222         * kjs/number_object.cpp:
 223         (NumberObjectImp::getValueProperty): Use jsNumberCell instead of
 224         jsNumber.
 225
 226 2007-10-31  Adam Roben  <aroben@apple.com>
 227
 228         Windows build fix
 229
 230         * kjs/ExecState.h:
 231
 232 2007-10-31  Maciej Stachowiak  <mjs@apple.com>
 233
 234         Reviewed by Oliver.
 235
 236         - shave some cycles off of local storage access for a 1% SunSpider speedup
 237
 238         Keep the LocalStorage pointer in the ExecState, instead of getting
 239         it from the ActivationImp all the time.
 240
 241         * kjs/ExecState.cpp:
 242         (KJS::ExecState::updateLocalStorage):
 243         * kjs/ExecState.h:
 244         (KJS::ExecState::localStorage):
 245         * kjs/nodes.cpp:
 246         (KJS::LocalVarAccessNode::evaluate):
 247         (KJS::LocalVarFunctionCallNode::evaluate):
 248         (KJS::PostIncLocalVarNode::evaluate):
 249         (KJS::PostDecLocalVarNode::evaluate):
 250         (KJS::LocalVarTypeOfNode::evaluate):
 251         (KJS::PreIncLocalVarNode::evaluate):
 252         (KJS::PreDecLocalVarNode::evaluate):
 253         (KJS::ReadModifyLocalVarNode::evaluate):
 254         (KJS::AssignLocalVarNode::evaluate):
 255         (KJS::FunctionBodyNode::processDeclarationsForFunctionCode):
 256
 257 2007-10-31  Adam Roben  <aroben@apple.com>
 258
 259         Fix a crash on launch due to a static initializer race
 260
 261         We now use fast inline assembler spinlocks which can be statically
 262         initialized at compile time.
 263
 264         As a side benefit, this speeds up SunSpider by 0.4%.
 265
 266         Reviewed by Oliver.
 267
 268         * wtf/FastMalloc.cpp:
 269         * wtf/TCSpinLock.h:
 270         (TCMalloc_SpinLock::Lock):
 271         (TCMalloc_SpinLock::Unlock):
 272         (TCMalloc_SlowLock):
 273         * wtf/TCSystemAlloc.cpp:
 274
 275 2007-10-31  Kevin McCullough  <kmccullough@apple.com>
 276
 277         Reviewed by Sam.
 278
 279         - Corrected spelling.
 280
 281         * wtf/HashTraits.h:
 282
 283 2007-10-31  Mark Rowe  <mrowe@apple.com>
 284
 285         Further Gtk build fixage.
 286
 287         * kjs/regexp_object.cpp:
 288
 289 2007-10-31  Mark Rowe  <mrowe@apple.com>
 290
 291         Gtk build fix.
 292
 293         * kjs/regexp.h:
 294
 295 2007-10-31  Darin Adler  <darin@apple.com>
 296
 297         Reviewed by Maciej.
 298
 299         - fix http://bugs.webkit.org/show_bug.cgi?id=15749
 300           RegExp/RegExpObjectImp cause needless UString creation
 301
 302         Speeds things up 0.4% according to SunSpider.
 303
 304         * kjs/config.h: Define USE(PCRE16) instead of HAVE(PCREPOSIX),
 305         because this library doesn't use the real PCRE -- it uses its
 306         own PCRE that works on UTF-16.
 307
 308         * kjs/regexp.h: Removed a few unused functions. Changed the ifdef.
 309         Use Noncopyable. Change the return value of match.
 310         * kjs/regexp.cpp:
 311         (KJS::RegExp::RegExp): Call pcre_compile2, for a slight speed boost.
 312         (KJS::RegExp::~RegExp): PCRE16 rather than PCREPOSIX.
 313         (KJS::RegExp::match): Change to return the position as an int and the
 314         ovector as a OwnArrayPtr<int> for efficiency and clearer storage management.
 315
 316         * kjs/regexp_object.h: Change performMatch and arrayOfMatches to no longer
 317         require a result string.
 318         * kjs/regexp_object.cpp:
 319         (RegExpProtoFunc::callAsFunction): Update for new signature of performMatch.
 320         (RegExpObjectImp::performMatch): Change so it doesn't return a string.
 321         (RegExpObjectImp::arrayOfMatches): Simplify by unifying the handling of
 322         the main result with the backreferences; now it doesn't need to take
 323         a result parameter.
 324         (RegExpObjectImp::getBackref): Minor tweaks.
 325         (RegExpObjectImp::getLastParen): Ditto.
 326         (RegExpObjectImp::getLeftContext): Ditto.
 327         (RegExpObjectImp::getRightContext): Ditto.
 328         (RegExpObjectImp::getValueProperty): Change LastMatch case to call
 329         getBackref(0) so we don't need a separate getLastMatch function.
 330
 331         * kjs/string_object.cpp:
 332         (KJS::replace): Update to use new performMatch, including merging the
 333         matched string section with the other substrings.
 334         (KJS::StringProtoFunc::callAsFunction): Update functions to use the
 335         new performMatch and match. Also change to use OwnArrayPtr.
 336
 337 2007-10-31  Oliver Hunt  <oliver@apple.com>
 338
 339         * kjs/nodes.h: include OwnPtr.h
 340
 341 2007-10-31  Oliver Hunt  <oliver@apple.com>
 342
 343         Reviewed by Maciej.
 344
 345         Remove SourceCodeElement class and replaced with a Vector for a 0.8% gain on sunspider
 346
 347         * kjs/grammar.y:
 348         * kjs/nodes.cpp:
 349         (KJS::statementListPushFIFO):
 350         (KJS::statementListGetDeclarations):
 351         (KJS::statementListInitializeDeclarationStacks):
 352         (KJS::statementListInitializeVariableAccessStack):
 353         (KJS::statementListExecute):
 354         (KJS::BlockNode::optimizeVariableAccess):
 355         (KJS::BlockNode::BlockNode):
 356         (KJS::BlockNode::getDeclarations):
 357         (KJS::BlockNode::execute):
 358         (KJS::CaseClauseNode::optimizeVariableAccess):
 359         (KJS::CaseClauseNode::getDeclarations):
 360         (KJS::CaseClauseNode::evalStatements):
 361         (KJS::FunctionBodyNode::initializeDeclarationStacks):
 362         (KJS::FunctionBodyNode::optimizeVariableAccess):
 363         * kjs/nodes.h:
 364         * kjs/nodes2string.cpp:
 365         (KJS::statementListStreamTo):
 366         (KJS::BlockNode::streamTo):
 367         (KJS::CaseClauseNode::streamTo):
 368
 369 2007-10-30  Mark Rowe  <mrowe@apple.com>
 370
 371         * kjs/property_map.cpp: Added a missing using directive to fix the build
 372         for non-Mac ports. Mac worked only because it does the AllInOneFile compile.
 373
 374 2007-10-31  Maciej Stachowiak  <mjs@apple.com>
 375
 376         * kjs/property_map.cpp: Include HashTable.h the right way to fix the build
 377         for non-Mac ports.
 378
 379 2007-10-31  Alexey Proskuryakov  <ap@webkit.org>
 380
 381         Reviewed by Darin.
 382
 383         http://bugs.webkit.org/show_bug.cgi?id=11001
 384         WebKit doesn't support RegExp.compile method
 385
 386         Test: fast/js/regexp-compile.html
 387
 388         * kjs/regexp_object.cpp:
 389         (RegExpPrototype::RegExpPrototype):
 390         (RegExpProtoFunc::callAsFunction):
 391         * kjs/regexp_object.h:
 392         (KJS::RegExpProtoFunc::):
 393         Added RegExp.compile.
 394
 395         * tests/mozilla/expected.html: js1_2/regexp/compile.js now passes.
 396
 397 2007-10-31  Maciej Stachowiak  <mjs@apple.com>
 398
 399         Reviewed by Oliver.
 400
 401         - get rid of integer divide in PropertyMap and HashTable for 1% SunSpider speedup
 402
 403         Integer divide sucks. Fortunately, a bunch of shifts and XORs
 404         biased towards the high bits is sufficient to provide a good
 405         double hash. Besides the SunSpider win, I used the dump statistics
 406         mode for both to verify that collisions did not increase and that
 407         the longest collision chain is not any longer.
 408
 409         * kjs/property_map.cpp:
 410         (KJS::doubleHash):
 411         (KJS::PropertyMap::get):
 412         (KJS::PropertyMap::getLocation):
 413         (KJS::PropertyMap::put):
 414         (KJS::PropertyMap::insert):
 415         (KJS::PropertyMap::remove):
 416         (KJS::PropertyMap::checkConsistency):
 417         * wtf/HashTable.h:
 418         (WTF::doubleHash):
 419         (WTF::::lookup):
 420         (WTF::::lookupForWriting):
 421         (WTF::::fullLookupForWriting):
 422         (WTF::::add):
 423
 424 2007-10-30  Adam Roben  <aroben@apple.com>
 425
 426         * kjs/collector.h: Make HeapType public so it can be used for non-member
 427         things like the HeapConstants struct template. Fixes the build on Windows.
 428
 429 2007-10-30  Adam Roben  <aroben@apple.com>
 430
 431         Change ALWAYS_INLINE and WTF_PRIVATE_INLINE to use __forceinline on Windows
 432
 433         Speeds up SunSpider by 0.4%.
 434
 435         Reviewed by Steve and Maciej.
 436
 437         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj: Disable
 438         a warning during LTCG in release builds about double -> float
 439         conversion.
 440         * wtf/AlwaysInline.h:
 441         * wtf/FastMalloc.h:
 442
 443 2007-10-30  Adam Roben  <aroben@apple.com>
 444
 445         Use GetCurrentThreadId instead of pthread_self in FastMalloc
 446
 447         Speeds up SunSpider by 0.3%.
 448
 449         Reviewed by Steve.
 450
 451         * wtf/FastMalloc.cpp:
 452         (WTF::TCMalloc_ThreadCache::InitTSD):
 453         (WTF::TCMalloc_ThreadCache::CreateCacheIfNecessary):
 454
 455 2007-10-30  Adam Roben  <aroben@apple.com>
 456
 457         Switch to a Win32 critical section implementation of spinlocks
 458
 459         Speeds up SunSpider by 0.4%.
 460
 461         Reviewed by Steve.
 462
 463         * wtf/FastMalloc.cpp:
 464         * wtf/TCSpinLock.h:
 465         (TCMalloc_SpinLock::TCMalloc_SpinLock):
 466         (TCMalloc_SpinLock::Init):
 467         (TCMalloc_SpinLock::Finalize):
 468         (TCMalloc_SpinLock::Lock):
 469         (TCMalloc_SpinLock::Unlock):
 470         * wtf/TCSystemAlloc.cpp:
 471
 472 2007-10-30  Adam Roben  <aroben@apple.com>
 473
 474         Fix Bug 15586: REGRESSION (r26759-r26785): Windows nightly builds crash with Safari 3 Public Beta
 475
 476         http://bugs.webkit.org/show_bug.cgi?id=15586
 477
 478         Also fixes: <rdar://5565303> Cannot use regsvr32.exe to register WebKit.dll
 479
 480         Use Win32 TLS functions instead of __declspec(thread), which breaks
 481         delay-loading.
 482
 483         Reviewed by Steve.
 484
 485         * wtf/FastMalloc.cpp:
 486         (WTF::getThreadHeap):
 487         (WTF::TCMalloc_ThreadCache::InitModule):
 488
 489 2007-10-30  Maciej Stachowiak  <mjs@apple.com>
 490
 491         Reviewed by Oliver.
 492
 493         - allocate numbers in half-size cells, for an 0.5% SunSpider speedup
 494         http://bugs.webkit.org/show_bug.cgi?id=15772
 495
 496         We do this by using a single mark bit per two number cells, and
 497         tweaking marking.
 498
 499         Besides being an 0.5% win overall, this is a 7.1% win on morph.
 500
 501         * kjs/collector.cpp:
 502         (KJS::):
 503         (KJS::Collector::heapAllocate):
 504         (KJS::Collector::markStackObjectsConservatively):
 505         (KJS::Collector::sweep):
 506         * kjs/collector.h:
 507         (KJS::SmallCollectorCell::):
 508
 509 2007-10-30  Geoffrey Garen  <ggaren@apple.com>
 510
 511         Reviewed by Adam Roben, Sam Weinig.
 512
 513         Made conflicts in grammar.y a persistent build failure.
 514
 515         * DerivedSources.make:
 516
 517 2007-10-30  Kevin McCullough  <kmccullough@apple.com>
 518
 519         Reviewed by Adam and Geoff.
 520
 521         - Added a new cast so all the casts are in the same place.
 522
 523         * API/APICast.h:
 524         (toGlobalRef):
 525
 526 2007-10-30  Geoffrey Garen  <ggaren@apple.com>
 527
 528         Reviewed by Darin Adler.
 529
 530         Fixed <rdar://problem/5567504> shift/reduce conflict introduced in r24457
 531
 532         JS tests, including
 533
 534             ecma_2/Statements/dowhile-001.js
 535             ecma_2/Statements/dowhile-002.js
 536             ecma_2/Statements/dowhile-003.js
 537             ecma_2/Statements/dowhile-004.js
 538             ecma_2/Statements/dowhile-005.js
 539             ecma_2/Statements/dowhile-006.js
 540             ecma_2/Statements/dowhile-007.js
 541             js1_2/statements/do_while.js
 542
 543         and layout tests, including
 544
 545             do-while-expression-value.html
 546             do-while-semicolon.html
 547             do-while-without-semicolon.html
 548
 549         pass.
 550
 551         * kjs/grammar.y: Use the explicit "error" production, as we do with other
 552         automatic semicolon insertions, to disambiguate "do { } while();" from
 553         "do { } while()" followed by ";" (the empty statement).
 554
 555 2007-10-29  Oliver Hunt  <oliver@apple.com>
 556
 557         Reviewed by Maciej.
 558
 559         Debranching remaining assignment nodes, and miscellaneous cleanup
 560
 561         Split read-modify code paths out of AssignBracketNode and AssignDotNode
 562         Removed now unnecessary check for write-only assignment in ReadModifyLocalVarNode
 563         and ReadModifyResolveNode evaluate methods
 564
 565         Leads to a 1% gain in SunSpider.
 566
 567         * kjs/grammar.y:
 568         * kjs/nodes.cpp:
 569         (KJS::ReadModifyLocalVarNode::evaluate):
 570         (KJS::ReadModifyResolveNode::evaluate):
 571         (KJS::AssignDotNode::evaluate):
 572         (KJS::ReadModifyDotNode::optimizeVariableAccess):
 573         (KJS::ReadModifyDotNode::evaluate):
 574         (KJS::AssignBracketNode::evaluate):
 575         (KJS::ReadModifyBracketNode::optimizeVariableAccess):
 576         (KJS::ReadModifyBracketNode::evaluate):
 577         * kjs/nodes.h:
 578         (KJS::AssignBracketNode::):
 579         (KJS::AssignBracketNode::precedence):
 580         (KJS::AssignDotNode::):
 581         (KJS::AssignDotNode::precedence):
 582         * kjs/nodes2string.cpp:
 583         (KJS::ReadModifyBracketNode::streamTo):
 584         (KJS::AssignBracketNode::streamTo):
 585         (KJS::ReadModifyDotNode::streamTo):
 586         (KJS::AssignDotNode::streamTo):
