473392 - ARM JIT fails the ecma3/Statements/eregress_74474_002/3 testcases

Status: VERIFIED FIXED

(Tamarin Graveyard :: Baseline JIT (CodegenLIR), defect)

Description

[Brent Baker]

The following acceptance testcases fail on ARM when using the JIT.

ecma3/Statements/eregress_74474_002.abc
ecma3/Statements/eregress_74474_003.abc

I have not been able to extract any additional information about the failures as this is all that is returned by the proxy tool that I use to run the test on an ARM device.

Comment 1

[Brent Baker]

Marked as expected failure in testconfig.txt (1290:ef275c60b255)

Comment 2

[Andrew Booker]

I supplied a patch to fix these tests just before Christmas: see http://hg.mozilla.org/users/edwsmith_adobe.com/armjit/rev/649bde8d8eda

The problem is to do with patching up branches where the final target is not known when the code is generated, and ends up being outside the range of a standard branch instruction.

Comment 3

[Edwin Smith]

i integrated that patch, it's in tamarin-redux as
http://hg.mozilla.org/tamarin-redux/rev/ec80d152e803

so... perhaps it's a different problem?  would you mind trying the tamain-redux tip to see if the problem is reproducable on arm linux (the build failing is on windows mobile)

Comment 4

[Andrew Booker]

Oops - sorry - I hadn't noticed the platform was set to Windows!

The tests do pass with the tamarin-redux tip on my Linux setup.

Comment 5

[Brent Baker]

These tests are now passing. Any idea why they might be working? Appears to started working with a combination of the latest nanojit restructuring for x64... The builds were also broken for a week or so unfortunately it is difficult to tell exactly when it started working... Removing the expectedfail from the testconfig.

Comment 6

[Edwin Smith]

no idea, very possibly a heisenbug that ducked under the covers.

Comment 7

[Rick Reitmaier]

if can no longer repro, can we close this ?

Comment 8

[Brent Baker]

These failures are intermittently causing the windows mobile acceptance run to fail. I am going to 'skip' the tests in that suite until this is resolved.

Comment 9

[Rick Reitmaier]

Seems like in latest runs we are *not* repro'ing this problem.  

So after fixing 429673, 476639 and 477228, and ensuring that all tests are running stably, we'll re-enable these tests to run.  Patch pending.

Comment 10

[Rick Reitmaier]

476639 can be removed from this dependency list since it is an enhancement

Comment 11

[Rick Reitmaier]

I just checked in fixes, so deps are met.  Should we re-enable the tests?

Comment 12

[Brent Baker]

Removed the skip since it has not be reproducible, pushed as 1596:74e23e3686ef

Comment 13

[Rick Reitmaier]

mark fixed.

Comment 14

[Brent Baker]

Tests have been passing without failure for the past 7 months, "verified fixed"

Comment 15

[Brent Baker]

removing QRB request, bug resolved/verified