1033856 - Assertion failure: false (MOZ_ASSERT_UNREACHABLE: unexpected type), at jit/Lowering.cpp:1772 with Symbol

Status: VERIFIED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Christian Holler (:decoder)]

The following testcase asserts on mozilla-central revision 7075808c3306 (run with --fuzzing-safe --ion-eager):


function TestCase(e, a)
  this.passed = getTestCaseResult(e, a);
function reportCompare (e, a) {
  printStatus ();
  var testcase = new TestCase(e, a);
}
function getTestCaseResult(e, a) {
    return a == e;
}
try {
test();
} catch(exc1) {}
function test()
  reportCompare();
new TestCase( .1, 0.1e0 );
var sym = Symbol("method");
evaluate("\
  reportCompare(test, sym, length)\
", { noScriptRval : true, compileAndGo : true});

Comment 1

[Christian Holler (:decoder)]

Attachment: [crash-signature] Machine-readable crash signature

Comment 2

[Christian Holler (:decoder)]

Needinfo from Jason, as this is related to the recent Symbol implementation.

Comment 3

[Jason Orendorff [:jorendorff]]

Attachment: bug-1033856-symbol-tonumber-jit-v1.patch

Again, would love some help regarding how to test each of these changes. There must be a less convoluted way.

The change to the comment (only) in LIRGenerator::visitToString is justified by the check for type == MIRType_Symbol in ToStringPolicy::staticAdjustInputs.

Comment 4

[Jan de Mooij [:jandem]]

Comment on attachment 8450168 [details] [diff] [review]
bug-1033856-symbol-tonumber-jit-v1.patch

Review of attachment 8450168 [details] [diff] [review]:
-----------------------------------------------------------------

(In reply to Jason Orendorff [:jorendorff] from comment #3)
> Again, would love some help regarding how to test each of these changes.
> There must be a less convoluted way.

Here's a test for the ToDouble case (fails with --ion-eager --ion-offthread-compile=off):

function f(x, y) {
    return x == y;
}
f(1.1, 2.2);
for (var i=0; i<5; i++)
    f(1, Symbol());


ToFloat32 is a bit more complicated unfortunately.

::: js/src/jit/MIR.h
@@ +3476,5 @@
>          return AliasSet::None();
>      }
>  
>      bool fallible() const {
> +        return input()->mightBeType(MIRType_Object) || input()->mightBeType(MIRType_Symbol);

You don't need this change. We give the LIR instruction a snapshot if fallible() returns true, so that CodeGenerator::visitValueToString can bail out.

For symbols we do a VM call that can still be "fallible" but in a different way, it's not affected by this method.

I think this method should be renamed (or just inlined into the only caller) to avoid confusion, but doesn't have to happen in this patch.

Comment 5

[Jason Orendorff [:jorendorff]]

Symbols regressed this, count on it.

Comment 6

[Jason Orendorff [:jorendorff]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/4c5ee514a3ac

Comment 7

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/4c5ee514a3ac

Comment 8

[Christian Holler (:decoder)]

JSBugMon: This bug has been automatically verified fixed.

Comment 9

[Christian Holler (:decoder)]

JSBugMon: This bug has been automatically verified fixed on Fx33