600596 - trace-test -mj failures for fennec/android on galaxy s

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

See sister bug 600488 for -j failures and https://wiki.mozilla.org/Mobile/Fennec/Android#JS_trace-tests for instructions on how to set up a testing environment.

However, we don't get very far, because we iloop at
(gdb) bt
#0  0xc03749b0 in ?? ()
#1  0xc03749b0 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

in trace-test/tests/arguments/args5.js.

Comment 1

[Robert Sayre]

which mozilla-central or tm cset are you working from?

Comment 2

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

m-c a9d1ad0bc386.

Comment 3

[Robert Sayre]

that is going to be too old:

you should pull tip now.

Comment 4

[Robert Sayre]

also, thanks for getting this up and running.

Comment 5

[Robert Sayre]

with TM tip on my Nexus One, I get 1 failure: date-format-tofte. It is not a
segfault. It's just a failing test. Anyone with a Galaxy want to try that?

Comment 6

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

When run by itself, yes

/data/local/js/trace-test/tests/sunspider/check-date-format-tofte.js:302: Error: Assertion failed: got "2007-01-01Monday,[...]

Comment 7

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Building from m-c
changeset:   54781:23605bdd871a
tag:         qparent
user:        Sid Stamm <[e-mail]>
date:        Thu Sep 23 15:38:47 2010 -0700
summary:     Bug 594446 - CSP report-uri should accept relative URIs [r=jst
a=blocking2.0: beta7+]

We're getting beyond arguments/args5.js, which is good.  Seeing a few "InternalError: too much recursion", which looks suspicious.

Bad failures first: these two tests hung.  See upcoming stack for
testDivisionFloat.  I couldn't get a stack from the second hang with
gdb, for whatever reason.

basic/testDivisionFloat.js
HANG
jaeger/getelem-sanity-3.js
HANG

I did three trials again, same setup as in bug 600488.

  3/3
 -----
arguments/strict-nested-assign-shadow-function-call.js
Exit code: -11

arguments/strict-nested-assign-shadow-function-name.js
Exit code: -11

basic/math-trace-tests.js
Exit code: -4 (twice)
/data/local/js/trace-test/tests/basic/math-trace-tests.js:46: TypeError: mapfunc is not a function
Exit code: 3 (once)

basic/testBug504520Harder.js
Infinity Infinity === Infinity
Infinity Infinity === -Infinity
Infinity Infinity === 0
Infinity Infinity === NaN
Infinity Infinity !== Infinity
Infinity Infinity !== -Infinity
Infinity Infinity !== 0
Exit code: -11

basic/testNewString.js
Exit code: -11

basic/testSlowArrayPopNestedTrees.js
Exit code: -11

basic/testincops.js
Exit code: -11

basic/truthies.js
Exit code: -4

sunspider/check-date-format-tofte.js
Exit code: 3


  2/3
 -----
arguments/strict-nested-assign-shadowed-catch.js
Exit code: -11

basic/testApplyCall.js
Exit code: -11

basic/deep2.js
Exit code: -11

basic/testBug552248.js
Exit code: -4

basic/testComparisons.js
Exit code: -11

basic/testShiftRightLogical.js
Exit code: -11
/data/local/js/trace-test/tests/basic/testShiftRightLogical.js:35: Error: Assertion failed
Exit code: 3

basic/testSlowNativeWithNullThis.js
Exit code: -11

closures/bug541239.js
/data/local/js/trace-test/tests/closures/bug541239.js:4: InternalError: too much recursion
Exit code: 3

closures/lambda.js
/data/local/js/trace-test/tests/closures/lambda.js:15: Error: Assertion failed: got 1, expected 2
Exit code: 3

pic/bug558099.js
Exit code: -11

Comment 8

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Trial 1

Comment 9

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Trial 2

Comment 10

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Trial 3

Comment 11

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Stack of testDivisionFloat hang and other failures from t-m build

Building t-m
changeset:   54638:da5a3ef843f8
tag:         qparent
user:        Nicholas Nethercote <[e-mail]>
date:        Wed Sep 29 16:12:04 2010 -0700
summary:     Bug 600489 - TM: allow for insBranch() returning NULL.  r=gal.

I don't see much difference.

basic/testDivisionFloat.js
HANG
jaeger/scriptedICs-1.js
HANG
v8-v5/check-deltablue.js
(MAYBE HANG?)

Other common failures are present.

Comment 12

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

And why not, -mj passes all tests in a debug build, except for being OOM-killed during the OOM test, and failing the broken date test.

Comment 13

[Robert Sayre]

I tested with a --disable-thumb2 build. This works on my Nexus One, but again fails on the Samsung.

Comment 14

[Robert Sayre]

Attachment: logcat of segfaults

Comment 15

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Oh, nice.  Does that happen automatically or did you enable it somehow?

Comment 16

[Robert Sayre]

it happens automatically.

Comment 17

[Robert Sayre]

Attachment: nm --demangle js

Comment 18

[Jacob Bramley [:jbramley]]

I couldn't reproduce on my Samsung board, but I got a different failure:

ubuntu@hbird-ubuntu:~/jacbra01/tm/js/src/trace-test$ python trace-test.py /home/ubuntu/jacbra01/tm.r/js
[ 857|   2| 859] 100% ===============================================>|   57.5s
FAILURES:
    -j /home/ubuntu/jacbra01/mozilla.org/tracemonkey/js/src/trace-test/tests/jaeger/bug588338.js
    -j /home/ubuntu/jacbra01/mozilla.org/tracemonkey/js/src/trace-test/tests/sunspider/check-date-format-tofte.js

(Ignore the date one.)

Got to run to a meeting.

Comment 19

[Jacob Bramley [:jbramley]]

(In reply to comment #18)
> FAILURES:
> [...]/trace-test/tests/jaeger/bug588338.js

I see the same failure on my other boards too, but only with release builds. I suspect that I'm seeing a totally different issue. (Actually, it looks like a Python error, but that seems unlikely to be the true cause, given that a debug build works.)

Comment 20

[Doug Turner (:dougt)]

on samsung devices we don't jit, moving this out to 2.0.