 587
 588 2007-10-29  Oliver Hunt  <oliver@apple.com>
 589
 590         Debranching various Node::evaluate implementations
 591
 592         Reviewed by Maciej.
 593
 594         Split the read-modify-write assignment cases out of AssignResolveNode and into ReadModifyResolveNode
 595         Split the increment and decrement cases for Prefix- and Postfix- ResolveNode, BracketNode, and DotNode
 596
 597         Gains 1.6% on SunSpider
 598
 599         * JavaScriptCore.xcodeproj/project.pbxproj:
 600         * kjs/grammar.y:
 601         * kjs/nodes.cpp:
 602         (KJS::PostIncResolveNode::optimizeVariableAccess):
 603         (KJS::PostIncResolveNode::evaluate):
 604         (KJS::PostIncLocalVarNode::evaluate):
 605         (KJS::PostDecResolveNode::optimizeVariableAccess):
 606         (KJS::PostDecResolveNode::evaluate):
 607         (KJS::PostDecLocalVarNode::evaluate):
 608         (KJS::PostIncBracketNode::evaluate):
 609         (KJS::PostDecBracketNode::evaluate):
 610         (KJS::PostIncDotNode::evaluate):
 611         (KJS::PostDecDotNode::evaluate):
 612         (KJS::PreIncResolveNode::optimizeVariableAccess):
 613         (KJS::PreIncLocalVarNode::evaluate):
 614         (KJS::PreIncResolveNode::evaluate):
 615         (KJS::PreDecResolveNode::optimizeVariableAccess):
 616         (KJS::PreDecLocalVarNode::evaluate):
 617         (KJS::PreDecResolveNode::evaluate):
 618         (KJS::PreIncBracketNode::evaluate):
 619         (KJS::PreDecBracketNode::evaluate):
 620         (KJS::PreIncDotNode::evaluate):
 621         (KJS::PreDecDotNode::evaluate):
 622         (KJS::ReadModifyResolveNode::optimizeVariableAccess):
 623         (KJS::AssignResolveNode::optimizeVariableAccess):
 624         (KJS::AssignLocalVarNode::evaluate):
 625         (KJS::AssignResolveNode::evaluate):
 626         * kjs/nodes.h:
 627         (KJS::PostDecResolveNode::):
 628         (KJS::PostDecResolveNode::precedence):
 629         (KJS::PostDecLocalVarNode::):
 630         (KJS::PostfixBracketNode::):
 631         (KJS::PostfixBracketNode::precedence):
 632         (KJS::PostIncBracketNode::):
 633         (KJS::PostIncBracketNode::isIncrement):
 634         (KJS::PostDecBracketNode::):
 635         (KJS::PostDecBracketNode::isIncrement):
 636         (KJS::PostfixDotNode::):
 637         (KJS::PostfixDotNode::precedence):
 638         (KJS::PostIncDotNode::):
 639         (KJS::PostIncDotNode::isIncrement):
 640         (KJS::PostDecDotNode::):
 641         (KJS::PreIncResolveNode::):
 642         (KJS::PreDecResolveNode::):
 643         (KJS::PreDecResolveNode::precedence):
 644         (KJS::PreDecLocalVarNode::):
 645         (KJS::PrefixBracketNode::):
 646         (KJS::PrefixBracketNode::precedence):
 647         (KJS::PreIncBracketNode::):
 648         (KJS::PreIncBracketNode::isIncrement):
 649         (KJS::PreDecBracketNode::):
 650         (KJS::PreDecBracketNode::isIncrement):
 651         (KJS::PrefixDotNode::):
 652         (KJS::PrefixDotNode::precedence):
 653         (KJS::PreIncDotNode::):
 654         (KJS::PreIncDotNode::isIncrement):
 655         (KJS::PreDecDotNode::):
 656         (KJS::ReadModifyResolveNode::):
 657         (KJS::ReadModifyLocalVarNode::):
 658         (KJS::AssignResolveNode::):
 659         (KJS::AssignResolveNode::precedence):
 660         * kjs/nodes2string.cpp:
 661         (KJS::PostIncResolveNode::streamTo):
 662         (KJS::PostDecResolveNode::streamTo):
 663         (KJS::PostfixBracketNode::streamTo):
 664         (KJS::PostfixDotNode::streamTo):
 665         (KJS::PreIncResolveNode::streamTo):
 666         (KJS::PreDecResolveNode::streamTo):
 667         (KJS::ReadModifyResolveNode::streamTo):
 668         (KJS::AssignResolveNode::streamTo):
 669
 670 2007-10-29  Maciej Stachowiak  <mjs@apple.com>
 671
 672         Not reviewed, build fix.
 673
 674         - Include Vector.h in a way that actually works.
 675
 676         * kjs/LocalStorage.h:
 677
 678 2007-10-29  Maciej Stachowiak  <mjs@apple.com>
 679
 680         Not reviewed, build fix.
 681
 682         - Install LocalStorage.h as a private header.
 683
 684         * JavaScriptCore.xcodeproj/project.pbxproj:
 685
 686 2007-10-29  Maciej Stachowiak  <mjs@apple.com>
 687
 688         Reviewed by Darin.
 689
 690         - Define good VectorTraits for LocalStorage entry for 0.5% speed improvement on SunSpider.
 691
 692         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
 693         * JavaScriptCore.xcodeproj/project.pbxproj:
 694         * kjs/LocalStorage.h: Added.
 695         (KJS::LocalStorageEntry::LocalStorageEntry):
 696         (WTF::):
 697         * kjs/function.h:
 698         * kjs/nodes.cpp:
 699         (KJS::FunctionBodyNode::processDeclarationsForFunctionCode):
 700
 701 2007-10-29  Geoffrey Garen  <ggaren@apple.com>
 702
 703         Reviewed by Oliver Hunt.
 704
 705         Some small tweaks that I notice while reviewing Oliver's last patch.
 706
 707         Includes removal of an unnecessary KJS_CHECKEXCEPTIONVALUE.
 708
 709         No change in SunSpider because SunSpider doesn't take the code path that
 710         would execute the unnecessary KJS_CHECKEXCEPTIONVALUE much.
 711
 712         * kjs/nodes.cpp:
 713         (KJS::LocalVarPostfixNode::evaluate):
 714         (KJS::TypeOfResolveNode::optimizeVariableAccess):
 715         (KJS::LocalVarTypeOfNode::evaluate):
 716         (KJS::PrefixResolveNode::optimizeVariableAccess):
 717         (KJS::LocalVarPrefixNode::evaluate):
 718         (KJS::AssignResolveNode::optimizeVariableAccess):
 719         (KJS::LocalVarAssignNode::evaluate):
 720         * kjs/nodes.h:
 721         (KJS::LocalVarTypeOfNode::):
 722         (KJS::PrefixResolveNode::):
 723         (KJS::LocalVarPrefixNode::):
 724         (KJS::AssignResolveNode::):
 725         (KJS::LocalVarAssignNode::):
 726
 727 2007-10-29  Eric Seidel  <eric@webkit.org>
 728
 729         Reviewed by Maciej.
 730
 731         SunSpider claims this was a 0.7% speedup.
 732
 733         * kjs/string_object.cpp:
 734         (KJS::StringProtoFunc::callAsFunction): avoid mallocing a jsString in the common case
 735
 736 2007-10-29  Maciej Stachowiak  <mjs@apple.com>
 737
 738         Reviewed by Mark.
 739
 740         - re-enable asserts for access to empty or deleted keys
 741
 742         * wtf/HashTable.h:
 743         (WTF::::lookup):
 744         (WTF::::lookupForWriting):
 745         (WTF::::fullLookupForWriting):
 746         (WTF::::add):
 747
 748 2007-10-29  Eric Seidel  <eric@webkit.org>
 749
 750         Build fix only, no review.
 751
 752         * JavaScriptCore.exp: Export symbol for new StringInstance::getOwnPropertySlot
 753
 754 2007-10-29  Mark Rowe  <mrowe@apple.com>
 755
 756         Gtk build fix.  Move struct declarations into nodes.h.
 757
 758         * kjs/grammar.y:
 759         * kjs/nodes.h:
 760
 761 2007-10-29  Eric Seidel  <eric@webkit.org>
 762
 763         Reviewed by darin.
 764
 765         Give StringInstance a getOwnPropertySlot(ExecState, unsigned, PropertySlot) fastpath, just like Arrays.
 766         Make it a compile time error to use toString(ExecState) on a StringInstance
 767
 768         SunSpider claims this was a 6.6% speedup overall (22% on string-base64)
 769
 770         * kjs/internal.h:
 771         (KJS::StringImp::getLength):
 772         * kjs/string_object.cpp:
 773         (KJS::StringInstance::lengthGetter):
 774         (KJS::StringInstance::inlineGetOwnPropertySlot):
 775         (KJS::StringInstance::getOwnPropertySlot):
 776         * kjs/string_object.h:
 777
 778 2007-10-28  Oliver Hunt  <oliver@apple.com>
 779
 780         Reviewed by Darin.
 781
 782         Add nodes to allow Assignment, TypeOf, and prefix operators to
 783         make use of the new optimised local variable look up.
 784
 785         5% gain on sunspider
 786
 787         * kjs/nodes.cpp:
 788         (KJS::TypeOfResolveNode::optimizeVariableAccess):
 789         (KJS::LocalTypeOfAccessNode::evaluate):
 790         (KJS::PrefixResolveNode::optimizeVariableAccess):
 791         (KJS::PrefixLocalAccessNode::evaluate):
 792         (KJS::AssignResolveNode::optimizeVariableAccess):
 793         (KJS::AssignLocalAccessNode::evaluate):
 794         * kjs/nodes.h:
 795         (KJS::TypeOfResolveNode::):
 796         (KJS::TypeOfResolveNode::precedence):
 797         (KJS::LocalTypeOfAccessNode::):
 798         (KJS::PrefixResolveNode::):
 799         (KJS::PrefixResolveNode::precedence):
 800         (KJS::PrefixLocalAccessNode::):
 801         (KJS::AssignResolveNode::):
 802         (KJS::AssignLocalAccessNode::):
 803
 804 2007-10-28  Maciej Stachowiak  <mjs@apple.com>
 805
 806         Reviewed by Darin.
 807
 808         - avoid creating and then breaking circular lists in the parser, instead track head and tail pointers at parse time
 809         http://bugs.webkit.org/show_bug.cgi?id=15748
 810
 811         Not a significant speedup or slowdown on SunSpider.
 812
 813         * kjs/Parser.cpp:
 814         (KJS::clearNewNodes):
 815         * kjs/Parser.h:
 816         * kjs/grammar.y:
 817         * kjs/nodes.cpp:
 818         (KJS::BlockNode::BlockNode):
 819         (KJS::CaseBlockNode::CaseBlockNode):
 820         (KJS::FunctionBodyNode::FunctionBodyNode):
 821         (KJS::SourceElementsNode::SourceElementsNode):
 822         (KJS::ProgramNode::ProgramNode):
 823         * kjs/nodes.h:
 824         (KJS::ElementNode::):
 825         (KJS::ArrayNode::):
 826         (KJS::PropertyListNode::):
 827         (KJS::ObjectLiteralNode::):
 828         (KJS::ArgumentListNode::):
 829         (KJS::ArgumentsNode::):
 830         (KJS::VarDeclListNode::):
 831         (KJS::VarStatementNode::):
 832         (KJS::ForNode::):
 833         (KJS::ParameterNode::):
 834         (KJS::FuncExprNode::):
 835         (KJS::FuncDeclNode::):
 836         (KJS::SourceElementsNode::):
 837         (KJS::CaseClauseNode::):
 838         (KJS::ClauseListNode::):
 839
 840 2007-10-28  Mark Rowe  <mrowe@apple.com>
 841
 842         Disable assertions in a manner that doesn't break the Qt Windows build.
 843
 844         * wtf/HashTable.h:
 845         (WTF::::lookup):
 846         (WTF::::lookupForWriting):
 847         (WTF::::fullLookupForWriting):
 848
 849 2007-10-28  Geoffrey Garen  <ggaren@apple.com>
 850
 851         Temporarily disabling some ASSERTs I introduced in my last check-in
 852         because of http://bugs.webkit.org/show_bug.cgi?id=15747
 853         Lots of layout tests fail the !HashTranslator::equal(KeyTraits::emptyValue() ASSERT
 854
 855         * wtf/HashTable.h:
 856         (WTF::::lookup):
 857         (WTF::::lookupForWriting):
 858         (WTF::::fullLookupForWriting):
 859         (WTF::::add):
 860
 861 2007-10-28  Geoffrey Garen  <ggaren@apple.com>
 862
 863         Reviewed by Darin Adler.
 864
 865         Fixed http://bugs.webkit.org/show_bug.cgi?id=15746
 866         #ifndef ASSERT_DISABLED is no good!
 867
 868         Replaced with #if !ASSERT_DISABLED.
 869
 870         * wtf/HashTable.h:
 871         (WTF::::lookup):
 872         (WTF::::lookupForWriting):
 873         (WTF::::fullLookupForWriting):
 874         (WTF::::add):
 875
 876 2007-10-28  Geoffrey Garen  <ggaren@apple.com>
 877
 878         Reviewed by Darin Adler.
 879
 880         Added FunctionCallResolveNode, PostfixResolveNode, and DeleteResolveNode
 881         to the AST transfom that replaces slow resolve nodes with fast local
 882         variable alternatives.
 883
 884         2.5% speedup on SunSpider.
 885
 886         Also added some missing copyright notices.
 887
 888         * kjs/nodes.cpp:
 889         (KJS::FunctionCallResolveNode::optimizeVariableAccess):
 890         (KJS::FunctionCallResolveNode::evaluate):
 891         (KJS::LocalVarFunctionCallNode::evaluate):
 892         (KJS::PostfixResolveNode::optimizeVariableAccess):
 893         (KJS::PostfixResolveNode::evaluate):
 894         (KJS::LocalVarPostfixNode::evaluate):
 895         (KJS::DeleteResolveNode::optimizeVariableAccess):
 896         (KJS::DeleteResolveNode::evaluate):
 897         (KJS::LocalVarDeleteNode::evaluate):
 898         * kjs/nodes.h:
 899         (KJS::FunctionCallResolveNode::):
 900         (KJS::LocalVarFunctionCallNode::LocalVarFunctionCallNode):
 901         (KJS::PostfixResolveNode::):
 902         (KJS::LocalVarPostfixNode::LocalVarPostfixNode):
 903         (KJS::DeleteResolveNode::):
 904         (KJS::LocalVarDeleteNode::LocalVarDeleteNode):
 905
 906 2007-10-28  Eric Seidel  <eric@webkit.org>
 907
 908         Reviewed by darin.
 909
 910         Inline UString::Rep::deref() for a 0.8% improvement in SunSpider
 911         Add virtual keyword to a few virtual functions previously unmarked.
 912
 913         * kjs/internal.h:
 914         (KJS::StringImp::type):
 915         (KJS::NumberImp::type):
 916         * kjs/ustring.h:
 917         (KJS::UString::Rep::deref):
 918
 919 2007-10-28  Darin Adler  <darin@apple.com>
 920
 921         - fix "broken everything" from the storage leak fix
 922
 923         * wtf/RefPtr.h: (WTF::RefPtr::RefPtr): Added a PlacementNewAdopt constructor.
 924         * kjs/ustring.h: (KJS::UString::UString): Pass PlacementNewAdopt along to RefPtr.
 925
 926 2007-10-28  Darin Adler  <darin@apple.com>
 927
 928         Reviewed by Adam.
 929
 930         - turn on unused parameter waring on Mac OS X because it's already on elsewhere
 931
 932         * Configurations/Base.xcconfig: Took out -wno-unused-parameter.
 933
 934         * API/JSNode.c:
 935         * API/JSNodeList.c:
 936         * API/minidom.c:
 937         * API/testapi.c:
 938         Fixed unused variables by using them or marked them with UNUSED_PARAM.
 939
 940         * kjs/CollectorHeapIntrospector.h: (KJS::CollectorHeapIntrospector::zoneCalloc):
 941         Removed parameter names to indicate they are unused.
 942
 943 2007-10-28  Darin Adler  <darin@apple.com>
 944
 945         Reviewed by Maciej.
 946
 947         - fix a storage leak where we ref the UString every time we replace
 948           a ResolveNode with a LocalVarAccessNode
 949
 950         * kjs/identifier.h: (KJS::Identifier::Identifier): Added a constructor
 951         that takes PlacementNewAdopt.
 952
 953         * kjs/nodes.h: (KJS::ResolveNode::ResolveNode): Initialize the ident
 954         with PlacementNewAdopt instead of the old value of ident.
 955
 956         * kjs/ustring.h: (KJS::UString::UString): Added a constructor that
 957         takes PlacementNewAdopt.
 958
 959 2007-10-28  Darin Adler  <darin@apple.com>
 960
 961         - Windows build fix; get rid of unused parameter
 962
 963         * kjs/nodes.cpp: (KJS::ResolveNode::optimizeVariableAccess): Don't pass it.
 964         * kjs/nodes.h: (KJS::LocalVarAccessNode::LocalVarAccessNode): Remove it.
 965         The assertions weren't all that helpful.
 966
 967 2007-10-28  Mark Rowe  <mrowe@apple.com>
 968
 969         Gtk build fix.  Add include of MathExtras.h.
 970
 971         * kjs/string_object.cpp:
 972
 973 2007-10-28  Mark Rowe  <mrowe@apple.com>
 974
 975         Reviewed by Maciej and Tim.
 976
 977         Replace uses of isNaN and isInf with isnan and isinf, and
 978         remove isNaN and isInf.
 979
 980         * kjs/config.h: Remove unused HAVE_'s.
