Added benchmarks for Johan's LRP Parser

Portability: fixed RBLiteralValueNode>>isLiteralNumber to work under both - Pharo and Smalltalk/X

Portability fixes * do not use Object>>asString. Not all Smalltalks implement it. * do not use Object>>name. Not all Smalltalks implement it. * do not use Dictionary keysAndValuesRemove:. Not all Smalltalks implement it. * do not use Class>>methods The semantics is different among Smalltalks. Use `Class methodDictionary values` instead. * do not modify dictionary in #at:ifAbsentPut: block!

Portability: override #new for PPCScannerCodeGenerator. Not all smalltalks send #initialize by default.

Merge

Updated to PetitCompiler-JanKurs.160, PetitCompiler-Tests-JanKurs.112, PetitCompiler-Extras-Tests-JanKurs.25, PetitCompiler-Benchmarks-JanKurs.17 Name: PetitCompiler-JanKurs.160 Author: JanKurs Time: 17-08-2015, 09:52:26.291 AM UUID: 3b4bfc98-8098-4951-af83-a59e2585b121 Name: PetitCompiler-Tests-JanKurs.112 Author: JanKurs Time: 16-08-2015, 05:00:32.936 PM UUID: 85613d47-08f3-406f-9823-9cdab451e805 Name: PetitCompiler-Extras-Tests-JanKurs.25 Author: JanKurs Time: 16-08-2015, 05:00:10.328 PM UUID: 09731810-51a1-4151-8d3a-56b636fbd1f7 Name: PetitCompiler-Benchmarks-JanKurs.17 Author: JanKurs Time: 05-08-2015, 05:29:32.407 PM UUID: e544b5f1-bcf8-470b-93a6-d2363e4dfc8a

Fixed PPCSetUpBefore...Resource to work on Pharo. Few othr minor Pharo fixes.

Portability fix: do not use "/ style comments in PetitParser code.

Added smoke tests for LRPParser

Added LRPParser from Live Robots Prograaming project

Tests refactoring - use generated test cases to make sure all posibilities are tested. Do not generate resource for all combinations, use PPCSetUpBeforeTearDownAfterResource instead that delegates parser compilation to the testcase itself (it calls it's #setUpBefore method).

Inline child of an action node only is its unnamed sequence node. Named nodes should not be inlined as they should make a method. There's little point in inlining non-sequence nodes, so don't enforce inlining on those. Some (JK :-) may prefer them non-inlined (for debugging purposes)

Temporarily disabled #acceptsEpsilon test in PPCTokenizingCodeGenerator>>visitChoiceNode: JK should have a look and come up with proper solution.

Workaround for stc bug.

Added static analysis of blocks when inlining. Allow inlining only when block is functional ...i.e., does not access any shared state (in instance or class variables). If the block does a self-send, the sent method has to be (transitively) functional too. To allow for self-sends in action blocks, copy (transitively) self-sent methods to target parser. This is safe as these self-sent methods are guarnateed to be functional.

Post-merge fixes. PPCCodeBlock refactoring.

Portability fix: override #new for class that implements #initialize #initialize is not sent by default.

Merge

Updated to PetitCompiler-JanVrany.135, PetitCompiler-Tests-JanKurs.93, PetitCompiler-Extras-Tests-JanVrany.16, PetitCompiler-Benchmarks-JanKurs.12 Name: PetitCompiler-JanVrany.135 Author: JanVrany Time: 22-07-2015, 06:53:29.127 PM UUID: 890178b5-275d-46af-a2ad-1738998f07cb Ancestors: PetitCompiler-JanVrany.134 Name: PetitCompiler-Tests-JanKurs.93 Author: JanKurs Time: 20-07-2015, 11:30:10.283 PM UUID: 6473e671-ad70-42ca-b6c3-654b78edc531 Ancestors: PetitCompiler-Tests-JanKurs.92 Name: PetitCompiler-Extras-Tests-JanVrany.16 Author: JanVrany Time: 22-07-2015, 05:18:22.387 PM UUID: 8f6f9129-dbba-49b1-9402-038470742f98 Ancestors: PetitCompiler-Extras-Tests-JanKurs.15 Name: PetitCompiler-Benchmarks-JanKurs.12 Author: JanKurs Time: 06-07-2015, 02:10:06.901 PM UUID: cb24f1ac-46a4-494d-9780-64576f0f0dba Ancestors: PetitCompiler-Benchmarks-JanKurs.11, PetitCompiler-Benchmarks-JanVrany.e29bd90f388e.20150619081300

Moved PPMappedActionParser>>asCompilerNode to protocol 'converting' If in '*petitparser' it makes some problems with Monticello.

Use RBVariableNode>>name: to change variable's name. The old way - `node token value:` is working in Pharo - there's no token in RBVariableNode. Sigh.

Portability fix: do not use #? ...use `notNil ifTrue:[] ifFalse:[]` instead. #? is not supported in Squeak/Pharo

Moved PPCMappedActionNode to category PetitCompiler-Nodes

Do not use RBProgramNode>>variableNodesDo: as this method is not present in Pharo. Use local helper method instead.

