704795 - Assertion failure: isGCThing(), at ../../jsapi.h:536

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Christian Holler (:decoder)]

The following test asserts on mozilla-central revision 6f998cc964be (options -m -n -a):


gczeal(4);
function setprop() {
  var obj = { a:({   } )  };
  var obj2 = { b:-1, a:-1 };
  for (var i = 0; i < 20; (length(resultsY.global0, 42))) {
    obj2.b = obj.a = i;
  }
}
assertEq(setprop(), "19,-1,19");



Guessing this is only related to incremental GC due to gczeal(4), therefore not s-s.

Comment 1

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Another testcase:

Function("\
  gczeal(4,false);\
  function f(){\
    \"use strict\";\
    this.x = Object\
  }\
  for each(y in[0,0]){\
    try{\
      new f\
    }\
    catch(e){}\
  }\
")()

autoBisect shows this is probably related to the following changeset:

The first bad revision is:
changeset:   80143:65dcb557b3f6
user:        Bill McCloskey
date:        Fri Aug 05 10:25:23 2011 -0700
summary:     Bug 641027 - Add snapshot-based verifier for JS write barriers (r=luke)

Comment 2

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Attachment: fix

I couldn't reproduce the problem in comment 0, but the test in comment 1 worked. It was a dumb mistake.

Comment 3

[Christian Holler (:decoder)]

(In reply to Bill McCloskey (:billm) from comment #2)
> I couldn't reproduce the problem in comment 0, but the test in comment 1
> worked.

I think the test in comment 0 might be 32 bit only.

Comment 4

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

(In reply to Christian Holler (:decoder) from comment #3)
> (In reply to Bill McCloskey (:billm) from comment #2)
> > I couldn't reproduce the problem in comment 0, but the test in comment 1
> > worked.
> 
> I think the test in comment 0 might be 32 bit only.

Ah, thanks. I can reproduce now, and the patch I posted does not fix the problem. I'll start a new bug for the comment 0 testcase.

Comment 5

[Brian Hackett [Laid off!]]

Comment on attachment 577788 [details] [diff] [review]
fix

Mmm, the testing here still doesn't look right.  In the !isObject case, the write barrier is always performed and the address passed in is the payload + fixed slot offset.  But no object test has occurred, and the payload may not be an object's address.  I think you need to test for an object here, and in the non-object case skip the barrier entirely (the setprop will go through a stub which will hit barriers in the VM).

Comment 6

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Attachment: fix v2

Yes, you're right.

Comment 7

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a48fe9aef820

Comment 8

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/a48fe9aef820

Comment 9

[Christian Holler (:decoder)]

A testcase for this bug was automatically identified at js/src/jit-test/tests/basic/bug704795.js.