 981         * kjs/date_object.cpp:
 982         (KJS::DateInstance::getTime):
 983         (KJS::DateInstance::getUTCTime):
 984         (KJS::DateProtoFunc::callAsFunction):
 985         (KJS::DateObjectImp::construct):
 986         (KJS::DateObjectFuncImp::callAsFunction):
 987         * kjs/function.cpp:
 988         (KJS::GlobalFuncImp::callAsFunction):
 989         * kjs/math_object.cpp:
 990         (MathFuncImp::callAsFunction):
 991         * kjs/nodes2string.cpp:
 992         (KJS::isParserRoundTripNumber):
 993         * kjs/number_object.cpp:
 994         (NumberProtoFunc::callAsFunction):
 995         * kjs/operations.cpp:
 996         * kjs/operations.h:
 997         * kjs/string_object.cpp:
 998         (KJS::StringProtoFunc::callAsFunction):
 999         * kjs/ustring.cpp:
1000         (KJS::UString::from):
1001         * kjs/value.cpp:
1002         (KJS::JSValue::toInteger):
1003         (KJS::JSValue::toInt32SlowCase):
1004         (KJS::JSValue::toUInt32SlowCase):
1005
1006 2007-10-28  Geoffrey Garen  <ggaren@apple.com>
1007
1008         Build fix: use the new-fangled missingSymbolMarker().
1009
1010         * kjs/nodes.cpp:
1011         (KJS::ResolveNode::optimizeVariableAccess):
1012         * kjs/nodes.h:
1013         (KJS::LocalVarAccessNode::LocalVarAccessNode):
1014
1015 2007-10-28  Geoffrey Garen  <ggaren@apple.com>
1016
1017         Reviewed by Maciej Stachowiak, Darin Adler.
1018
1019         Much supporting work done by Maciej Stachowiak, Maks Orlovich, and
1020         Cameron Zwarich.
1021
1022         AST transfom to replace slow resolve nodes with fast local variable
1023         alternatives that do direct memory access. Currently, only ResolveNode
1024         provides a fast local variable alternative. 6 others are soon to come.
1025
1026         16.7% speedup on SunSpider.
1027
1028         Most of this patch is just scaffolding to support iterating all the
1029         resolve nodes in the AST through optimizeResolveNodes(). In
1030         optimizeResolveNodes(), most classes just push their child nodes onto
1031         the processing stack, while ResolveNodes actually replace themselves in
1032         the tree with more optimized alternatives, if possible.
1033
1034         Here are the interesting bits:
1035
1036         * kjs/nodes.h: Added PlacementNewAdoptTag, along with implementations
1037         in Node and ResolveNode. This tag allows you to use placement new to
1038         swap out a base class Node in favor of a subclass copy that holds the
1039         same data. (Without this tag, default initialization would NULL out
1040         RefPtrs, change line numbers, etc.)
1041
1042         * kjs/nodes.cpp:
1043         (KJS::ResolveNode::evaluate): Since we're taking the slow path, ASSERT
1044         that the fast path is impossible, to make sure we didn't leave anything
1045         on the table.
1046
1047         (KJS::FunctionBodyNode::optimizeResolveNodes): Here's where the AST
1048         transformation happens.
1049
1050         (KJS::ResolveNode::optimizeResolveNodes): Here's where the ResolveNode
1051         optimization happens.
1052
1053         * kjs/function.h: Added symbolTable() accessor for, for the sake of
1054         an ASSERT.
1055
1056 2007-10-28  Mark Rowe  <mrowe@apple.com>
1057
1058         Reviewed by Maciej.
1059
1060         Fix "AllInOneFile.o has a global initializer in it".
1061
1062         Some versions of gcc generate a global initializer for std::numeric_limits<size_t>::max().
1063         We can avoid this by moving it inside an inline function.
1064
1065         * kjs/SymbolTable.h:
1066         (KJS::missingSymbolMarker):
1067         * kjs/function.cpp:
1068         (KJS::ActivationImp::getOwnPropertySlot):
1069         (KJS::ActivationImp::put):
1070
1071 2007-10-28  Maciej Stachowiak  <mjs@apple.com>
1072
1073         Reviewed by Mark.
1074
1075         - Added assertions to protect against adding empty or deleted keys to a HashTable
1076
1077         * wtf/HashTable.h:
1078         (WTF::HashTable::lookup):
1079         (WTF::HashTable::lookupForWriting):
1080         (WTF::HashTable::fullLookupForWriting):
1081         (WTF::HashTable::add):
1082
1083 2007-10-28  Darin Adler  <darin@apple.com>
1084
1085         - fix GTK build
1086
1087         * kjs/nodes2string.cpp: (KJS::isParserRoundTripNumber):
1088         Use isNaN and isInf instead of isnan and isinf.
1089
1090 2007-10-28  Darin Adler  <darin@apple.com>
1091
1092         Reviewed by Maciej.
1093
1094         - http://bugs.webkit.org/show_bug.cgi?id=15735
1095           remove GroupNode to simplify AST and possibly get a modest speedup
1096
1097         This patch removes 4 node types: GroupNode, PropertyNameNode,
1098         FunctionCallParenBracketNode, and FunctionCallParenDotNode.
1099
1100         To remove GroupNode, we add knowledge of precedence to the tree nodes,
1101         and use that when serializing to determine where parentheses are needed.
1102         This means we no longer have to represent parentheses in the tree.
1103
1104         The precedence values are named after productions in the grammar from the
1105         JavaScript standard.
1106
1107         SunSpider says this is an 0.4% speedup.
1108
1109         * kjs/function.h:
1110         * kjs/function.cpp: Removed escapeStringForPrettyPrinting -- it's part of
1111         serialization, so I moved it to the file that takes care of that.
1112
1113         * kjs/grammar.y: Changed makeGetterOrSetterPropertyNode to use 0 to
1114         indicate failure instead of a separate boolean. Got rid of PropertyNameNode
1115         by merging the PropertyName rule into the Property rule (which was easier
1116         than figuring out how to pass the Identifier from one node to another).
1117         Got rid of GroupNode, nodeInsideAllParens(), FunctionCallParenBracketNode,
1118         and FunctionCallParenDotNode.
1119
1120         * kjs/nodes.h: Removed unused forward declarations and Operator values.
1121         Added Precedence enum, and precedence function to all nodes. Removed
1122         nodeInsideAllParens. Added streamBinaryOperator function for serialization.
1123         Removed GroupNode and PropertyNameNode. Made PropertyNode store an Identifier.
1124         Removed FunctionCallParenBracketNode and FunctionCallParenDotNode.
1125
1126         * kjs/nodes.cpp: Removed Node::nodinsideAllParens, GroupNode, and PropertyNameNode.
1127         (KJS::PropertyListNode::evaluate): Changed code to get name directly instead
1128         of converting it from an Identifier to a jsString then back to a UString
1129         then into an Identifier again!
1130
1131         * kjs/nodes2string.cpp: Changed special-token implementation to use a separate
1132         function for each of Endl, Indent, Unindent, and DotExpr instead of using a
1133         single function with a switch. Added a precedence that you can stream in, to
1134         cause the next node serialized to add parentheses based on that precedence value.
1135         (KJS::operatorString): Moved to the top of the file.
1136         (KJS::escapeStringForPrettyPrinting): Moved here from function.cpp. Removed old
1137         workaround for snprintf, since StringExtras.h takes care of that.
1138         (KJS::operator<<): Made the char and char* versions faster by using UString's
1139         character append functions instead of constructing a UString. Added the logic
1140         to the Node* version to add parentheses if needed.
1141         (KJS::Node::streamLeftAssociativeBinaryOperator): Added helper function.
1142         (KJS::ElementNode::streamTo): Use PrecAssignment for the elements.
1143         (KJS::BracketAccessorNode::streamTo): Use PrecCall for the expression before
1144         the bracket.
1145         (KJS::DotAccessorNode::streamTo): Use PrecCall for the expression before the dot.
1146         (KJS::ArgumentListNode::streamTo): Use PrecAssignment for the arguments.
1147         (KJS::NewExprNode::streamTo): Use PrecMember for the expression.
1148         (KJS::FunctionCallValueNode::streamTo): Use PrecCall.
1149         (KJS::FunctionCallBracketNode::streamTo): Ditto.
1150         (KJS::FunctionCallDotNode::streamTo): Ditto.
1151         (KJS::PostfixBracketNode::streamTo): Ditto.
1152         (KJS::PostfixDotNode::streamTo): Ditto.
1153         (KJS::PostfixErrorNode::streamTo): Use PrecLeftHandSide.
1154         (KJS::DeleteBracketNode::streamTo): Use PrecCall.
1155         (KJS::DeleteDotNode::streamTo): Ditto.
1156         (KJS::DeleteValueNode::streamTo): Use PrecUnary.
1157         (KJS::VoidNode::streamTo): Ditto.
1158         (KJS::TypeOfValueNode::streamTo): Ditto.
1159         (KJS::PrefixBracketNode::streamTo): Use PrecCall.
1160         (KJS::PrefixDotNode::streamTo): Ditto.
1161         (KJS::PrefixErrorNode::streamTo): Use PrecUnary.
1162         (KJS::UnaryPlusNode::streamTo): Ditto.
1163         (KJS::NegateNode::streamTo): Ditto.
1164         (KJS::BitwiseNotNode::streamTo): Ditto.
1165         (KJS::LogicalNotNode::streamTo): Ditto.
1166         (KJS::MultNode::streamTo): Use streamLeftAssociativeBinaryOperator.
1167         (KJS::DivNode::streamTo): Ditto.
1168         (KJS::ModNode::streamTo): Ditto.
1169         (KJS::AddNode::streamTo): Ditto.
1170         (KJS::SubNode::streamTo): Ditto.
1171         (KJS::LeftShiftNode::streamTo): Ditto.
1172         (KJS::RightShiftNode::streamTo): Ditto.
1173         (KJS::UnsignedRightShiftNode::streamTo): Ditto.
1174         (KJS::LessNode::streamTo): Ditto.
1175         (KJS::GreaterNode::streamTo): Ditto.
1176         (KJS::LessEqNode::streamTo): Ditto.
1177         (KJS::GreaterEqNode::streamTo): Ditto.
1178         (KJS::InstanceOfNode::streamTo): Ditto.
1179         (KJS::InNode::streamTo): Ditto.
1180         (KJS::EqualNode::streamTo): Ditto.
1181         (KJS::NotEqualNode::streamTo): Ditto.
1182         (KJS::StrictEqualNode::streamTo): Ditto.
1183         (KJS::NotStrictEqualNode::streamTo): Ditto.
1184         (KJS::BitAndNode::streamTo): Ditto.
1185         (KJS::BitXOrNode::streamTo): Ditto.
1186         (KJS::BitOrNode::streamTo): Ditto.
1187         (KJS::LogicalAndNode::streamTo): Ditto.
1188         (KJS::LogicalOrNode::streamTo): Ditto.
1189         (KJS::ConditionalNode::streamTo): Ditto.
1190         (KJS::AssignResolveNode::streamTo): Use PrecAssignment for the right side.
1191         (KJS::AssignBracketNode::streamTo): Use PrecCall for the expression before
1192         the bracket and PrecAssignment for the right side.
1193         (KJS::AssignDotNode::streamTo): Ditto.
1194         (KJS::AssignErrorNode::streamTo): Use PrecLeftHandSide for the left side
1195         and PrecAssignment for the right side.
1196         (KJS::CommaNode::streamTo): Use PrecAssignment for both expressions.
1197         (KJS::AssignExprNode::streamTo): Use PrecAssignment.
1198
1199 2007-10-28  Kevin Ollivier  <kevino@theolliviers.com>
1200
1201         Define wx port and set wx port USE options.
1202
1203         Reviewed by Adam Roben.
1204
1205         * wtf/Platform.h:
1206
1207 2007-10-28  Mark Rowe  <mrowe@apple.com>
1208
1209         We don't include "config.h" in headers.
1210
1211         * bindings/jni/jni_instance.h:
1212         * kjs/regexp.h:
1213         * wtf/TCPageMap.h:
1214         * wtf/TCSpinLock.h:
1215
1216 2007-10-28  Maciej Stachowiak  <mjs@apple.com>
1217
1218         Rubber stamped by Mark.
1219
1220         - avoid using non-portable SIZE_T_MAX in favor of std::numeric_limits
1221
1222         * kjs/SymbolTable.h:
1223         (KJS::SymbolTableIndexHashTraits::emptyValue):
1224         * kjs/function.cpp:
1225         (KJS::ActivationImp::getOwnPropertySlot):
1226         (KJS::ActivationImp::put):
1227
1228 2007-10-28  Maciej Stachowiak  <mjs@apple.com>
1229
1230         Reviewed by Eric.
1231
1232         - switch SymbolTable to be a HashMap instead of a PropertyMap for 3% SunSpider speedup
1233
1234         * kjs/SymbolTable.h:
1235         (KJS::IdentifierRepHash::hash): Special hash function for identifier reps.
1236         (KJS::IdentifierRepHash::equal): ditto
1237         (KJS::SymbolTableIndexHashTraits::emptyValue): Special HashTraits for the index value.
1238         (KJS::SymbolTable): change to a typedef for a HashMap.
1239         * kjs/function.cpp:
1240         (KJS::ActivationImp::getOwnPropertySlot): Adjusted for new SymbolTable API.
1241         (KJS::ActivationImp::deleteProperty): ditto
1242         (KJS::ActivationImp::put): ditto
1243
1244         * kjs/nodes.cpp:
1245         (KJS::FunctionBodyNode::initializesymbolTable): Adjusted, since
1246         you now have to store a UString::rep, not an identifier.
1247
1248 2007-10-27  Maciej Stachowiak  <mjs@apple.com>
1249
1250         Reviewed by Oliver.
1251
1252         - numerous HashTable performance improvements
1253
1254         This does not quite add up to a measurable win on SunSpider, but it allows a
1255         follow-on > 3% improvement and probably helps WebCore too.
1256
1257         I made the following improvements, among others:
1258
1259         - Made HashFunctions note whether it is ok to compare a real value with the equal() function
1260         to the empty or deleted value, and used this to optimize the comparisons done in hash lookup.
1261
1262         - Specialized lookup so it doesn't have to do so many extra branches and build so many extra
1263         std::pairs for cases that don't need them. There are now four versions, one for read-only access,
1264         two for writing, and one folded directly into add() (these all were improvments).
1265
1266         - Made HashMap::get() use lookup() directly instead of find() to avoid having to build iterators.
1267
1268         - Made a special constructor for iterators that knows it points to
1269         a valid filled cell and so skips updating itself.
1270
1271         - Reordered memory accesses in the various lookup functions for better code generation
1272
1273         - Made simple translators avoid passing a hash code around
1274
1275         - Other minor tweaks
1276
1277         * wtf/HashTable.h:
1278         (WTF::):
1279         (WTF::HashTableConstIterator::HashTableConstIterator):
1280         (WTF::HashTableIterator::HashTableIterator):
1281         (WTF::IdentityHashTranslator::translate):
1282         (WTF::HashTable::end):
1283         (WTF::HashTable::lookup):
1284         (WTF::HashTable::lookupForWriting):
1285         (WTF::HashTable::makeKnownGoodIterator):
1286         (WTF::HashTable::makeKnownGoodConstIterator):
1287         (WTF::::lookup):
1288         (WTF::::lookupForWriting):
1289         (WTF::::fullLookupForWriting):
1290         (WTF::::add):
1291         (WTF::::addPassingHashCode):
1292         (WTF::::reinsert):
1293         (WTF::::find):
1294         (WTF::::contains):
1295         * kjs/identifier.cpp:
1296         (WTF::):
1297         * wtf/HashFunctions.h:
1298         (WTF::):
1299         * wtf/HashMap.h:
1300         (WTF::):
1301         (WTF::::get):
1302         * wtf/HashSet.h:
1303         (WTF::):
1304         (WTF::::add):
1305         * wtf/ListHashSet.h:
1306         (WTF::ListHashSetTranslator::translate):
1307
1308 2007-10-27  Darin Adler  <darin@apple.com>
1309
1310         Reviewed by Eric.
1311
1312         - fix ASCIICType.h for some Windows compiles
1313
1314         * wtf/ASCIICType.h: Check the compiler, not the OS, since it's the
1315         compiler/library that has the wchar_t that is just a typedef.
1316
1317 2007-10-27  Kevin McCullough  <kmccullough@apple.com>
1318
1319         - BuildFix
1320         - Forgot to change the build step when I changed the filename.
1321
1322         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1323
1324 2007-10-27  Geoffrey Garen  <ggaren@apple.com>
1325
1326         Reviewed by Darin Adler.
1327
1328         Fixed the rest of "ASSERTION FAILED: _hash in KJS::UString::Rep::
1329         computedHash()"
1330         http://bugs.webkit.org/show_bug.cgi?id=15718
1331
1332         * kjs/identifier.cpp: Fixed more cases where an Identifier didn't get a
1333         hash value. Also changed O(n) strlen to O(1) check for empty string.
1334         (KJS::Identifier::add):
1335
1336         * kjs/ustring.cpp: Changed O(n) strlens to O(1) checks for empty string.
1337         (KJS::UString::UString):
1338         (KJS::UString::operator=):
1339
1340 2007-10-27  Darin Adler  <darin@apple.com>
1341
1342         Reviewed by Eric.
1343
1344         - fix pow on Windows
1345
1346         * wtf/MathExtras.h: (wtf_pow): Add a special case for MSVC, which has
1347         a "pow" function that does not properly handle the case where arg1 is
1348         NaN and arg2 is 0.
1349
1350         * kjs/math_object.cpp: (MathFuncImp::callAsFunction): Don't explicity
1351         specify "::pow" -- just "pow" is fine.
