Bug 256796

Summary:

[compiler] dead code detection: "if (DEBUG) return;" should also be trivial IF stmt

Product:

[Eclipse Project] JDT

Reporter:

Markus Keller <markus.kell.r>

Component:

Core

Assignee:

Ayushman Jain <amj87.iitr>

Status:

VERIFIED FIXED

QA Contact:

Severity:

enhancement

Priority:

CC:

daniel_megert, frederic_fusier, jerome_lanneluc, Olivier_Thomann, philippe_mulet, srikanth_sankaran

Version:

3.5

Target Milestone:

3.8 M3

Hardware:

OS:

Windows XP

Whiteboard:

Bug Depends on:

256463

Bug Blocks:

Attachments:

Description	Flags
proposed fix v1.0 + regression tests	none
patch for NegativeTest	none

Description Markus Keller

2008-11-27 10:53:01 EST

+++ This bug was initially created as a clone of Bug #256463 +++

HEAD after bug 256463

COMPILER_PB_DEAD_CODE_IN_TRIVIAL_IF_STATEMENT should also suppress dead code problems *after* the if statement, not only in the else branch:

public class X {
    private static final boolean DEBUG= true;

    void foo() {
        if (DEBUG)
            return;
        else
            System.out.println("hello"); // not dead (good)
        System.out.println("world"); // dead (but should not be)
    }
    
    void bar() {
        if (DEBUG) {
            return;
        }
        System.out.println("hello"); // dead (but should not be)
    }
}

Comment 1 Philipe Mulet

2008-11-28 04:48:58 EST

This feels a corner situation to me.

Comment 2 Philipe Mulet

2008-11-28 06:52:18 EST

Thinking more of this, I believe this is actually an evil pattern, which shouldn't exist... so I wouldn't treat it gracefully.

        if (DEBUG)
            return;

Comment 3 Jerome Lanneluc

2008-12-08 05:21:38 EST

Reopen to assign

Comment 4 Jerome Lanneluc

2008-12-08 05:22:36 EST

(In reply to comment #2)
> Thinking more of this, I believe this is actually an evil pattern, which
> shouldn't exist... so I wouldn't treat it gracefully.

Comment 5 Jerome Lanneluc

2008-12-09 07:05:25 EST

Verified for 3.5M4

Comment 6 Dani Megert

2011-03-15 10:17:22 EDT

I disagree with the resolution. The field can have an arbitrary name/meaning and only adding support for the case where we the initial value is 'false' seems wrong to me.

Currently this bug is the main reason for the dead code warnings in JDT UI.

Comment 7 Ayushman Jain

2011-09-29 14:28:53 EDT

Created attachment 204316 [details]
proposed fix v1.0 + regression tests

This patch makes sure that the merged info from the if and else branches is not unreachable because of a known dead code pattern used as condition, when the COMPILER_PB_DEAD_CODE_IN_TRIVIAL_IF_STATEMENT is disabled. 

If the else statement is empty or does not otherwise return, the reachable bit of the flow info prior to if-else is taken as the bit for the subsequent info, otherwise the reachability is taken from the else statement

Comment 8 Ayushman Jain

2011-09-30 05:18:09 EDT

Created attachment 204355 [details]
patch for NegativeTest

Second part of patch from CVS

Comment 9 Ayushman Jain

2011-09-30 05:22:10 EDT

Fixed in HEAD with commit 5333a8d6e234b4d5bbcbee365cfa39aafade1032. I don't see the dead code warnings in jdt.ui anymore.

Comment 10 Dani Megert

2011-10-03 08:58:44 EDT

Thanks!

Comment 11 Srikanth Sankaran

2011-10-25 02:44:47 EDT

Verified for 3.8 M3 using build id: N20111022-2000