Adding locoGP benchmark problems

[codykenb]

Example sort algorithms and tests added.

[codykenb]

BubbleDouble, BubbleLoops, Heap, Quick, Merge, Radix have been run without exception. Cocktail throws exception.

[drdrwhite]

HuffmanCodeBook.java doesn't parse:

Exception in thread “main” com.github.javaparser.ParseProblemException: (line 83,col 1) private is not allowed here.
(line 98,col 1) ‘private’ is not allowed here.

    at com.github.javaparser.JavaParser.simplifiedParse(JavaParser.java:390)
    at com.github.javaparser.JavaParser.parse(JavaParser.java:261)
    at gin.SourceFile.<init>(SourceFile.java:23)
    at gin.LocalSearch.<init>(LocalSearch.java:47)
    at gin.LocalSearch.main(LocalSearch.java:33)'

[codykenb]

Sure, I've yet to write a test case for it also, might be best to leave it out of master branch until it's ready?

[drdrwhite]

OK, removed HuffmanCodeBook from the branch.

[drdrwhite]

We still have two issues to resolve:

• the exception from JavaParser when optimising SortCocktail.
• // TODO why are we going out-of-bounds here? <--- these comments

Regarding the exception, I have recreated and am investigating. I can't see why this is happening yet.

Could you expand on the out-of-bounds comments please? e.g. could you give an example of it failing?

[drdrwhite]

Minimal example to recreate exception:

java gin.PatchAnalyser examples/locoGP/SortCocktail.java "| MOVE 16 -> 2:0 |"

[codykenb]

Question about out-of-bounds is, should we check parent.getChildNodes().size() > copy.destinationChildInBlock when applying a patch (happens when only one child node, index is 1), or should we investigate why a patch is referencing an out of bounds index?

On 27 June 2017 at 14:15, drdrwhite notifications@github.com wrote:

We still have two issues to resolve:

• the exception from JavaParser when optimising SortCocktail.
• // TODO why are we going out-of-bounds here? <--- these comments

Regarding the exception, I have recreated and am investigating. I can't see why this is happening yet.

Could you expand on the out-of-bounds comments please? e.g. could you give an example of it failing?

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/drdrwhite/gin/pull/1#issuecomment-311354428, or mute the thread https://github.com/notifications/unsubscribe-auth/AF0t4wBdYK6eW8NJgNbRdSyFwC7xzdrAks5sIQBigaJpZM4OFYtU .

[drdrwhite]

A patch should never reference an out-of-bounds index, so there's a bug if that's happening. Perhaps that could explain the Exception problem, which is mystifying me so far.

DEL X = Delete statement X (you can see the way the statements are numbered by viewing the .statements output of PatchAnalyser)

MOVE or COPY X -> B:P = place X at Block B (you can see the block numbering via .blocks output from PatchAnalyser) and point P within the block.

"Point P" means "before statement P". There are special conditions to handle where there are no statements, i.e. the insertion is going into an empty block.

[drdrwhite]

So I can confirm that the bug is actually caused by the delete operation and not the insertion.

I suspect some sort of issue with the cloning of the original statement here.

[drdrwhite]

I think I've identified the problem. The .remove method in JavaParser's Node doesn't fully remove a node, and the AST becomes garbled.

Looking at a ModifierVisitor instead. This makes things much more complex that they should be from our POV. Not good.

https://github.com/javaparser/javaparser/wiki/Manual

[codykenb]

I was hoping javaparser would give us a simple way to modify ast's, strange...

[drdrwhite]

New minimal example: "DEL 16" is enough to cause the problem

[drdrwhite]

New absolutely minimal example:

public class SortCocktail {
  public static void example() {
    boolean swapped;
    swapped=false;
    swapped=false;
  }
}

with | DEL 3 |

[drdrwhite]

Bug is clearly related to the sample expression statement appear twice in a block.