1352
1353 2007-10-27  Darin Adler  <darin@apple.com>
1354
1355         Reviewed by Maciej.
1356
1357         - http://bugs.webkit.org/show_bug.cgi?id=15711
1358           force JSImmediate to be inlined for roughly 1.2% SunSpider speedup
1359
1360         * kjs/JSImmediate.h: Put ALWAYS_INLINE on everything.
1361
1362         * kjs/object.h: Removed redundant includes.
1363         * kjs/value.h: Ditto.
1364
1365 2007-10-27  Maciej Stachowiak  <mjs@apple.com>
1366
1367         Reviewed by Mark.
1368
1369         - fixed "ASSERTION FAILED: _hash in KJS::UString::Rep::computedHash()"
1370         http://bugs.webkit.org/show_bug.cgi?id=15718
1371
1372         * kjs/identifier.cpp:
1373         (KJS::Identifier::addSlowCase): Ensure that empty Identifiers have a hash computed,
1374         now that we count on all Identifiers already having one.
1375
1376 2007-10-27  Mark Rowe  <mrowe@apple.com>
1377
1378         Silence a warning.
1379
1380         * kjs/SymbolTable.h:
1381
1382 2007-10-27  Mark Rowe  <mrowe@apple.com>
1383
1384         Gtk build fix.
1385
1386         * kjs/function.h:
1387
1388 2007-10-26  Kevin McCullough  <kmccullough@apple.com>
1389
1390        Rubber stamp by Adam.
1391
1392         - Renamed JSStringRefCOM to JSStringRefBSTR since it he only thing the
1393         files contain are functions that operate on BSTRs.
1394
1395         * API/JSStringRefBSTR.cpp: Copied from API/JSStringRefCOM.cpp.
1396         * API/JSStringRefBSTR.h: Copied from API/JSStringRefCOM.h.
1397         * API/JSStringRefCOM.cpp: Removed.
1398         * API/JSStringRefCOM.h: Removed.
1399         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1400
1401 2007-10-26  Kevin McCullough  <kmccullough@apple.com>
1402
1403         Reviewed by Adam.
1404
1405         - Made JSStringCreateWithBSTR capable of handling null BSTRs.
1406
1407         * API/JSStringRefCOM.cpp:
1408         (JSStringCreateWithBSTR):
1409
1410 2007-10-26  Sam Weinig  <sam@webkit.org>
1411
1412         Windows build fix.
1413
1414         * kjs/SymbolTable.h: Add header gaurd.
1415         * kjs/nodes.h: #include "SymbolTable.h"
1416
1417 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1418
1419         Suggested by Anders Carlsson.
1420
1421         Fixed tyop.
1422
1423         * kjs/function.cpp:
1424         (KJS::ActivationImp::getOwnPropertySlot):
1425
1426 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1427
1428         Suggested by Darin Adler.
1429
1430         Use computedHash(), which is safer than just directly accessing _hash.
1431
1432         * kjs/lookup.cpp:
1433         (KJS::Lookup::findEntry):
1434         (KJS::Lookup::find):
1435
1436 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1437
1438         Build fix: svn add SymbolTable.h
1439
1440         * kjs/SymbolTable.h: Added.
1441         (KJS::SymbolTable::set):
1442         (KJS::SymbolTable::get):
1443
1444 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1445
1446         Build fix: export SymbolTable.h to WebCore.
1447
1448         * JavaScriptCore.xcodeproj/project.pbxproj:
1449
1450 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1451
1452         Comment tweak suggested by Maciej.
1453
1454         * kjs/function.cpp:
1455         (KJS::ActivationImp::getOwnPropertySlot):
1456
1457 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1458
1459         Reviewed by Maciej Stachowiak.
1460
1461         Tweaked property maps to remove 2 branches. 2.5% speedup on SunSpider.
1462
1463         * kjs/property_map.cpp: Use a special no branch accessor to the UString's
1464         hash value. Also, return immediately instead of branching to the end
1465         of the loop if the value is not found.
1466         (KJS::PropertyMap::get):
1467         (KJS::PropertyMap::getLocation):
1468         (KJS::PropertyMap::put):
1469         (KJS::PropertyMap::insert):
1470         (KJS::PropertyMap::remove):
1471         (KJS::PropertyMap::checkConsistency):
1472
1473         * kjs/ustring.h:
1474         (KJS::UString::Rep::computedHash): Special no branch accessor to the
1475         UString's hash value. Used when the caller knows that the hash value
1476         has already been computed. (For example, if the caller got the UString
1477         from an Identifier.)
1478
1479 2007-10-26  Geoffrey Garen  <ggaren@apple.com>
1480
1481         Reviewed by Maciej Stachowiak.
1482
1483         Switched ActivationImp to using a symbol table. For now, though, all
1484         clients take the slow path.
1485
1486         Net .6% speedup on SunSpider.
1487
1488         Slowdowns:
1489             - ActivationImp now mallocs in its constructor
1490             - Local variable hits use an extra level of indirection to retrieve
1491             data
1492             - Local variable misses do two lookups
1493
1494         Speedups:
1495             - Fast initialization of local variables upon function entry
1496
1497         * JavaScriptCore.xcodeproj/project.pbxproj: Added SymbolTable.h
1498
1499         * kjs/function.cpp:
1500         (KJS::ActivationImp::ActivationImp): Malloc a private structure to hold
1501         data that won't fit in a JSCell.
1502         (KJS::ActivationImp::argumentsGetter): Use slow symbol table path for
1503         lookup.
1504         (KJS::ActivationImp::getOwnPropertySlot): ditto
1505         (KJS::ActivationImp::deleteProperty): ditto
1506         (KJS::ActivationImp::put): ditto
1507         (KJS::ActivationImp::createArgumentsObject): ditto
1508
1509         (KJS::ActivationImp::mark): Call JSObject::mark first so that one of
1510         our properties doesn't try to recursively mark us. (This caused a crash
1511         in earlier testing. Not sure why we haven't run into it before.)
1512
1513         * kjs/nodes.cpp: Functions now build a symbol table the first time
1514         they're called.
1515         (KJS::VarDeclNode::evaluate):
1516         (KJS::FunctionBodyNode::FunctionBodyNode):
1517         (KJS::FunctionBodyNode::initializeSymbolTable):
1518         (KJS::FunctionBodyNode::processDeclarations):
1519         (KJS::FunctionBodyNode::processDeclarationsForFunctionCode):
1520         (KJS::FunctionBodyNode::processDeclarationsForProgramCode):
1521
1522         * kjs/nodes.h:
1523         (KJS::FunctionBodyNode::symbolTable):
1524
1525         * wtf/Forward.h: Added Vector.
1526
1527 2007-10-26  Kevin McCullough  <kmccullough@apple.com>
1528
1529         - Corrected function name mistake in this changelog.
1530
1531 2007-10-26  Kevin McCullough  <kmccullough@apple.com>
1532         Reviewed by Sam and Steve.
1533
1534         - Added convenience methods for converting between BSTR and JSStringRefs
1535
1536         * API/JSStringRefCOM.cpp: Added.
1537         (JSStringCreateWithBSTR):
1538         (JSStringCopyBSTR):
1539         * API/JSStringRefCOM.h: Added.
1540         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1541
1542 2007-10-26  Mark Rowe  <mrowe@apple.com>
1543
1544         Windows build fix.
1545
1546         * kjs/collector.cpp:
1547         (KJS::Collector::collect):
1548
1549 2007-10-26  Oliver Hunt  <oliver@apple.com>
1550
1551         Reviewed by Maciej.
1552
1553         Make the JSC GC use a separate heap for JSNumbers to get a 0.7-1.4% progression in SunSpider.
1554
1555         * kjs/CollectorHeapIntrospector.cpp:
1556         (KJS::CollectorHeapIntrospector::init):
1557         (KJS::CollectorHeapIntrospector::enumerate):
1558         * kjs/CollectorHeapIntrospector.h:
1559         * kjs/collector.cpp:
1560         (KJS::Collector::recordExtraCost):
1561         (KJS::Collector::heapAllocate):
1562         (KJS::Collector::allocate):
1563         (KJS::Collector::allocateNumber):
1564         (KJS::Collector::registerThread):
1565         (KJS::Collector::markStackObjectsConservatively):
1566         (KJS::Collector::markMainThreadOnlyObjects):
1567         (KJS::Collector::sweep):
1568         (KJS::Collector::collect):
1569         * kjs/collector.h:
1570         * kjs/internal.h:
1571         (KJS::NumberImp::operator new):
1572           Force numbers to be allocated in the secondary heap.
1573
1574 2007-10-26  Maciej Stachowiak  <mjs@apple.com>
1575
1576         Reviewed by Oliver.
1577
1578         - encourage GCC a little harder to inline a few hot functions for 1.5% improvement on SunSpider.
1579
1580         * kjs/value.h:
1581         (KJS::JSValue::getUInt32):
1582         (KJS::JSValue::getTruncatedInt32):
1583         (KJS::JSValue::toNumber):
1584         * wtf/PassRefPtr.h:
1585         (WTF::PassRefPtr::~PassRefPtr):
1586         * wtf/RefPtr.h:
1587         (WTF::RefPtr::operator->):
1588
1589 2007-10-26  Mark Rowe  <mrowe@apple.com>
1590
1591         Gtk build fix.
1592
1593         * kjs/ExecState.h:
1594
1595 2007-10-26  Maciej Stachowiak  <mjs@apple.com>
1596
1597         Reviewed by Mark.
1598
1599         - Merge Context class fully into ExecState, since they are always created and used together.
1600
1601         No measurable performance impact but this is a useful cleanup.
1602
1603         * JavaScriptCore.pri:
1604         * kjs/ExecState.cpp:
1605         (KJS::ExecState::ExecState):
1606         (KJS::ExecState::~ExecState):
1607         (KJS::ExecState::mark):
1608         (KJS::ExecState::lexicalInterpreter):
1609         * kjs/ExecState.h:
1610         (KJS::ExecState::dynamicInterpreter):
1611         (KJS::ExecState::setException):
1612         (KJS::ExecState::clearException):
1613         (KJS::ExecState::exception):
1614         (KJS::ExecState::exceptionSlot):
1615         (KJS::ExecState::hadException):
1616         (KJS::ExecState::scopeChain):
1617         (KJS::ExecState::callingExecState):
1618         (KJS::ExecState::propertyNames):
1619         * kjs/collector.cpp:
1620         (KJS::Collector::reportOutOfMemoryToAllInterpreters):
1621         * kjs/function.cpp:
1622         (KJS::FunctionImp::callAsFunction):
1623         (KJS::FunctionImp::argumentsGetter):
1624         (KJS::FunctionImp::callerGetter):
1625         (KJS::GlobalFuncImp::callAsFunction):
1626         * kjs/interpreter.cpp:
1627         (KJS::Interpreter::Interpreter):
1628         (KJS::Interpreter::init):
1629         (KJS::Interpreter::evaluate):
1630         (KJS::Interpreter::mark):
1631         * kjs/interpreter.h:
1632         (KJS::Interpreter::setCurrentExec):
1633         (KJS::Interpreter::currentExec):
1634         * kjs/nodes.cpp:
1635         (KJS::currentSourceId):
1636         (KJS::currentSourceURL):
1637         (KJS::ThisNode::evaluate):
1638         (KJS::ResolveNode::evaluate):
1639         (KJS::FunctionCallResolveNode::evaluate):
1640         (KJS::PostfixResolveNode::evaluate):
1641         (KJS::DeleteResolveNode::evaluate):
1642         (KJS::TypeOfResolveNode::evaluate):
1643         (KJS::PrefixResolveNode::evaluate):
1644         (KJS::AssignResolveNode::evaluate):
1645         (KJS::VarDeclNode::evaluate):
1646         (KJS::DoWhileNode::execute):
1647         (KJS::WhileNode::execute):
1648         (KJS::ForNode::execute):
1649         (KJS::ForInNode::execute):
1650         (KJS::ContinueNode::execute):
1651         (KJS::BreakNode::execute):
1652         (KJS::ReturnNode::execute):
1653         (KJS::WithNode::execute):
1654         (KJS::SwitchNode::execute):
1655         (KJS::LabelNode::execute):
1656         (KJS::TryNode::execute):
1657         (KJS::FunctionBodyNode::processDeclarationsFunctionCode):
1658         (KJS::FunctionBodyNode::processDeclarationsProgramCode):
1659         (KJS::FunctionBodyNode::processDeclarations):
1660         (KJS::FuncDeclNode::makeFunction):
1661         (KJS::FuncExprNode::evaluate):
1662
1663 2007-10-26  Mark Rowe  <mrowe@apple.com>
1664
1665         Windows build fix.
1666
1667         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1668
1669 2007-10-26  Mark Rowe  <mrowe@apple.com>
1670
1671         Gtk build fix.
1672
1673         * JavaScriptCore.pri:
1674         * kjs/ExecState.cpp:
1675
1676 2007-10-26  Maciej Stachowiak  <mjs@apple.com>
1677
1678         Reviewed by Oliver.
1679
1680         - moved Context class into ExecState.{h,cpp} in preparation for merging
1681         ExecState and Context classes.
1682
1683         * kjs/ExecState.h: Moved CodeType enum and Context class here in
1684         preparation for merging ExecState and Context.
1685         * kjs/ExecState.cpp: Moved Context class here from Context.cpp.
1686         (KJS::Context::Context):
1687         (KJS::Context::~Context):
1688         (KJS::Context::mark):
1689         * kjs/context.h: Removed.
1690         * kjs/Context.cpp: Removed.
1691         * kjs/function.h: Removed CodeType enum.
1692         * kjs/LabelStack.h: Added. Pulled LabelStack class out of internal.h.
1693         * kjs/internal.h: Removed LabelStack.
1694         * JavaScriptCore.xcodeproj/project.pbxproj: Added new file, removed ones that are gone.
1695         * kjs/collector.cpp: Fixed includes.
1696         * kjs/function.cpp: ditto
1697         * kjs/internal.cpp: ditto
1698         * kjs/interpreter.cpp: ditto
1699         * kjs/lookup.h: ditto
1700         * kjs/nodes.cpp: ditto
1701
1702 2007-10-26  Mark Rowe  <mrowe@apple.com>
1703
1704         Windows build fix.
1705
1706         * kjs/string_object.cpp:
1707         (KJS::StringObjectFuncImp::callAsFunction):
1708
1709 2007-10-25  Darin Adler  <darin@apple.com>
1710
1711         Reviewed by Maciej.
1712
1713         - http://bugs.webkit.org/show_bug.cgi?id=15703
1714           fix numeric functions -- improve correctness and speed
1715
1716         Gives about 1% gain on SunSpider.
1717
1718         * kjs/value.h: Added toIntegerPreserveNan, removed toUInt16.
1719         (KJS::JSValue::toInt32): Changed to call getTruncatedInt32 in a way that works
1720         with both immediate and number values.
1721         (KJS::JSValue::toUInt32): Ditto.
1722         * kjs/value.cpp:
1723         (KJS::JSValue::toInteger): Moved the logic from roundValue here, with a couple
1724         differences. One is that it now correctly returns 0 for NaN, and another is that
1725         there's no special case for 0 or infinity, since the general case already handles
1726         those correctly.
1727         (KJS::JSValue::toIntegerPreserveNaN): Added. Like toInteger, but without the
1728         check for NaN.
1729         (KJS::JSValue::toInt32SlowCase): Call toNumber instead of roundValue. The
1730         truncation done by the typecast already does the necessary truncation that
1731         roundValue was doing.
1732         (KJS::JSValue::toUInt32SlowCase): Ditto.
1733         (KJS::JSValue::toUInt16): Removed.
1734
1735         * kjs/internal.h: Removed roundValue.
1736         * kjs/internal.cpp: Ditto.
1737
1738         * kjs/array_object.cpp: (KJS::ArrayProtoFunc::callAsFunction): Remove unneeded
1739         code to handle NaN in Array.slice; toInteger now never returns NaN as specified.
1740
1741         * kjs/date_object.cpp:
1742         (KJS::fillStructuresUsingTimeArgs): Replaced call to roundValue with a call to
1743         toNumber as specified.
1744         (KJS::DateProtoFunc::callAsFunction): In SetTime case, replaced call to roundValue
1745         with a call to toNumber and timeClip as specified.
1746         (KJS::DateObjectImp::construct): Removed unnecessary checks of numArgs in cases
1747         where the default behavior of toInt32 (returning 0) was already correct. Replaced
1748         call to roundValue with a call to toNumber as specified.
1749         (KJS::DateObjectFuncImp::callAsFunction): Ditto.
1750
1751         * kjs/math_object.cpp: (MathFuncImp::callAsFunction): Removed unnecessary special
1752         cases for the pow function that the library already handles correctly.
1753
1754         * kjs/number_object.cpp: (NumberProtoFunc::callAsFunction): Changed ToString to
1755         call toIntegerPreserveNaN, so we can continue to handle the NaN case differently.
