Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 359857

Summary:	boxing happening when it should not, and not when it should
Product:	z_Archived	Reporter:	Jeff Douglas <jeffdouglas>
Component:	EDT	Assignee:	Huang Ji Yong <hjiyong>
Status:	CLOSED FIXED	QA Contact:
Severity:	major
Priority:	P1	CC:	broy2, carrollk, jqian, mayunf, pharmon, svihovec, tww
Version:	unspecified
Target Milestone:	---
Hardware:	PC
OS:	Windows XP
Whiteboard:

Description Jeff Douglas

2011-10-04 11:23:48 EDT

Paul.

Tim and I discussed this and it seems that boxing is occurring when it should not be and is not occurring when it should.

Take this example:

package acme.driver;

program bug10
    myVar decimal(18,6);    
    results int;
   
    function main()
    	results = 1 - 1 + 1;
    	results = 1 / 1 + 1;
    	results = 1 * 1 + 1;
        results = MathLib.decimals(myVar);
    end
end

In the case of 1 / 1 + 1, there is a boxing occurring (likely caused by the EGL definition file EInt having the return of EDecimal for the $divide method). It might be okay to do this, but it causes a lot of execution code to be created.
Tim thinks this is unnecessary.

For the Mathlib.decimals function. IRUtils makeExprCompatibleToType method is passing this, here:

		if (exprType.equals(type) || exprType.getClassifier().equals(type)) {
			return expr;
		}

exprType is FixedPrecisionType and exprType's classifier is ParameterizableType, which matches the EGL definition for mathlib.decimal, as it takes decimal in as the arg. The problem is that it needs to do a boxing, and caused the problem in defect 359745

Can you please talk with Tim and Matt and see what needs to be done? Thanks

Comment 1 Jeff Douglas

2011-10-04 11:24:16 EDT

*** Bug 359745 has been marked as a duplicate of this bug. ***

Comment 2 Paul Harmon

2011-10-12 18:11:11 EDT

I looked into the first problem and was able to check in a fix in IRUtils to remove the excess boxing when converting value types to the Reference types of Number, Decimal and Timestamp.


However:

I am routing this defect back to javaGen because of the second problem (Mathlib.decimals() not working). We CANNOT do a box for operations that involve things like Decimal(10,3) to Decimal. The reason for this is that references to the Decimal (reference) type is generated as BigDecimal, not EDecimal. If makeExprCompatibleWithType made a boxing expression for this, it would break javaGen (and probably JS gen). Since things are working with Decimal in other regards, I think the generators will need to have special code to box the argument to the Mathlib.decimals() function.

Assuming you do add special code to java gen to handle this system function, please route the bug to JS gen when complete so that a similar change can be made to that generator

Comment 3 Jeff Douglas

2011-10-28 06:01:06 EDT

I've implemented code in reorg code to check for mathlib.decimals and mathlib.precision to box the argument if of type decimal, so that the values are preserved.

Routing this to JSgen, to determine if they wish this same logic activated.

Comment 4 Huang Ji Yong

2011-11-17 20:21:11 EST

fixed already

Comment 5 Jeff Douglas

2011-11-18 11:58:20 EST

ok