469616 - tail call optimization will cause several testcase failed on sparc.

Status: VERIFIED WORKSFORME

(Tamarin Graveyard :: Virtual Machine, defect)

Description

[Leon Sha]

ecma3/Array/splice1.abc : exhaustive splice w/no optional args 1 = false FAILED! expected: true
ecma3/Array/splice2.abc : exhaustive splice w/2 optional args 1 = false FAILED! expected: true

The problem is caused by the tail call optimization for String::concatStrings.  After the optimization the arg rightStr is not on the stack. So there is a chance that rightStr is released while there are still other functions using its data. I believe this is a all platform bug.

Comment 1

[Leon Sha]

Attachment: Patch

I believe this is a all platform bug. So I removed the #ifdef part for win 64. Also I tried it on try server, everything looks OK. 
http://tamarin-builds.mozilla.org/sandbox/builders/windows64/builds/5
The thing I do is to put rightStr on stack and break the tail optimization. Also I need some input of the comments for this. Another solution would be addref for rightStr, but I don't know where to deref it.

Comment 2

[Steven Johnson]

Leon, we're concerned there's a more general problem here (ie, the use of interior pointers might bite us elsewhere as well)... I've added other folks to the cc list, and while your patch looks OK as-is I'm going to hold off review+ for now since a more sweeping fix might be called for.

Comment 3

[Leon Sha]

Ok, Thanks. If this general problem fixed, I'll close this bug.

Comment 4

[Lars T Hansen]

Tommy says it's a GC issue that is also biting hard on https://bugzilla.mozilla.org/show_bug.cgi?id=465506.  Probably this bug is a dup of that.

Comment 5

[Leon Sha]

This bug is not reproduced any more. I am not sure if it is a duplicate bug of 465506. So just mark it as "works for me".