1756         The real toInteger now returns 0 for NaN. Took out unneeded special case in
1757         ToFixed for undefined; was only needed because our toInteger was wrong. Same
1758         thing in ToExponential. Changed ToPrecision to call toIntegerPreserveNaN.
1759
1760         * kjs/string_object.cpp:
1761         (KJS::StringProtoFunc::callAsFunction): Took out CharAt and CharCodeAt special
1762         cases for undefined that were only needed because toInteger was wrong. Same in
1763         IndexOf, and was able to remove some special cases. In LastIndexOf, used
1764         toIntegerPreserveNaN, but was able to remove some special cases there too.
1765         Changed Substr implementation to preserve correct behavior with the change
1766         to toInteger and match the specification. Also made sure we weren't converting
1767         an out of range double to an int.
1768         (KJS::StringObjectFuncImp::callAsFunction): Changed constructor to just use
1769         toUInt32, because truncating toUInt32 to 16 bits is the same thing and there's
1770         no reason to have toUInt16 as a second, less-optimized function that's only
1771         called at this one call site.
1772
1773         * wtf/MathExtras.h: Added trunc function for Windows.
1774
1775 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1776
1777         Reviewed by Maciej Stachowiak.
1778
1779         Tweaked the inner hashtable lookup loop to remove a branch in the "not
1780         found" case. .5% speedup on SunSpider.
1781
1782         * JavaScriptCore.xcodeproj/project.pbxproj:
1783         * wtf/HashTable.h:
1784         (WTF::::lookup):
1785
1786 2007-10-25  Maciej Stachowiak  <mjs@apple.com>
1787
1788         Reviewed by Oliver.
1789
1790         - fold together toPrimitive() and toNumber() conversions for 0.5% gain on SunSpider
1791
1792         * kjs/nodes.cpp:
1793         (KJS::SubNode::evaluate): Subtract directly, since toPrimitive() is not
1794         adding any value over toNumber() here.
1795         (KJS::valueForReadModifyAssignment): Ditto.
1796         (KJS::lessThan): Use new getPrimitiveNumber() method to avoid some virtual calls
1797         and branches.
1798         (KJS::lessThanEq): Ditto.
1799         * JavaScriptCore.exp: Export new functions as needed.
1800         * kjs/value.h:
1801         (KJS::JSValue::toPrimitive): Fixed formatting.
1802         (KJS::JSValue::getPrimitiveNumber): New method - this simultaneously converts
1803         to number and tells you whether a toPrimitive() conversion with a Number hint
1804         would have given a string.
1805         * kjs/internal.cpp:
1806         (KJS::StringImp::getPrimitiveNumber): Implemented.
1807         (KJS::NumberImp::getPrimitiveNumber): ditto
1808         (KJS::GetterSetterImp::getPrimitiveNumber): ditto
1809         (KJS::StringImp::toPrimitive): Fixed formatting.
1810         (KJS::NumberImp::toPrimitive): ditto
1811         (KJS::GetterSetterImp::toPrimitive): ditto
1812         * kjs/internal.h:
1813         * kjs/object.cpp:
1814         (KJS::JSObject::getPrimitiveNumber): Implemented.
1815         * kjs/object.h:
1816
1817 2007-10-25  Sam Weinig  <sam@webkit.org>
1818
1819         Reviewed by Adam Roben.
1820
1821         Remove JSStringRefCFHack from windows as it is no longer needed.
1822
1823         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1824
1825 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1826
1827         Reviewed by Oliver Hunt.
1828
1829         Rolled out my last patch. It turns out that I needed 2 words, not 1,
1830         so it didn't help.
1831
1832 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1833
1834         Reviewed by Oliver Hunt.
1835
1836         Fixed http://bugs.webkit.org/show_bug.cgi?id=15694
1837         Shrink the size of an activation object by 1 word
1838
1839         This is in preparation for adding a symbol table to the activation
1840         object.
1841
1842         The basic strategy here is to rely on the mutual exclusion between
1843         the arguments object pointer and the function pointer (you only need
1844         the latter in order to create the former), and store them in the same
1845         place. The LazyArgumentsObject class encapsulates this strategy.
1846
1847         Also inlined the ArgumentsImp constructor, for good measure.
1848
1849         SunSpider reports no regression. Regression tests pass.
1850
1851         * JavaScriptCore.xcodeproj/project.pbxproj:
1852         * kjs/Context.cpp:
1853         (KJS::Context::~Context):
1854         * kjs/function.cpp:
1855         (KJS::ActivationImp::LazyArgumentsObject::createArgumentsObject):
1856         (KJS::ActivationImp::LazyArgumentsObject::mark):
1857         (KJS::):
1858         (KJS::ActivationImp::argumentsGetter):
1859         (KJS::ActivationImp::mark):
1860         * kjs/function.h:
1861         (KJS::ActivationImp::LazyArgumentsObject::LazyArgumentsObject):
1862         (KJS::ActivationImp::LazyArgumentsObject::getOrCreate):
1863         (KJS::ActivationImp::LazyArgumentsObject::resetArguments):
1864         (KJS::ActivationImp::LazyArgumentsObject::setArgumentsObject):
1865         (KJS::ActivationImp::LazyArgumentsObject::argumentsObject):
1866         (KJS::ActivationImp::LazyArgumentsObject::setFunction):
1867         (KJS::ActivationImp::LazyArgumentsObject::function):
1868         (KJS::ActivationImp::LazyArgumentsObject::createdArgumentsObject):
1869         (KJS::ActivationImp::LazyArgumentsObject::):
1870         (KJS::ActivationImp::ActivationImp::ActivationImp):
1871         (KJS::ActivationImp::resetArguments):
1872
1873 2007-10-25  Adam Roben  <aroben@apple.com>
1874
1875         Change JavaScriptCore.vcproj to use DerivedSources.make
1876
1877         We were trying to emulate the logic of make in
1878         build-generated-files.sh, but we got it wrong. We now use a
1879         build-generated-files very much like the one that WebCore uses to
1880         invoke make.
1881
1882         We also now only have a Debug configuration of dftables which we build
1883         even when doing a Release build of JavaScriptCore. dftables also no
1884         longer has the "_debug" name suffix.
1885
1886         Changes mostly made by Darin, reviewed by me.
1887
1888         * DerivedSources.make: Add a variable to set the extension used for
1889         the dftables executable.
1890         * JavaScriptCore.vcproj/JavaScriptCore.sln: Updated to use Debug
1891         dftables in Release configurations.
1892         * JavaScriptCore.vcproj/JavaScriptCoreSubmit.sln: Ditto.
1893         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
1894            - Updated include path to point to the new location of the derived
1895              sources.
1896            - Modified pre-build event to pass the right arguments to
1897              build-generated-files.sh and not call dftables directly.
1898            - Added the derived source files to the project.
1899            - Removed grammarWrapper.cpp, which isn't needed now that we're
1900              compiling grammar.cpp directly.
1901         * JavaScriptCore.vcproj/JavaScriptCore/build-generated-files.sh:
1902         Slightly modified from the WebCore version.
1903         * JavaScriptCore.vcproj/JavaScriptCore/grammarWrapper.cpp: Removed.
1904         * JavaScriptCore.vcproj/dftables/dftables.vcproj:
1905             - Changed the output location to match Mac.
1906             - Removed the Release configuration.
1907             - Removed the _debug suffix.
1908
1909 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1910
1911         Reviewed by Eric Seidel.
1912
1913         Slightly elaborated the differences between declaration procesing in
1914         Function Code and Program Code.
1915
1916         .3% speedup on SunSpider.
1917
1918         * kjs/nodes.cpp:
1919         (KJS::FunctionBodyNode::processDeclarationsFunctionCode):
1920         (KJS::FunctionBodyNode::processDeclarationsProgramCode): Store a
1921         minimum set of attributes instead of recomputing all the time. Also,
1922         ignore m_parameters, since programs don't have arguments.
1923
1924 2007-10-25  Eric Seidel  <eric@webkit.org>
1925
1926         Reviewed by Maciej.
1927
1928         More preparation work before adding long-running mode to testkjs.
1929
1930         * kjs/testkjs.cpp:
1931         (TestFunctionImp::callAsFunction):
1932         (prettyPrintScript):
1933         (runWithScripts):
1934         (parseArguments):
1935         (kjsmain):
1936         (fillBufferWithContentsOfFile):
1937
1938 2007-10-25  Eric Seidel  <eric@webkit.org>
1939
1940         Reviewed by Maciej.
1941
1942         Bring testkjs code out of the dark ages in preparation for more
1943         radical improvements (like long-running testing support!)
1944
1945         * kjs/testkjs.cpp:
1946         (TestFunctionImp::callAsFunction):
1947         (setupInterpreter):
1948         (doIt):
1949         (fillBufferWithContentsOfFile):
1950
1951 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1952
1953         Reviewed by Maciej Stachowiak.
1954
1955         Make a fast path for declaration processing inside Function Code.
1956
1957         Lifted declaration processing code up from individual declaration nodes
1958         and into processDeclarations.
1959
1960         Broke out processDeclarations into two cases, depending on the type of
1961         code. This eliminates 2 branches, and facilitates more radical
1962         divergeance in the future.
1963
1964         2.5% SunSpider speedup.
1965
1966         * JavaScriptCore.xcodeproj/project.pbxproj:
1967         * kjs/nodes.cpp:
1968         (KJS::FunctionBodyNode::initializeDeclarationStacks):
1969         (KJS::FunctionBodyNode::processDeclarationsFunctionCode):
1970         (KJS::FunctionBodyNode::processDeclarationsProgramCode):
1971         (KJS::FunctionBodyNode::execute):
1972         (KJS::FuncDeclNode::makeFunction):
1973         * kjs/nodes.h:
1974         (KJS::):
1975
1976 2007-10-25  Maciej Stachowiak  <mjs@apple.com>
1977
1978         Reviewed by Adam.
1979
1980         - add header includes needed on platforms that don't use AllInOneFile.cpp
1981
1982         * API/JSCallbackObject.cpp:
1983         * kjs/Context.cpp:
1984         * kjs/ExecState.cpp:
1985         * kjs/array_instance.cpp:
1986         * kjs/function_object.cpp:
1987         * kjs/interpreter.cpp:
1988         * kjs/nodes.cpp:
1989
1990 2007-10-25  Eric Seidel  <eric@webkit.org>
1991
1992         Reviewed by Geoff.
1993
1994         * JavaScriptCore.xcodeproj/project.pbxproj: re-mark JSGlobalObject.h as private
1995
1996 2007-10-25  Geoffrey Garen  <ggaren@apple.com>
1997
1998         Reviewed by Maciej Stachowiak.
1999
2000         Fixed http://bugs.webkit.org/show_bug.cgi?id=15683
2001         Re-order declaration initialization to avoid calling hasProperty inside
2002         VarDeclNode::processDeclaration
2003
2004         .7% speedup on SunSpider.
2005
2006         * kjs/function.h:
2007         * kjs/function.cpp: Merged parameter processing into FunctionBodyNode's
2008         other processing of declared symbols, so the order of execution could
2009         change.
2010
2011         * kjs/nodes.cpp:
2012         (KJS::VarDeclNode::getDeclarations): Added special case for the
2013         "arguments" property name, explained in the comment.
2014
2015         (KJS::VarDeclNode::processDeclaration): Removed call to hasProperty
2016         in the case of function code, since we know the declared symbol
2017         management will resolve conflicts between symbols. Yay!
2018
2019         (KJS::VarDeclListNode::getDeclarations): Now that VarDeclNode's
2020         implementation of getDeclarations is non-trivial, we can't take a
2021         short-cut here any longer -- we need to put the VarDecl node on the
2022         stack so it gets processed normally.
2023
2024         (KJS::FunctionBodyNode::processDeclarations): Changed the order of
2025         processing to enforce mutual exclusion rules.
2026
2027         * kjs/nodes.h:
2028         (KJS::DeclarationStacks::DeclarationStacks): Structure includes an
2029         ExecState now, for fast access to the "arguments" property name.
2030
2031 2007-10-24  Eric Seidel  <eric@webkit.org>
2032
2033         Reviewed by Maciej.
2034
2035         Add a JSGlobalObject class and remove the InterpreterMap
2036         http://bugs.webkit.org/show_bug.cgi?id=15681
2037
2038         This required making JSCallbackObject a template class to allow for
2039         JSGlobalObjects with JSCallbackObject functionality.
2040
2041         SunSpider claims this was a 0.5% speedup.
2042
2043         * API/JSCallbackObject.cpp:
2044         (KJS::):
2045         * API/JSCallbackObject.h:
2046         * API/JSCallbackObjectFunctions.h: Copied from API/JSCallbackObject.cpp.
2047         (KJS::::JSCallbackObject):
2048         (KJS::::init):
2049         (KJS::::~JSCallbackObject):
2050         (KJS::::initializeIfNeeded):
2051         (KJS::::className):
2052         (KJS::::getOwnPropertySlot):
2053         (KJS::::put):
2054         (KJS::::deleteProperty):
2055         (KJS::::implementsConstruct):
2056         (KJS::::construct):
2057         (KJS::::implementsHasInstance):
2058         (KJS::::hasInstance):
2059         (KJS::::implementsCall):
2060         (KJS::::callAsFunction):
2061         (KJS::::getPropertyNames):
2062         (KJS::::toNumber):
2063         (KJS::::toString):
2064         (KJS::::setPrivate):
2065         (KJS::::getPrivate):
2066         (KJS::::inherits):
2067         (KJS::::cachedValueGetter):
2068         (KJS::::staticValueGetter):
2069         (KJS::::staticFunctionGetter):
2070         (KJS::::callbackGetter):
2071         * API/JSClassRef.cpp:
2072         (OpaqueJSClass::prototype):
2073         * API/JSContextRef.cpp:
2074         (JSGlobalContextCreate):
2075         * API/JSObjectRef.cpp:
2076         (JSObjectMake):
2077         (JSObjectGetPrivate):
2078         (JSObjectSetPrivate):
2079         * API/JSValueRef.cpp:
2080         (JSValueIsObjectOfClass):
2081         * JavaScriptCore.exp:
2082         * JavaScriptCore.xcodeproj/project.pbxproj:
2083         * bindings/c/c_utility.cpp:
2084         (KJS::Bindings::convertValueToNPVariant):
2085         * bindings/jni/jni_jsobject.cpp:
2086         * bindings/objc/objc_utility.mm:
2087         (KJS::Bindings::convertValueToObjcValue):
2088         * kjs/Context.cpp:
2089         (KJS::Context::Context):
2090         * kjs/ExecState.cpp:
2091         (KJS::ExecState::lexicalInterpreter):
2092         * kjs/JSGlobalObject.h: Added.
2093         (KJS::JSGlobalObject::JSGlobalObject):
2094         (KJS::JSGlobalObject::isGlobalObject):
2095         (KJS::JSGlobalObject::interpreter):
2096         (KJS::JSGlobalObject::setInterpreter):
2097         * kjs/array_instance.cpp:
2098         * kjs/context.h:
2099         * kjs/function.cpp:
2100         (KJS::FunctionImp::callAsFunction):
2101         (KJS::GlobalFuncImp::callAsFunction):
2102         * kjs/interpreter.cpp:
2103         (KJS::Interpreter::Interpreter):
2104         (KJS::Interpreter::init):
2105         (KJS::Interpreter::~Interpreter):
2106         (KJS::Interpreter::globalObject):
2107         (KJS::Interpreter::initGlobalObject):
2108         (KJS::Interpreter::evaluate):
2109         * kjs/interpreter.h:
2110         * kjs/lookup.h:
2111         (KJS::cacheGlobalObject):
2112         * kjs/object.h:
2113         (KJS::JSObject::isGlobalObject):
2114         * kjs/testkjs.cpp:
2115
2116 2007-10-24  Eric Seidel  <eric@webkit.org>
2117
2118         Build fix for Gtk, no review.
2119
2120         * kjs/collector.cpp: #include "context.h"
2121
2122 2007-10-24  Eric Seidel  <eric@webkit.org>
2123
2124         Reviewed by Maciej.
2125
2126         Stop checking isOutOfMemory after every allocation, instead let the collector
2127         notify all ExecStates if we ever hit this rare condition.
2128
2129         SunSpider claims this was a 2.2% speedup.
2130
2131         * kjs/collector.cpp:
2132         (KJS::Collector::collect):
2133         (KJS::Collector::reportOutOfMemoryToAllInterpreters):
2134         * kjs/collector.h:
2135         * kjs/nodes.cpp:
2136         (KJS::TryNode::execute):
2137
2138 2007-10-24  Mark Rowe  <mrowe@apple.com>
2139
2140         Gtk build fix.
2141
2142         * kjs/identifier.h:  Remove extra qualification.
2143
2144 2007-10-24  Geoffrey Garen  <ggaren@apple.com>
2145
2146         Reviewed by Sam Weinig.
2147
2148         Disable ALWAYS_INLINE in debug builds, since it drives the debugger
2149         crazy.
