412619 - Math failures in Tamarin Tracing on OSX 10.5 with intel hardware

Status: RESOLVED FIXED

(Tamarin Graveyard :: Tracing Virtual Machine, defect)

Description

[Chris Peyer]

FAILURES:
  ecma3/Math/e15_8_2_6.abc : Infinity/Math.ceil(-Number.MIN_VALUE) = Infinity FAILED! expected: -Infinity
  ecma3/Math/e15_8_2_6.abc : Infinity/Math.ceil(-0.9) = Infinity FAILED! expected: -Infinity

These are failing in JIT and interp.
These pass on 10.5 on PPC
 
 
 
Transferred Comments:

Brent Baker - Thu Nov 15 10:42:36 CST 2007
These testcases are also failing in Tamarin.

Brent Baker - Fri Nov 30 09:11:01 CST 2007
Steven can you take a look at this mac failure. IT IS also happening in tamarin at this stage and it appears to be an issue ONLY when using 10.5 on intel hardware.

Comment 1

[Brent Baker]

This a duplication of https://bugzilla.mozilla.org/show_bug.cgi?id=401095 The testcases fails in both TT and TC

Comment 2

[Steven Johnson]

Err, the "wrong" values actually look correct to me! Consider:


Math.ceil(-0.9) == 0.0 (ditto for MIN_VALUE)

Infinity / 0.0 == Infinity, not -Infinity (as "expected")

FYI, executing the same snippet in SpiderMonkey (FF2) on my IntelMac on Leopard also gives me Infinity.

What am I missing here?

Comment 3

[Steven Johnson]

Ah. I'm guessing that Math.ceil is expected to return -0 here (not 0). Hmm...

Comment 4

[Steven Johnson]

It's easy to special-case a fix for this, but it appears to be broken in the ceil() function call in Leopard...

Comment 5

[Steven Johnson]

Fixed in changeset:   306:c811c024a458

Comment 6

[Brent Baker]

Do not expect these testcases to fail anymore (pushed changeset: 307:c4383132e8b0)