297750 - wrong sequence of declaration in union fd_twoints on x86

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Nian Liu(n/a in a long time)]

1.start mozilla on S10 x86
2.run javascript Math.pow(2,2) in venkman

expected result: output is 4
actural result: output is 0

the reason is that in union fd_twoints, __LITTLE_ENDIAN should be defined while
acturally not

Comment 1

[Nian Liu(n/a in a long time)]

Attachment: patch

Comment 2

[Brendan Eich [:brendan]]

Comment on attachment 186297 [details] [diff] [review]
patch

rs+a=me, assuming this has been tested on all relevant configurations.

I would like to do away with fdlibm where system libm is correct and (always
the case, AFAIK) performant.  Can someone from Sun comment on the effect of
using libm instead of fdlibm on the JS testsuite and any other tests that
stress floating point and Math methods?

/be

Comment 3

[Nian Liu(n/a in a long time)]

(In reply to comment #2)
> (From update of attachment 186297 [details] [diff] [review] [edit])
> rs+a=me, assuming this has been tested on all relevant configurations.
> 
> I would like to do away with fdlibm where system libm is correct and (always
> the case, AFAIK) performant.  Can someone from Sun comment on the effect of
> using libm instead of fdlibm on the JS testsuite and any other tests that
> stress floating point and Math methods?
> 
> /be
> 

Brendan,

If you can tell me how to run the JS testsuite, I'd like to do the test.

Comment 4

[Bob Clary [:bc] (inactive)]

(In reply to comment #3)

# from mozilla/js directory, checkout tests
cvs -z3 -q update -d -P tests

# from mozilla/js/src directory, checkout config, editline
cvs -z3 -q update -d -A config editline

# in mozilla/js/src directory, build optimized shell
make -f Makefile.ref BUILD_OPT=1

# in mozilla/js/src directory, build debug shell
make -f Makefile.ref

# in mozilla/tests directory, get help on running tests
perl jsDriver.pl --help

Comment 5

[Bob Clary [:bc] (inactive)]

oops, that last should have been
# in mozilla/js/tests directory, get help on running tests
perl jsDriver.pl --help

Comment 6

[Brendan Eich [:brendan]]

*** Bug 298298 has been marked as a duplicate of this bug. ***

Comment 7

[Leon Sha]

Comment on attachment 186297 [details] [diff] [review]
patch

I have two reasons to ask the approval.

1. For solaris x86 users, it is a lost of feature problem.
It cause all the math method in js not work.

2. This fix is a low risk fix.
It will not change other platform.

Comment 8

[Brendan Eich [:brendan]]

This was approved for b3, was it checked into the trunk?  It should be now if not.

/be

Comment 9

[Pete Zha]

checked in ->FIXED

Comment 10

[Bob Clary [:bc] (inactive)]

There are existing tests for Math.pow.

Comment 11

[Iain MacDonnell]

Solaris x86 users really need this simple fix. It works fine in trunk and when
applied to 1.7 and aviary branches - can we get it checked in there, please?

Comment 12

[Pete Zha]

Asa,

Could you please approve to fix this bug in aviary1.0.8? This is a low risk fix
but it is very important for Solaris x86

-Pete

Comment 13

[Daniel Veditz [:dveditz]]

Comment on attachment 186297 [details] [diff] [review]
patch

a=dveditz for drivers

Comment 14

[Bob Clary [:bc] (inactive)]

I can't verify this on solaris, but didn't find any other issues with Math.pow except as noted in bug 320770.

Nian, it isn't quite kosher, but if you can verify this on Solaris using a nightly build of Firefox 1.0.8 and Mozilla 1.7.13 I would very much appreciate it.