2150
2151         * wtf/AlwaysInline.h:
2152
2153 2007-10-24  Geoffrey Garen  <ggaren@apple.com>
2154
2155         Reviewed by Sam Weinig.
2156
2157         Inlined the fast path for creating an Identifier from an Identifier.
2158
2159         This is a .4% speedup on SunSpider overall, but as big as a 2.5%
2160         speedup on certain individual tests. 65% of the Identifiers creating
2161         by SunSpider are already Identifiers.
2162
2163         (The main reason I'm making this change is that it resolves a large
2164         regression in a patch I haven't checked in yet.)
2165
2166         * JavaScriptCore.exp:
2167         * kjs/identifier.cpp:
2168         (KJS::Identifier::addSlowCase):
2169         * kjs/identifier.h:
2170         (KJS::Identifier::Identifier::add):
2171
2172 2007-10-24  Lars Knoll  <lars@trolltech.com>
2173
2174         Reviewed by Simon.
2175
2176         some changes to the way JS values are converted to Qt values in the script bindings. Added support for converting JS arrays into QStringList's.
2177
2178         * bindings/qt/qt_instance.cpp:
2179         (KJS::Bindings::QtInstance::invokeMethod):
2180         * bindings/qt/qt_runtime.cpp:
2181         (KJS::Bindings::convertValueToQVariant):
2182         (KJS::Bindings::QtField::setValueToInstance):
2183
2184 2007-10-24  Oliver Hunt  <oliver@apple.com>
2185
2186         Reviewed by Darin.
2187
2188         Remove old relation method, replace with specialised LessThan and lessThenEq functions for a 0.5-0.6% improvement in SunSpider
2189
2190         * kjs/nodes.cpp:
2191         (KJS::lessThan):
2192         (KJS::lessThanEq):
2193         (KJS::LessNode::evaluate):
2194         (KJS::GreaterNode::evaluate):
2195         (KJS::LessEqNode::evaluate):
2196         (KJS::GreaterEqNode::evaluate):
2197         * kjs/operations.cpp:
2198         * kjs/operations.h:
2199
2200 2007-10-24  Eric Seidel  <eric@webkit.org>
2201
2202         Reviewed by darin.
2203
2204         * kjs/nodes.h:
2205         (KJS::ImmediateNumberNode::): Fix ASSERT correctness (and debug build!)
2206
2207 2007-10-24  Darin Adler  <darin@apple.com>
2208
2209         Reviewed by Eric.
2210
2211         * kjs/object.cpp: (KJS::JSObject::defaultValue): Get rid of a little
2212         Identifier ref/deref for what SunSpider claims is a 0.4% speedup.
2213
2214 2007-10-24  Darin Adler  <darin@apple.com>
2215
2216         Reviewed by Maciej.
2217
2218         - separate out the code to create a hash table the first time from the code
2219           to rehash
2220
2221         SunSpider claims this was a 0.7% speedup.
2222
2223         * kjs/property_map.cpp:
2224         (KJS::PropertyMap::expand): Changed to call either createTable or rehash.
2225         (KJS::PropertyMap::createTable): Added. For the case where we had no table.
2226         (KJS::PropertyMap::rehash): Removed code needed only in the case where we
2227         had no table.
2228         * kjs/property_map.h: Added createTable.
2229
2230 2007-10-24  Eric Seidel  <eric@webkit.org>
2231
2232         Reviewed by darin.
2233
2234         Add ImmediateNumberNode to hold a JSValue* instead of a double for numbers
2235         which can be represented by JSImmediate.
2236
2237         SunSpider claims this was a 0.6% speedup.
2238
2239         * kjs/grammar.y:
2240         * kjs/nodes.cpp:
2241         (KJS::NumberNode::evaluate):
2242         (KJS::ImmediateNumberNode::evaluate):
2243         * kjs/nodes.h:
2244         (KJS::Node::):
2245         (KJS::ImmediateNumberNode::):
2246         * kjs/nodes2string.cpp:
2247         (ImmediateNumberNode::streamTo):
2248
2249 2007-10-24  Darin Adler  <darin@apple.com>
2250
2251         Reviewed by Maciej.
2252
2253         - http://bugs.webkit.org/show_bug.cgi?id=15657
2254           change static hash tables to use powers of two for speed
2255
2256         Seems to give 0.7% SunSpider speedup.
2257
2258         * kjs/create_hash_table: Updated to generate new format.
2259         * kjs/lookup.cpp:
2260         (KJS::keysMatch): Took out unneeded typecast.
2261         (KJS::findEntry): Updated to expect table type 3 -- changed the printf to a plain old assert.
2262         Replaced the modulus with a bit mask.
2263         (KJS::Lookup::findEntry): Get the hash directly, since we know identifiers already have computed
2264         their hash -- saves a branch.
2265         (KJS::Lookup::find): Ditto.
2266         * kjs/lookup.h: Changed attr from 2-byte value to one-byte value. Replaced hashSize with hashSizeMask.
2267
2268 2007-10-24  Maciej Stachowiak  <mjs@apple.com>
2269
2270         Reviewed by Darin.
2271
2272         - remove KJS_CHECKEXCEPTIONs in places where exceptions can't happen for 0.6% SunSpider speedup
2273
2274         * kjs/nodes.cpp:
2275         (KJS::DoWhileNode::execute):
2276         (KJS::WhileNode::execute):
2277         (KJS::ForNode::execute):
2278         (KJS::ForInNode::execute):
2279         (KJS::SourceElementsNode::execute):
2280
2281 2007-10-23  Darin Adler  <darin@apple.com>
2282
2283         Reviewed by Maciej.
2284
2285         * kjs/JSImmediate.h: (KJS::JSImmediate::getUInt32):
2286         Changed an && to an & for a 1% gain in SunSpider.
2287
2288 2007-10-23  Oliver Hunt  <oliver@apple.com>
2289
2290         Reviewed by Maciej.
2291
2292         Reduce branching in implementations of some operator implementations, yielding 1.3% boost to SunSpider.
2293
2294         * kjs/nodes.cpp:
2295         (KJS::MultNode::evaluate):
2296         (KJS::DivNode::evaluate):
2297         (KJS::ModNode::evaluate):
2298         (KJS::add):
2299         (KJS::sub):
2300         (KJS::AddNode::evaluate):
2301         (KJS::SubNode::evaluate):
2302         (KJS::valueForReadModifyAssignment):
2303         * kjs/operations.cpp:
2304         * kjs/operations.h:
2305
2306 2007-10-23  Oliver Hunt  <oliver@apple.com>
2307
2308         Reviewed by Maciej.
2309
2310         Separating all of the simple (eg. non-read-modify-write) binary operators
2311         into separate classes in preparation for further JS optimisations.
2312
2313         Happily this produces a 0.8% to 1.0% performance increase in SunSpider with
2314         no further work.
2315
2316         * JavaScriptCore.xcodeproj/project.pbxproj:
2317         * kjs/grammar.y:
2318         * kjs/nodes.cpp:
2319         (KJS::MultNode::evaluate):
2320         (KJS::DivNode::evaluate):
2321         (KJS::ModNode::evaluate):
2322         (KJS::AddNode::evaluate):
2323         (KJS::SubNode::evaluate):
2324         (KJS::LeftShiftNode::evaluate):
2325         (KJS::RightShiftNode::evaluate):
2326         (KJS::UnsignedRightShiftNode::evaluate):
2327         (KJS::LessNode::evaluate):
2328         (KJS::GreaterNode::evaluate):
2329         (KJS::LessEqNode::evaluate):
2330         (KJS::GreaterEqNode::evaluate):
2331         (KJS::InstanceOfNode::evaluate):
2332         (KJS::InNode::evaluate):
2333         (KJS::EqualNode::evaluate):
2334         (KJS::NotEqualNode::evaluate):
2335         (KJS::StrictEqualNode::evaluate):
2336         (KJS::NotStrictEqualNode::evaluate):
2337         (KJS::BitAndNode::evaluate):
2338         (KJS::BitXOrNode::evaluate):
2339         (KJS::BitOrNode::evaluate):
2340         (KJS::LogicalAndNode::evaluate):
2341         (KJS::LogicalOrNode::evaluate):
2342         * kjs/nodes.h:
2343         (KJS::MultNode::):
2344         (KJS::DivNode::):
2345         (KJS::ModNode::):
2346         (KJS::AddNode::):
2347         (KJS::SubNode::):
2348         (KJS::LeftShiftNode::):
2349         (KJS::RightShiftNode::):
2350         (KJS::UnsignedRightShiftNode::):
2351         (KJS::LessNode::):
2352         (KJS::GreaterNode::):
2353         (KJS::LessEqNode::):
2354         (KJS::GreaterEqNode::):
2355         (KJS::InstanceOfNode::):
2356         (KJS::InNode::):
2357         (KJS::EqualNode::):
2358         (KJS::NotEqualNode::):
2359         (KJS::StrictEqualNode::):
2360         (KJS::NotStrictEqualNode::):
2361         (KJS::BitAndNode::):
2362         (KJS::BitOrNode::):
2363         (KJS::BitXOrNode::):
2364         (KJS::LogicalAndNode::):
2365         (KJS::LogicalOrNode::):
2366         * kjs/nodes2string.cpp:
2367         (MultNode::streamTo):
2368         (DivNode::streamTo):
2369         (ModNode::streamTo):
2370         (AddNode::streamTo):
2371         (SubNode::streamTo):
2372         (LeftShiftNode::streamTo):
2373         (RightShiftNode::streamTo):
2374         (UnsignedRightShiftNode::streamTo):
2375         (LessNode::streamTo):
2376         (GreaterNode::streamTo):
2377         (LessEqNode::streamTo):
2378         (GreaterEqNode::streamTo):
2379         (InstanceOfNode::streamTo):
2380         (InNode::streamTo):
2381         (EqualNode::streamTo):
2382         (NotEqualNode::streamTo):
2383         (StrictEqualNode::streamTo):
2384         (NotStrictEqualNode::streamTo):
2385         (BitAndNode::streamTo):
2386         (BitXOrNode::streamTo):
2387         (BitOrNode::streamTo):
2388         (LogicalAndNode::streamTo):
2389
2390 2007-10-23  Darin Adler  <darin@apple.com>
2391
2392         Reviewed by Maciej.
2393
2394         - fix http://bugs.webkit.org/show_bug.cgi?id=15639
2395           fix Math.abs(0), Math.ceil(-0), and Math.floor(-0)
2396
2397         Test: fast/js/math.html
2398
2399         * kjs/math_object.cpp: (MathFuncImp::callAsFunction):
2400         Fix abs to look at the sign bit. Add a special case for values in the range
2401         between -0 and -1 and a special case for ceil and for -0 for floor.
2402
2403 2007-10-23  Darin Adler  <darin@apple.com>
2404
2405         Reviewed by Eric.
2406
2407         - streamline exception handling code for a >1% speed-up of SunSpider
2408
2409         * kjs/nodes.cpp: Changed macros to use functions for everything that's not
2410         part of normal execution. We'll take function call overhead when propagating
2411         an exception or out of memory.
2412         (KJS::createOutOfMemoryCompletion): Added.
2413         (KJS::substitute): Use append instead of the relatively inefficient + operator.
2414         (KJS::Node::rethrowException): Added.
2415         * kjs/nodes.h: Added rethrowException.
2416
2417 2007-10-22  Darin Adler  <darin@apple.com>
2418
2419         Reviewed by Maciej.
2420
2421         - fix http://bugs.webkit.org/show_bug.cgi?id=15636
2422           some JavaScriptCore regression tests are failing due to numeric conversion
2423
2424         This should restore correctness and make speed better too, restoring some
2425         of the optimization we lost in my last check-in.
2426
2427         * kjs/JSImmediate.h:
2428         (KJS::JSImmediate::getTruncatedInt32): Added. Uses the range checking idiom
2429         I used in my patch yesterday.
2430         (KJS::JSImmediate::getTruncatedUInt32): Ditto.
2431
2432         * kjs/internal.h: Removed getInt32 and added getTruncatedInt/UInt32.
2433         * kjs/internal.cpp:
2434         (KJS::NumberImp::getUInt32): Changed to always use double, since I can't find
2435         a way to write this more efficiently for float.
2436         (KJS::NumberImp::getTruncatedInt32): Added.
2437         (KJS::NumberImp::getTruncatedUInt32): Added.
2438
2439         * kjs/value.h: Removed getInt32 and added getTruncatedInt/UInt32.
2440         (KJS::JSValue::getUInt32):
2441         (KJS::JSValue::getTruncatedInt32): Added.
2442         (KJS::JSValue::getTruncatedUInt32): Added.
2443         (KJS::JSValue::toInt32): Changed getInt32 call to getTruncatedInt32.
2444         (KJS::JSValue::toUInt32): Changed getUInt32 call to getTruncatedUInt32.
2445         * kjs/value.cpp:
2446         (KJS::JSCell::getTruncatedInt32): Added.
2447         (KJS::JSCell::getTruncatedUInt32): Added.
2448         (KJS::JSValue::toInteger): Changed getUInt32 call to getTruncatedInt32.
2449         (KJS::JSValue::toInt32SlowCase): Removed extra getInt32 call I accidentally
2450         had left in here.
2451         (KJS::JSValue::toUInt32SlowCase): Ditto.
2452         (KJS::JSValue::toUInt16): Changed getUInt32 call to getTruncatedUInt32.
2453
2454         * JavaScriptCore.exp: Updated.
2455
2456 2007-10-22  Darin Adler  <darin@apple.com>
2457
2458         Reviewed by Geoff.
2459
2460         - fix http://bugs.webkit.org/show_bug.cgi?id=15632
2461           js1_5/Array/array-001.js test failing
2462
2463         One of the JavaScriptCore tests was failing; it failed because of
2464         my change to NumberImp::getUInt32. The incorrect code I copied was
2465         from JSImmediate::getUInt32, and was a pre-existing bug.
2466
2467         This patch fixes correctness, but will surely slow down SunSpider.
2468         We may be able to code this tighter and get the speed back.
2469
2470         * kjs/JSImmediate.h:
2471         (KJS::JSImmediate::getInt32): Renamed from toInt32 to more accurately
2472         reflect the fact that this function only returns true if the value is
2473         accurate (no fractional part, etc.). Changed code so that it returns
2474         false when the value has a fraction.
2475         (KJS::JSImmediate::getUInt32): Ditto.
2476
2477         * kjs/internal.cpp:
2478         (KJS::NumberImp::getInt32): Changed code so that it returns false when
2479         the value has a fraction. Restores the old behavior.
2480         (KJS::NumberImp::getUInt32): Ditto.
2481
2482         * kjs/value.h:
2483         (KJS::JSValue::getInt32): Updated for name change.
2484         (KJS::JSValue::getUInt32): Ditto.
2485         (KJS::JSValue::toInt32): Ditto.
2486         (KJS::JSValue::toUInt32): Ditto.
2487
2488 2007-10-22  Darin Adler  <darin@apple.com>
2489
2490         Reviewed by Brady.
2491
2492         - fix crash seen when running JavaScriptCore tests
2493
2494         * kjs/array_instance.cpp: (KJS::ArrayInstance::mark):
2495         Copy and paste error: I accidentally had code here that was
2496         making a copy of the HashMap -- that's illegal inside a mark
2497         function and was unnecessary. The other callsite was modifying
2498         the map as it iterated it, but this function is not.
2499
2500 2007-10-22  Maciej Stachowiak  <mjs@apple.com>
2501
2502         Reviewed by Oliver.
2503
2504         - Avoid moving floats into integer registers in jsNumber() for 3% speedup on SunSpider
2505         http://bugs.webkit.org/show_bug.cgi?id=15627
2506
2507         * kjs/JSImmediate.h:
2508         (KJS::JSImmediate::fromDouble): Avoid moving floats to integer
2509         registers since this is very slow.
2510
2511 2007-10-22  Darin Adler  <darin@apple.com>
2512
2513         Reviewed by Eric Seidel.
2514
2515         - http://bugs.webkit.org/show_bug.cgi?id=15617
2516           improve speed of integer conversions
2517
2518         Makes SunSpider 6% faster.
2519
2520         * kjs/JSImmediate.h: Added toInt32 and toUInt32, with separate versions for
2521         32-bit and 64-bit.
2522         * kjs/value.h:
2523         (KJS::JSValue::getUInt32): Call JSImmediate::toUInt32.
2524
2525         * kjs/internal.h: Added getInt32.
2526         * kjs/internal.cpp:
2527         (KJS::NumberImp::getInt32): Added.
2528         (KJS::NumberImp::getUInt32): Replaced with more-optimal implementation
2529         stolen from JSValue.
2530
2531         * kjs/value.h:
2532         (KJS::jsNumber): Marked ALWAYS_INLINE, because this wasn't getting
2533         inlined.
2534         (KJS::JSValue::getInt32): Added.