Reverted fc3dbe5654c5: sending #copy should be OK (copy set's parent properly)

Send #sourceNode instead of #ast as there's no #ast in Pharo ...I was mistaken.

Oops, removed EOL comments as they;re not supported in Squeak / Pharo

Fixed target mcz - exit Smalltalk after all packages are exported.

Use #deepCopy instead of #copy when copying RB parse tree ...because #copy make a copy if child nodes but does not set their parents properly. Therefore node replacing does not work because it replaces it in wrong tree (original, not the copy).

Codegen: added support for smart action node compiling. Avoid creation of intermediate result collection for action nodes if all references to action block's argument (i.e., the nodes collection) is in form of: * <nodes> at: <numeric constant> * <nodes> first (second, third...

Oops, renamed testMappedNode3/4/5 to testActionNode3/4/5 which is the correct name!

Fix in codegen for inlined sequence nodes. For inlined sequence nodes, generate nested ifs rather than sequential code which does not work when inlined. The reason is that #codeReturn: in inline generates instvar assignment, not method return, so in sequential code the next child of a sequence will be probed even if previous failed. If that happends, the whole sequence fail and therefore we must generate nested ifs to correctly handle this w.r.t. inlining.

Avoid creation of intermediate collection for mapped action nodes.

Fix in codegen: fixed compilation of unknown node: do not hardcode return variable name ...use `self retvalVar` instead

Renamed #codeStoreValueOf:intoVariable: to #codeAssignParsedValueOf:to: to use same naming as #codeAssign:to:

Fixed codegen for empty (mapped) action block. Although empty blocks are useless in practice, they may occur (and do occur in tests)

Do not create intermediate collection when parsing sequence if not necesary. The collection is not needed when the result of a choice is being used in mapped parser. In that case, store parsed objects in variables and inline action code to use these variables to access parsed objects.

Oops, fixed subproject definition (wrong package names)

Updated .mcz exporting code to include a commit message digest ...instead of just exported revision commit message.

Introduced PPCMappedActionNode for PPMappedActionParser. It's compilation is not yet ideal, bot works for now.

Introduced PPMappedActionParser, a specialized parser for map:ped actions. In order to compile nicer code, compiler has to be able to tell between simple (==>) action parser and mapped (map:) action parser.

Action inlining [1/x]: Initial support for inlining actions parsers (i.e., ==>) The code of the action is now inlined into parsing method rather then delegated to stored block. Mapping parser (i.e., map:[...]) are not supported and not detected, so using them cause crash. This will be fixed later.

Codegen refactoring [2/x]: Added PPCCompiler>>codeIf:then:else: this allows for more structured coding of conditionals. then/else code is evaluated inside new block so it may allocate its own temporaries.

Codegen refactoring [1/x]: Introduced a PPCCodeBlock A PPCCodeBlock is an abstraction of a block of statements with temporaries. This will allow for scoped temporary management in code generator - temporary variables could be allocated for block only. (i.e., make them block-temporaries)

Fixed choice inlining.

Add inspector tab to present parser tree.

More fixes for inlining of nodes

Oops, merged code which contained Pharoism's

Avoid using of a bridge parser for top-most composite parser ...by converting PPCompositeParser to PPCDelegatePrser.

Added two more tests for inlining

Added inspector representation to show PPC node tree.

Oops. re-added classes lost during merge...

Indent by 4 spaces on Smalltalk/X

Fixed code generation for deterministic choice w.r.t inlining For PPCDeterministicChoiceNode generate code in form if () else if () else if () else error instead of sequence of ifs. The former is safe w.r.t inlining.

Fixed MNU in PPCSpecializingVisitor>>visitTrimmingTokenNode: node Set child of PPCTrimmingCharacterTokenNode before setting name as child is used to compute meaningful name.

Merge

Updated to PetitCompiler-JanKurs.118, PetitCompiler-Tests-JanKurs.46, PetitCompiler-Extras-Tests-JanKurs.11, and PetitCompiler-Benchmarks-JanKurs.11 Name: PetitCompiler-JanKurs.118 Author: JanKurs Time: 13-05-2015, 03:59:01.292 PM UUID: 4a8ccd94-3131-4cc7-9098-528f8e5ea0b5 Name: PetitCompiler-Tests-JanKurs.46 Author: JanKurs Time: 04-05-2015, 04:25:06.162 PM UUID: 9f4cf8b7-876e-4a13-9579-b833f016db66 Name: PetitCompiler-Extras-Tests-JanKurs.11 Author: JanKurs Time: 13-05-2015, 04:27:27.940 PM UUID: e9f30c31-fbd0-4e96-ad2a-868f88d20ea8 Name: PetitCompiler-Benchmarks-JanKurs.11 Author: JanKurs Time: 13-05-2015, 02:21:49.932 PM UUID: 6a23fd1e-a86f-46db-8221-cc41b778d32c

Small improvement in inlining: inline child of an action node.

Added benchmark of a smalltalk parser with empty actions. This is used to check the cost of compiled actions

Portability: fixes for Smalltalk/X * Do not use #crShow: - not present in Smalltalk/X * Do not use Array class>>with:withAll: * do not use detect:ifFound:ifAbsent: