727223 - "Assertion failure: !obj->isBlock(),"

Status: VERIFIED FIXED

(Core :: JavaScript Engine, defect)

Description

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

Attachment: stack

try {
    mjitChunkLimit(1)
    function x() {}
} catch (e) {}
(function() {
    for (let c in [0, 0, 0]) {
        let c
        for (y in decodeURI()) {
            (function() {
                c
            }())
        }
    }
}())

asserts js debug shell on m-c changeset ebafee0cea36 with -m, -n and -a when the testcase is passed in as a CLI argument, at Assertion failure: !obj->isBlock(). Tested on 32-bit 10.6 js shell.

Guessing related to chunk patch in bug 706914 due to presence of mjitChunkLimit.

Comment 1

[Brian Hackett [Laid off!]]

Attachment: patch

This is actually an old bug in the BINDNAME IC (not sure how old, but seems to predate objshrink).  It tests shapes incorrectly (testing the scope's shape twice, and then off by one going up to the object.  This normally means it will miss, but it can get spurious hits when adjacent objects on the scope chain have identical shapes.  Not quite sure why mjitChunkLimit is required, but I think a specific order of invalidation/compilation is required for this testcase to assert.

Comment 2

[David Anderson [:dvander] - inactive, e-mail if emergency]

Comment on attachment 597979 [details] [diff] [review]
patch

Review of attachment 597979 [details] [diff] [review]:
-----------------------------------------------------------------

Bleh. Nice catch.

Comment 3

[Jan de Mooij [:jandem]]

Comment on attachment 597979 [details] [diff] [review]
patch

Review of attachment 597979 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/methodjit/PolyIC.cpp
@@ +1770,5 @@
> +            JSObject *tobj = &scopeChain->asScope().enclosingScope();
> +            Address parent(pic.objReg, ScopeObject::offsetOfEnclosingScope());
> +            while (tobj) {
> +                if (!IsCacheableNonGlobalScope(tobj))
> +                    return disable("non-cacheable obj in scope chain");

AFAICS this no longer calls IsCacheableNonGlobalScope for the first link on the scope chain, right? I'm working on a BINDNAME IC for IonMonkey (bug 728311) so I'm investigating what JM does.

Comment 4

[Brian Hackett [Laid off!]]

Yeah, IsCacheableNonGlobalScope should be tested for all objects on the scope chain.  I'll fix this before checking in.  I don't know whether it actually matters --- 'with' objects won't be the topmost object, since JSOP_WITH isn't compiled, but there could be problems with non-global scopes for event handlers and such.

Comment 5

[Brian Hackett [Laid off!]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/8610a9da1742

Comment 6

[Tom S [:evilpie]]

You might want to check-in the test case.

Comment 7

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/8610a9da1742

Comment 8

[Ed Morley [:emorley]]

Gary, don't suppose you could check in the testcase? :-)

Comment 9

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/325746d17f50

Comment 10

[Ed Morley [:emorley]]

Thank you :-)

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/325746d17f50

Comment 12

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

Test was landed in testsuite. -> VERIFIED