2535         (KJS::JSValue::getUInt32): Changed to call the new JSImmediate::toUInt32
2536         to avoid converting from float to double.
2537         (KJS::JSValue::toInt32): Made inline, separated out the slow case.
2538         (KJS::JSValue::toUInt32): Ditto.
2539         * kjs/value.cpp:
2540         (KJS::JSCell::getInt32): Added.
2541         (KJS::JSValue::toInt32SlowCase): Renamed from toInt32. Changed to use the
2542         new getInt32. Added a faster case for in-range numbers.
2543         (KJS::JSValue::toUInt32SlowCase): Ditto.
2544         (KJS::JSValue::toUInt16): Added a faster case for in-range numbers.
2545
2546         * JavaScriptCore.exp: Updated for changes.
2547
2548 2007-10-22  Adam Roben  <aroben@apple.com>
2549
2550         Windows build fix
2551
2552         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj: Turn off
2553         warning about implicit conversion to bool.
2554
2555 2007-10-22  Mark Rowe  <mrowe@apple.com>
2556
2557         Gtk build fix.
2558
2559         * kjs/array_instance.cpp:
2560
2561 2007-10-22  Darin Adler  <darin@apple.com>
2562
2563         Reviewed by Maciej.
2564
2565         - http://bugs.webkit.org/show_bug.cgi?id=15606
2566           make cut-off for sparse vs. dense arrays smarter for speed with large arrays
2567
2568         Makes the morph test in SunSpider 26% faster, and the overall
2569         benchmark 3% faster.
2570
2571         This also fixes some small problems we had with the distinction
2572         between nonexistent and undefined values in arrays.
2573
2574         * kjs/array_instance.h: Tweaked formatting and naming.
2575         * kjs/array_instance.cpp: Copied from kjs/array_object.cpp.
2576         (KJS::storageSize): Added. Computes the size of the storage given a vector length.
2577         (KJS::increasedVectorLength): Added. Implements the rule for resizing the vector.
2578         (KJS::isDenseEnoughForVector): Added.
2579         (KJS::ArrayInstance::ArrayInstance): Initialize the new fields.
2580         (KJS::ArrayInstance::~ArrayInstance): Since m_storage is now never 0, delete it.
2581         (KJS::ArrayInstance::getItem): Updated for name changes.
2582         (KJS::ArrayInstance::lengthGetter): Ditto.
2583         (KJS::ArrayInstance::inlineGetOwnPropertySlot): Added. Allows both versions of
2584         getOwnPropertySlot to share more code.
2585         (KJS::ArrayInstance::getOwnPropertySlot): Just refactored, no code change.
2586         (KJS::ArrayInstance::put): Added logic for extending the vector as long as the
2587         array is dense enough. Also keep m_numValuesInVector up to date.
2588         (KJS::ArrayInstance::deleteProperty): Added code to keep m_numValuesInVector
2589         up to date.
2590         (KJS::ArrayInstance::getPropertyNames): Fixed bug where this would omit names
2591         for array indices with undefined values.
2592         (KJS::ArrayInstance::increaseVectorLength): Renamed from resizeStorage. Also
2593         simplified to only handle getting larger.
2594         (KJS::ArrayInstance::setLength): Added code to update m_numValuesInVector, to
2595         zero out the unused part of the vector and to delete the map if it's no longer
2596         needed.
2597         (KJS::ArrayInstance::mark): Tweaked formatting.
2598         (KJS::compareByStringForQSort): Ditto.
2599         (KJS::ArrayInstance::sort): Ditto.
2600         (KJS::CompareWithCompareFunctionArguments::CompareWithCompareFunctionArguments):
2601         Ditto.
2602         (KJS::compareWithCompareFunctionForQSort): Ditto.
2603         (KJS::ArrayInstance::compactForSorting): Fixed bug where this would turn
2604         undefined values into nonexistent values in some cases.
2605
2606         * kjs/array_object.h: Removed MAX_ARRAY_INDEX.
2607         * kjs/array_object.cpp: Removed ArrayInstance. Moved to a separate file.
2608
2609         * JavaScriptCore.pri: Added array_instance.cpp.
2610         * JavaScriptCore.xcodeproj/project.pbxproj: Ditto.
2611         * kjs/AllInOneFile.cpp: Ditto.
2612
2613 2007-10-22  Andrew Wellington  <proton@wiretapped.net>
2614
2615         Reviewed by Mark Rowe.
2616
2617         Fix for local database support after r26879
2618         Ensure that ENABLE_DATABASE and ENABLE_ICONDATABASE are correctly set
2619
2620         * Configurations/JavaScriptCore.xcconfig:
2621
2622 2007-10-22  Simon Hausmann  <hausmann@kde.org>
2623
2624         Reviewed by Alp.
2625
2626         Build fix for the non-qmake builds.
2627
2628         * wtf/Platform.h: Default to enabling the database features unless
2629         otherwise specified. (similar to ENABLE_ICONDATABASE)
2630
2631 2007-10-22  Holger Freyther  <zecke@selfish.org>
2632
2633         Reviewed by Simon Hausmann <hausmann@kde.org>.
2634
2635         * Do not build testkjs as an application bundle. This is
2636         needed for run-javascriptcore-tests on OSX.
2637         * Also, based on r26633, allow to test the WebKit/Qt port on OSX.
2638         * Set DYLD_LIBRARY_PATH if it was set in the environment. It must be set
2639         as we do not have -rpath on OSX.
2640
2641         * kjs/testkjs.pro:
2642
2643 2007-10-21  Mark Rowe  <mrowe@apple.com>
2644
2645         Reviewed by Alp.
2646
2647         http://bugs.webkit.org/show_bug.cgi?id=15575
2648         Bug 15575: [GTK] Implement threading using GThread
2649
2650         * wtf/Platform.h: Do not enable pthreads for Gtk.
2651
2652 2007-10-21  Mark Rowe  <mrowe@apple.com>
2653
2654         Reviewed by Mitz.
2655
2656         Fix http://bugs.webkit.org/show_bug.cgi?id=15603
2657         Bug 15603: Regression(r26847): Crash when sorting an empty array from JavaScript
2658
2659         * kjs/array_object.cpp:
2660         (KJS::freeStorage): Reinstate null-check that was removed in r26847.
2661
2662 2007-10-21  Darin Adler  <darin@apple.com>
2663
2664         - fix Windows build
2665
2666         * kjs/array_instance.h: Removed unused ExecState parameter.
2667         * kjs/array_object.cpp:
2668         (KJS::ArrayInstance::put): Ditto.
2669         (KJS::ArrayInstance::setLength): Ditto.
2670
2671 2007-10-21  Darin Adler  <darin@apple.com>
2672
2673         * kjs/array_object.cpp: (KJS::ArrayInstance::put):
2674         Add missing assignment that was causing regression test crash.
2675
2676 2007-10-21  Darin Adler  <darin@apple.com>
2677
2678         Reviewed by Maciej.
2679
2680         - http://bugs.webkit.org/show_bug.cgi?id=15585
2681           speed up sparse arrays by using a custom map
2682
2683         Speeds up SunSpider by 10%.
2684
2685         * kjs/array_object.cpp:
2686         (allocateStorage): Leave room for an additional pointer.
2687         (reallocateStorage): Ditto.
2688         (freeStorage): Ditto.
2689         (ArrayInstance::~ArrayInstance): Delete the overflow map if present.
2690         (ArrayInstance::getItem): Read values from the overflow map if present.
2691         Removed the check of length, since it slows down the common case.
2692         (ArrayInstance::getOwnPropertySlot): Ditto. Also removed the fallback
2693         to the property map.
2694         (ArrayInstance::put): Write values into the overflow map as needed.
2695         Also create overflow map when needed.
2696         (ArrayInstance::deleteProperty): Remove values from the overflow map
2697         as appropriate.
2698         (ArrayInstance::getPropertyNames): Add a name for each identifier in
2699         the property map. This is extremely inefficient.
2700         (ArrayInstance::setLength): Remove any values in the overflow map
2701         that are past the new length, as we formerly did with the property map.
2702         (ArrayInstance::mark): Mark any values in the overflow map.
2703         (compareByStringForQSort): Removed unneeded undefined case, since
2704         compactForSorting guarantees we will have no undefined values.
2705         (compareWithCompareFunctionForQSort): Ditto.
2706         (ArrayInstance::compactForSorting): Copy all the values out of the
2707         overflow map and destroy it.
2708
2709         * kjs/property_map.h: Removed now-unused getSparseArrayPropertyNames.
2710         * kjs/property_map.cpp: Ditto.
2711
2712 2007-10-20  Darin Adler  <darin@apple.com>
2713
2714         Reviewed by Maciej.
2715
2716         - http://bugs.webkit.org/show_bug.cgi?id=15579
2717           stop churning identifier reference counts copying Completion objects
2718
2719         * kjs/completion.h: Replace the Identifier with an Identifier*.
2720         * kjs/nodes.cpp:
2721         (ForInNode::execute): Update for change to Completion constructor.
2722         (ContinueNode::execute): Ditto.
2723         (BreakNode::execute): Ditto.
2724
2725 2007-10-20  Mark Rowe  <mrowe@apple.com>
2726
2727         Reviewed by Alp.
2728
2729         Gtk changes needed to enable HTML 5 client-side database storage.
2730
2731         * wtf/Platform.h: Have Gtk use pthreads for now.
2732
2733 2007-10-20  Geoffrey Garen  <ggaren@apple.com>
2734
2735         Reviewed by Maciej Stachowiak.
2736
2737         Fixed http://bugs.webkit.org/show_bug.cgi?id=15570
2738         Store gathered declaration nodes in the function body node.
2739
2740         This means that you only have to gather the declaration nodes the first
2741         time the function executes. Performance gain of 2.10% on SunSpider,
2742         0.90% on command-line JS iBench.
2743
2744         * kjs/nodes.cpp: Split declaration stack initialization code off into
2745         initializeDeclarationStacks().
2746         (FunctionBodyNode::FunctionBodyNode):
2747         (FunctionBodyNode::initializeDeclarationStacks):
2748         (FunctionBodyNode::processDeclarations):
2749
2750         * kjs/nodes.h: Changed DeclarationStacks structure to hold references,
2751         since the actual Vectors are now stored either on the stack or in the
2752         function body node.
2753
2754 2007-10-19  Geoffrey Garen  <ggaren@apple.com>
2755
2756         Reviewed by Darin Adler.
2757
2758         http://bugs.webkit.org/show_bug.cgi?id=15559
2759         Moved processDeclarations call into FunctionBodyNode::execute
2760
2761         To improve encapsulation, moved processDeclarations call into
2762         FunctionBodyNode::execute. Also marked processDeclarations
2763         ALWAYS_INLINE, since it has only 1 caller now. This is a .71% speedup
2764         on command-line JS iBench.
2765
2766         * kjs/function.cpp:
2767         (KJS::FunctionImp::callAsFunction):
2768         (KJS::GlobalFuncImp::callAsFunction):
2769         * kjs/function.h:
2770         * kjs/interpreter.cpp:
2771         (KJS::Interpreter::evaluate):
2772         * kjs/nodes.cpp:
2773         (FunctionBodyNode::execute):
2774         * kjs/nodes.h:
2775
2776 2007-10-19  Brady Eidson  <beidson@apple.com>
2777
2778         Reviewed by Sam
2779
2780         Queue -> Deque! and small style tweaks
2781
2782         * JavaScriptCore.vcproj/WTF/WTF.vcproj:
2783         * JavaScriptCore/JavaScriptCore.xcodeproj/project.pbxproj
2784         * wtf/Deque.h: Added.
2785         (WTF::DequeNode::DequeNode):
2786         (WTF::Deque::Deque):
2787         (WTF::Deque::~Deque):
2788         (WTF::Deque::size):
2789         (WTF::Deque::isEmpty):
2790         (WTF::Deque::append):
2791         (WTF::Deque::prepend):
2792         (WTF::Deque::first):
2793         (WTF::Deque::last):
2794         (WTF::Deque::removeFirst):
2795         (WTF::Deque::clear):
2796         * wtf/Queue.h: Removed.
2797
2798
2799 2007-10-19  Brady Eidson <beidson@apple.com>
2800
2801         Reviewed by Oliver
2802
2803         Added a simple LinkedList based Queue to wtf
2804         We can make a better, more sophisticated an efficient one later, but have
2805         needed one for some time, now!
2806
2807         * JavaScriptCore.xcodeproj/project.pbxproj:
2808         * wtf/Queue.h: Added.
2809         (WTF::QueueNode::QueueNode):
2810         (WTF::Queue::Queue):
2811         (WTF::Queue::~Queue):
2812         (WTF::Queue::size):
2813         (WTF::Queue::isEmpty):
2814         (WTF::Queue::append):
2815         (WTF::Queue::prepend):
2816         (WTF::Queue::first):
2817         (WTF::Queue::last):
2818         (WTF::Queue::removeFirst):
2819         (WTF::Queue::clear):
2820
2821 2007-10-19  Nikolas Zimmermann  <zimmermann@kde.org>
2822
2823         Reviewed by Anders.
2824
2825         Try to fix Qt/Win build slave, by including windows.h also on Qt/Win.
2826
2827         * kjs/testkjs.cpp: Change PLATFORM(WIN) to PLATFORM(WIN_OS)
2828
2829 2007-10-19  Simon Hausmann  <hausmann@kde.org>
2830
2831         Reviewed by Lars.
2832
2833         Fix compilation on Windows when wchar_t is a typedef instead of a native type (triggered by -Zc:wchar_t-).
2834         Don't provide the wchar_t overloads then as they conflict with the unsigned short ones.
2835
2836         * wtf/ASCIICType.h:
2837         (WTF::isASCIIAlpha):
2838         (WTF::isASCIIAlphanumeric):
2839         (WTF::isASCIIDigit):
2840         (WTF::isASCIIHexDigit):
2841         (WTF::isASCIILower):
2842         (WTF::isASCIISpace):
2843         (WTF::toASCIILower):
2844         (WTF::toASCIIUpper):
2845
2846 2007-10-19  Simon Hausmann  <hausmann@kde.org>
2847
2848         Reviewed by Lars.
2849
2850         Another build fix for the windows/qt build: Apply the same fix as in revision 26686 also to kjs/config.h to disable the disallowctype feature.
2851
2852         * kjs/config.h:
2853
2854 2007-10-18  Maciej Stachowiak  <mjs@apple.com>
2855
2856         Reviewed by Adam.
2857
2858         - use __declspec(thread) for fast thread-local storage on Windows
2859
2860         - 2.2% speedup on sunspider (on Windows)
2861         - 7% speedup on the string section
2862         - 6% speedup on JS iBench
2863
2864         - fixed <rdar://problem/5473084> PLT on Windows got 2.5% slower between r25406 and r25422
2865         - fixed at least some of <rdar://5527965? i-Bench JS was 14% slower in 310A11 than 310A10
2866
2867
2868         * wtf/FastMalloc.cpp:
2869         (WTF::getThreadHeap):
2870         (WTF::setThreadHeap):
2871         (WTF::TCMalloc_ThreadCache::GetCache):
2872         (WTF::TCMalloc_ThreadCache::GetCacheIfPresent):
2873         (WTF::TCMalloc_ThreadCache::CreateCacheIfNecessary):
2874
2875 2007-10-17  Darin Adler  <darin@apple.com>
2876
2877         Reviewed by Mark Rowe.
2878
2879         - fix http://bugs.webkit.org/show_bug.cgi?id=15543
2880           <rdar://problem/5545639> REGRESSION (r26697):
2881           GoogleDocs: Can't create new documents or open existing ones
2882
2883         Test: fast/js/regexp-non-character.html
2884
2885         * pcre/pcre_compile.c: (check_escape): Take out the checks for valid characters
2886         in the \u sequences -- not needed and actively harmful.
2887
2888 2007-10-17  Anders Carlsson  <andersca@apple.com>
2889
2890         Reviewed by Oliver.
2891
2892         * wtf/Platform.h:
2893         #define USE_PTHREADS on Mac.
2894
2895 2007-10-17  Geoffrey Garen  <ggaren@apple.com>
2896
2897         Reviewed by Darin Adler.
2898
2899         Merged DeclaredFunctionImp into FunctionImp (the base class) because
2900         the distinction between the two was unused.
2901
2902         Removed codeType() from FunctionImp because FunctionImp and its
2903         subclasses all returned FunctionCode, so it was unused, practically
2904         speaking.
2905
2906         Removed a different codeType() from GlobalFuncImp because it was unused.
2907         (Perhaps it was vestigial from a time when GlobalFuncImp used to
2908         inherit from FunctionImp.)
2909
2910         * bindings/runtime_method.cpp:
2911         * bindings/runtime_method.h:
2912         * kjs/function.cpp:
2913         (KJS::):
2914         (KJS::FunctionImp::FunctionImp):
2915         (KJS::FunctionImp::callAsFunction):
2916         (KJS::FunctionImp::construct):
2917         (KJS::FunctionImp::execute):
2918         (KJS::FunctionImp::processVarDecls):
2919         * kjs/function.h:
2920         (KJS::FunctionImp::implementsConstruct):
2921         (KJS::FunctionImp::scope):
2922         * kjs/function_object.cpp:
2923         (FunctionProtoFunc::callAsFunction):
2924         (FunctionObjectImp::construct):
2925         * kjs/nodes.cpp:
2926         (FuncDeclNode::processFuncDecl):
2927         (FuncExprNode::evaluate):
2928
2929 2007-10-17  Adam Roben  <aroben@apple.com>
2930
2931         Windows build fix part 2.
2932
2933         Fix was by Darin, reviewed by Anders and Adam.
2934
2935         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj: Add
2936         FastMallocPCRE.cpp to the project, and let Visual Studio have its way
2937         with the post-build step.
2938         * pcre/pcre.h: Don't DLL export the entry points just because this
2939         is Win32 -- this is an internal copy of PCRE and should be private.
2940         * pcre/pcre_compile.c: Fix an uninitialized variable warning --
2941         there's no real problem but it's better to quiet the compiler by
2942         tweaking the code slightly than turn off the warning entirely.
2943
2944 2007-10-17  Adam Roben  <aroben@apple.com>
2945
2946         Windows build fix.
2947
2948         Reviewed by Anders.
2949
2950         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj: Disable
2951         some mismatched signed/unsigned comparison warnings.
2952         * pcre/pcre_exec.c:
2953         (match): #if-out some labels that don't seem to exist.
2954
2955 2007-10-17  Mark Rowe  <mrowe@apple.com>
2956
2957         Gtk build fix.
2958
2959         * JavaScriptCore.pri: Add FastMallocPCRE.cpp.
2960         * pcre/pcre_get. #if out two functions that depend on pcre_get_stringnumber, which
2961         is currently unavailable for UTF-16.
2962
2963 2007-10-16  Darin Adler  <darin@apple.com>
2964
2965         Reviewed by Geoff.
2966
2967         - merged PCRE changes between 6.4 and 6.5
2968
2969         * JavaScriptCore.vcproj/JavaScriptCore/JavaScriptCore.vcproj:
2970         * JavaScriptCore.xcodeproj/project.pbxproj:
2971         Removed pcre_config.c, pcre_globals.c, pcre_info.c, pcre_maketables.c,
2972         pcre_printint.src, pcre_refcount.c, pcre_study.c, pcre_try_flipped.c,
2973         pcre_ucp_findchar.c, pcre_version.c, and ucptable.c. Added pcre_ucp_searchfuncs.c.
2974
2975         * pcre/AUTHORS:
2976         * pcre/LICENCE:
2977         * pcre/MERGING:
2978         * pcre/dftables.c:
2979         * pcre/pcre-config.h:
2980         * pcre/pcre.h:
2981         * pcre/pcre.pri:
2982         * pcre/pcre_compile.c:
2983         * pcre/pcre_exec.c:
2984         * pcre/pcre_fullinfo.c:
2985         * pcre/pcre_get.c:
2986         * pcre/pcre_internal.h:
2987         * pcre/pcre_maketables.c:
2988         * pcre/pcre_ord2utf8.c:
2989         * pcre/pcre_tables.c:
2990         * pcre/pcre_ucp_searchfuncs.c: Copied from pcre/pcre_ucp_findchar.c.
2991         * pcre/pcre_xclass.c:
2992         * pcre/ucp.h:
2993         * pcre/ucpinternal.h:
2994         * pcre/ucptable.c:
2995         Updated with new versions from the PCRE 6.5 release, merged with changes.
2996
2997         * pcre/pcre_config.c: Removed.
2998         * pcre/pcre_globals.c: Removed.
2999         * pcre/pcre_info.c: Removed.
3000         * pcre/pcre_printint.src: Removed.
3001         * pcre/pcre_refcount.c: Removed.
3002         * pcre/pcre_study.c: Removed.
3003         * pcre/pcre_try_flipped.c: Removed.
3004         * pcre/pcre_ucp_findchar.c: Removed.
3005         * pcre/pcre_version.c: Removed.
3006
3007 2007-10-16  Geoffrey Garen  <ggaren@apple.com>
3008
3009         Reviewed by Darin Adler.
3010
3011         Removed KJS_VERBOSE because it was getting in the way of readability,
3012         and the messages didn't seem very helpful.
3013
3014         * kjs/function.cpp:
3015         (KJS::FunctionImp::callAsFunction):
3016         (KJS::FunctionImp::passInParameters):
3017         * kjs/lookup.h:
3018         (KJS::lookupPut):
3019         * kjs/object.cpp:
3020         (KJS::JSObject::put):
3021         * kjs/value.h:
3022
3023 2007-10-16  Geoffrey Garen  <ggaren@apple.com>
3024
3025         Reviewed by Darin Adler.
3026
3027         Removed the Parameter class because it was a redundant wrapper around
3028         Identifier.
3029
3030         * kjs/function.cpp:
3031         (KJS::FunctionImp::passInParameters):
3032         (KJS::FunctionImp::getParameterName):
3033         * kjs/nodes.cpp:
3034         (FunctionBodyNode::addParam):
3035         * kjs/nodes.h:
3036         (KJS::FunctionBodyNode::):
3037
3038 2007-10-16  Geoffrey Garen  <ggaren@apple.com>
3039
3040         Reviewed by Darin Adler.
3041
3042         Global replace of assert with ASSERT.
3043
3044 2007-10-16  Adam Roben  <aroben@apple.com>
3045
3046         Make testkjs not delay-load WebKit
3047
3048         Soon, delay-loading WebKit will be impossible (because we will be
3049         using __declspec(thread) for thread-local storage). This change
3050         prepares testkjs for the future.
3051
3052         Reviewed by Sam.
3053
3054         * JavaScriptCore.vcproj/JavaScriptCore.sln: Removed WebKitInitializer,
3055         added FindSafari.
3056         * JavaScriptCore.vcproj/testkjs/testkjs.vcproj: Don't link against
3057         WebKitInitializer, don't delay-load WebKit.
3058         * kjs/testkjs.cpp: Don't use WebKitInitializer.
3059
3060 2007-10-16  Adam Roben  <aroben@apple.com>
3061
3062         Updated testkjs for the rename of WebKit_debug.dll to WebKit.dll for the Debug configuration
3063
3064         Reviewed by Kevin McCullough.
3065
3066         * JavaScriptCore.vcproj/debug.vsprops: Added WebKitDLLConfigSuffix.
3067         * JavaScriptCore.vcproj/debug_internal.vsprops: Ditto.
3068         * JavaScriptCore.vcproj/release.vsprops: Ditto.
3069         * JavaScriptCore.vcproj/testkjs/testkjs.vcproj: Use
3070         WebKitDLLConfigSuffix when referring to WebKit.dll, and fixed a typo
3071         in the name of icuuc36[_debug].dll.
3072
3073 2007-10-16  Geoffrey Garen  <ggaren@apple.com>
3074
3075         Reviewed by Maciej Stachowiak.
3076
3077         Re-structured variable and function declaration code.
3078
3079         Command-line JS iBench shows no regression.
3080
3081         Here are the changes:
3082
3083         1. Function declarations are now processed at the same time as var
3084         declarations -- namely, immediately upon entry to an execution context.
3085         This does not match Firefox, which waits to process a function
3086         declaration until the declaration's containing block executes, but it
3087         does match IE and the ECMA spec. (10.1.3 states that var and function
3088         declarations should be processed at the same time -- namely, "On
3089         entering an execution context." 12.2 states that "A Block does not
3090         define a new execution scope.")
3091
3092         2. Declaration processing proceeds iteratively now, rather than
3093         recursively, storing the nodes is finds in stacks. This will later
3094         facilitate an optimization to hold on to the gathered declaration nodes,
3095         rather than re-fetching them in every function call.
3096         [ http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3097
3098         Modified these tests because they expected the incorrect Mozilla
3099         behavior described above:
3100
3101         * tests/mozilla/ecma_3/Function/scope-001.js:
3102         * tests/mozilla/js1_5/Scope/regress-184107.js:
3103
3104 2007-10-16  Darin Adler  <darin@apple.com>
3105
3106         - try to fix the GTK build
3107
3108         * kjs/ustring.cpp: Include ASCIICType.h, not ASCIICtype.h.
3109
3110 2007-10-16  Darin Adler  <darin@apple.com>
3111
3112         - try to fix the Windows build
3113
3114         * kjs/date_object.cpp: (KJS::parseDate): A couple instances of isspace were
3115         in here. Not sure why it wasn't failing elsewhere. Changed to isASCIISpace.
3116
3117 2007-10-16  Darin Adler  <darin@apple.com>
3118
3119         - try to fix the GTK build
3120
3121         * kjs/ustring.cpp: Include ASCIICType.h.
3122
3123 2007-10-16  Darin Adler  <darin@apple.com>
3124
3125         Reviewed by Maciej and Geoff (and looked over by Eric).
3126
3127         - http://bugs.webkit.org/show_bug.cgi?id=15519
3128           eliminate use of <ctype.h> for processing ASCII
3129
3130         * wtf/ASCIICType.h: Added.
3131         * wtf/DisallowCType.h: Added.
3132
3133         * kjs/config.h: Include DisallowCType.h.
3134
3135         * kjs/date_object.cpp:
3136         (KJS::skipSpacesAndComments):
3137         (KJS::findMonth):
3138         (KJS::parseDate):
3139         * kjs/function.cpp:
3140         (KJS::decode):
3141         * kjs/ustring.cpp:
3142         (KJS::UString::toDouble):
3143         Use ASCIICType.h functions instead of ctype.h ones.
3144
3145 2007-10-14  Maciej Stachowiak  <mjs@apple.com>
3146
3147         Reviewed by Darin.
3148
3149         - fixes for "New JavaScript benchmark"
3150         http://bugs.webkit.org/show_bug.cgi?id=15515
3151
3152         * kjs/testkjs.cpp:
3153         (TestFunctionImp::callAsFunction): Implement "load" for compatibility
3154         with SpiderMonkey.
3155         (TestFunctionImp::): ditto
3156         (doIt): ditto
3157         (kjsmain): Drop useless --> from output.
3158
3159 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3160
3161         Removed unnecessary #include.
3162
3163         * API/JSObjectRef.cpp:
3164
3165 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3166
3167         Double-reverse build fix. My tree was out of date.
3168
3169         * kjs/nodes.cpp:
3170         (NumberNode::evaluate):
3171
3172 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3173
3174         Build fix.
3175
3176         * kjs/nodes.cpp:
3177         (NumberNode::evaluate):
3178
3179 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3180
3181         Reviewed by Darin Adler.
3182
3183         Removed surprising self-named "hack" that made nested functions
3184         available as named properties of their containing functions, and placed
3185         containing function objects in the scope chains of nested functions.
3186
3187         There were a few reasons to remove this "hack:"
3188
3189         1. It contradicted FF, IE, and the ECMA spec.
3190
3191         2. It incurred a performance penalty, since merely parsing a function
3192         required parsing its body for nested functions (and so on).
3193
3194         3. SVN history contains no explanation for why it was added. It was just
3195         legacy code in a large merge a long, long time ago.
3196
3197         [ Patch broken off from http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3198
3199         * kjs/nodes.cpp:
3200         (FuncDeclNode::processFuncDecl):
3201
3202 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3203
3204         Reviewed by Darin Adler.
3205
3206         Removed the concept of AnonymousCode. It was unused, and it doesn't
3207         exist in the ECMA spec.
3208
3209         [ Patch broken off from http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3210
3211         * kjs/Context.cpp:
3212         (KJS::Context::Context):
3213         * kjs/function.h:
3214         (KJS::):
3215         * kjs/nodes.cpp:
3216         (ReturnNode::execute):
3217
3218 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3219
3220         Reviewed by Darin Adler.
3221
3222         Made function parameters DontDelete. This matches FF and the vague
3223         description in ECMA 10.1.3. It's also required in order to make
3224         symbol table based lookup of function parameters valid. (If the
3225         parameters aren't DontDelete, you can't guarantee that you'll find
3226         them later in the symbol table.)
3227
3228         [ Patch broken off from http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3229
3230         * kjs/function.cpp:
3231         (KJS::FunctionImp::passInParameters):
3232
3233 2007-10-15  Geoffrey Garen  <ggaren@apple.com>
3234
3235         Reviewed by Maciej Stachowiak.
3236
3237         Some Vector optimizations. These are especially important when using
3238         Vector as a stack for implementing recursive algorithms iteratively.
3239
3240         [ Broken off from http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3241
3242         1. Added shrink(), which is a version of resize() that you can call
3243         to save a branch / improve code generation and inlining when you know
3244         that the vector is not getting bigger.
3245
3246         2. Changed subclassing relationship in VectorBuffer to remove a call to
3247         fastFree() in the destructor for the inlineCapacity != 0 template
3248         specialization. This brings inline Vectors one step closer to true
3249         stack-allocated arrays.
3250
3251         Also changed abort() to CRASH(), since the latter works better.
3252
3253         * wtf/Vector.h:
3254         (WTF::VectorBufferBase::allocateBuffer):
3255         (WTF::VectorBufferBase::deallocateBuffer):
3256         (WTF::VectorBufferBase::VectorBufferBase):
3257         (WTF::VectorBufferBase::~VectorBufferBase):
3258         (WTF::):
3259         (WTF::VectorBuffer::VectorBuffer):
3260         (WTF::VectorBuffer::~VectorBuffer):
3261         (WTF::VectorBuffer::deallocateBuffer):
3262         (WTF::VectorBuffer::releaseBuffer):
3263         (WTF::Vector::clear):
3264         (WTF::Vector::removeLast):
3265         (WTF::::operator):
3266         (WTF::::fill):
3267         (WTF::::shrink):
3268
3269 2007-10-12  Geoffrey Garen  <ggaren@apple.com>
3270
3271         Reviewed by Maciej Stachowiak.
3272
3273         Fixed http://bugs.webkit.org/show_bug.cgi?id=15490
3274         Iteration statements sometimes incorrectly evaluate to the empty value
3275         (KDE r670547).
3276
3277         [ Broken off from http://bugs.webkit.org/show_bug.cgi?id=14868 ]
3278
3279         This patch is a merge of KDE r670547, with substantial modification
3280         for performance.
3281
3282         It fixes do-while statements to evaluate to a value. (They used
3283         to evaluate to the empty value in all cases.)
3284
3285         It also fixes SourceElementsNode to maintain the value of abnormal
3286         completions like "break" and "continue."
3287
3288         It also re-works the main execution loop in SourceElementsNode so that
3289         it (1) makes a little more sense and (2) avoids unnecessary work. This
3290         is a .28% speedup on command-line JS iBench.
3291
3292         * kjs/nodes.cpp:
3293         (DoWhileNode::execute):
3294         (SourceElementsNode::execute):
3295
3296 2007-10-15  Simon Hausmann  <hausmann@kde.org>
3297
3298         Reviewed by Lars.
3299
3300         Fix compilation with gcc 4.3 by including 'limits' due to the use of std::numeric_limits.
3301
3302         * wtf/HashTraits.h:
3303
3304 2007-10-5  Kevin Ollivier  <kevino@theolliviers.com>
3305
3306         Reviewed by Adam.
3307
3308         Add support for MSVC7, and fix cases where PLATFORM(WIN) should
3309         be PLATFORM(WIN_OS) for other ports building on Windows.
3310
3311         * kjs/DateMath.cpp:
3312         (KJS::getDSTOffsetSimple):
3313         * kjs/JSImmediate.h:
3314         * wtf/Assertions.cpp:
3315         * wtf/Assertions.h:
3316         * wtf/Platform.h:
3317         * wtf/StringExtras.h:
3318         (snprintf):
3319         (vsnprintf):
3320
3321 2007-10-14  Cameron Zwarich  <cwzwarich@uwaterloo.ca>
3322
3323         Reviewed by Darin.
3324
3325         Adds NegateNode optimization from KJS. The relevant revision in KDE
3326         is 666736.
3327
3328         * kjs/grammar.y:
3329         * kjs/nodes.cpp:
3330         (NumberNode::evaluate):
3331         * kjs/nodes.h:
3332         (KJS::Node::):
3333         (KJS::NumberNode::):
3334         * kjs/nodes2string.cpp:
3335         (NumberNode::streamTo):
3336
3337 2007-10-14  Jason Foreman  <jason@threeve.org>
3338
3339         Reviewed by Maciej.
3340
3341         Fix http://bugs.webkit.org/show_bug.cgi?id=15145
3342
3343         Ensure that if adjusting n to minimize the difference of n*intPow10(e-p+1) to x,
3344         that the property n < intPow10(p) is maintained.
3345
3346         * kjs/number_object.cpp:
3347         (NumberProtoFunc::callAsFunction):
3348
3349 == Rolled over to ChangeLog-2007-10-14 ==